THE STRUCTURE AND MEANING OF COMPOUNDS
IN CHILD AND ADULT GRAMMARS
By
Jessica Lee Gamache

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
Linguistics—Doctor of Philosophy
2014

ABSTRACT
THE STRUCTURE AND MEANING OF COMPOUNDS
IN CHILD AND ADULT GRAMMARS
By
Jessica Lee Gamache
This dissertation presents a series of studies examining the structure and meaning of
compounds, such as mouse-trap and mouse-catcher, in child and adult grammars. The first part
of this dissertation examines how adults interpret compounds and the second part examines
how children acquire the adult compounding system. The goals are (i) to argue that a
generative lexical semantics is necessary to understand how speakers use compounds and (ii)
to argue that children use the same generative mechanism as adults to build compound
interpretations.
In the first part of this dissertation two adult studies examine the role of a generative
syntax and a generative lexical semantics (Pustejovsky 1995) in the interpretation of
compounds. The first study asks adults to interpret novel noun-noun compounds and verbal
compounds. The results support the hypothesis that organized lexical information, including
natural classes, must be taken into account in addition to syntactic structure to determine the
most likely meaning of compounds. The second study asks adults to rate different
interpretations of verbal compounds. The results support the hypothesis that the syntax of
verbal compounds constrains the relationship between the verb and modifier to a single
verb-internal argument interpretation. Taken together, these results support a model where
both the syntax and the lexical semantics are active, generative mechanisms.
In the second part of this dissertation five child studies test the hypothesis that, given
compositionality, children’s knowledge of syntax and lexical semantics go hand in hand with
their interpretation. Three studies examine children’s production and interpretation of noun-

noun compounds, with the goal of determining if children use lexical structure in the same
way as adults when producing and interpreting compounds. The first noun-noun study is a
corpus study. The results support the hypothesis that children and adults produce novel
noun-noun compounds according to the natural classes of the constituents in compounds,
but not the individual words in compounds. The second noun-noun compound study is an
elicited production task. The results provide evidence for children’s production of
compounds for the same range of meanings as adults, albeit to a lesser degree overall.
Finally, the third noun-noun compound study is a forced-choice interpretation task. The
results support the hypothesis that children interpret novel compounds according to their
knowledge of natural class behavior, rather than individual word behavior.
Next, two studies examine children’s production and interpretation of verbal
compounds. The goal of these studies is to examine whether children’s knowledge of verbal
compound syntax, as evidenced by their production, makes predictions about children’s
comprehension of verbal compounds. The first study is an elicited production task. The
results provide evidence for the idea that children’s non-adult verbal compound forms (catchmouse and catcher-mouse but not mouse-catch for mouse-catcher) reflect the steps in the adult
derivation of verbal compounds. The second study is a forced-choice interpretation task.
From the results we observe that children constrain their interpretation of verbal
compounds based on their knowledge of the verbal compound structure.
In summary, the results of the child studies provide evidence that children use the
same system as adults to interpret compounds, and any deviations from adult-like
interpretations can be explained by non-adult syntactic or lexical semantic knowledge.

ACKNOWLEDGMENTS

This dissertation would have never come to life (especially in the speed in which it
did) without the love, care, and support of so many people.
Mom, thank you for always setting an example of perseverance and strength. As I’ve
gotten older I’ve found myself more and more in awe of all that you have been able to do.
Dad, thank you for always supporting me in any decision I’ve made. Your faith in me
has helped me to push myself when I was too afraid to go after what I really wanted.
Brucie, thank you for being everything and more that a child could ask for in a
stepfather and an amazing husband to Mom.
Tressa and Becky, thank you for being my best friends and always being there to talk
to me when I need them (except for when they think I’m being an “arrogant …”). I’m really
glad nothing ever became of the butter knife incident. Soon enough we will find the ideal
vacation where no one ends up crying and Tress doesn’t feel the need to run away at 6 am
because we’re fighting over a broken comb. On the upside, us fighting gets Tressa emotional
and that’s entertaining.
Mema, thank you for always supporting me and being up for new adventures. I
promise that I will never drag you around another city (Chicago) or country (Spain) on foot
again.
Steve, I don’t know how you dealt with me for this past year of my life, but I will
never be able to thank you enough for being there for me and loving me through every freak
out and tear and the small monster I became. I can’t wait for all of the adventures we have
to come.

iv

Kerry, Molly, and Tori, thank you for still being a part of my life after all of these
years. I am so lucky to know that, even if we go months without talking, you will all be there
for me.
Tuuli and Kali, what would I have done in Michigan without you two? Period.
Having the New Yorker and the Texan as constants in my life gave me more of a sense of
balance than I could have ever imagined.
Greg and Curt, thank you for being my first friends and Michigan and two of the
people I always knew that I could go to when I needed to talk. Even if the talking involving
cry-laughing.
Cristina, thank you for all of the time and patience that you’ve given to me. You have
taught me so much about research and how to be a good mentor. I hope that I can embody
that as I go on into my career.
Alan, thank you for your insight and help over the past five years. Your easy-going
nature made me feel like I was right at home from the second I got to MSU.
Laura, thank you for taking me under your wing and teaching me so much of what I
know about experimental research. I hope that I was able to give back to the lab as much as
I got from it.
Yen-Hwei, thank you for always being patience and generous with your time and
advice. It was invaluable to my time at MSU.
Devin, thank you for being a more than reasonable boss, especially in my last year
when I was much more focused on my dissertation than being the cognitive science
assistant.
Mina and Kyle, the amount of enthusiasm and dedication that you brought to our
research was so inspiring to me. I am so glad that we’ve come to be friends. If either of you
v

ever want to go back to Europe, let me know! This time we’ll go for more than 4 days and
we’ll spend less time in a conference room.
Mrs. Elwood, I am so glad that we have stayed in touch for all of these years. While
you started as just my teacher, you’ve become a friend and mentor and your sound advice
has served me well so far, and I assume will for the rest of my life.
And finally, to the many, many schools that allowed me to come in and conduct my
research, thank you so, so much. Without you my research could truly not be done.

vi

TABLE OF CONTENTS

LIST OF TABLES .............................................................................................................................. xi
LIST OF FIGURES ......................................................................................................................... xiii
CHAPTER 1: INTRODUCTION ................................................................................................... 1
1.0 Introduction ................................................................................................................................. 1
1.1 Adults’ Interpretation of Compounds ..................................................................................... 2
1.2 Children’s Interpretation of Compounds ................................................................................ 5
1.3 Relevance ...................................................................................................................................... 8
1.4 Organization of Chapters........................................................................................................... 9
CHAPTER 2: LINGUISTICS BACKGROUND........................................................................ 11
2.0 Introduction ............................................................................................................................... 11
2.1 Basic assumptions ..................................................................................................................... 11
2.2 Compound Basics ..................................................................................................................... 12
2.3 Deriving Interpretations........................................................................................................... 13
2.3.1 The Syntax of Compounds ............................................................................................ 13
2.3.1.1 The Syntax of Noun-Noun Compounds ..............................................................13
2.3.1.2 The Syntax of Verbal Compounds .........................................................................15
2.3.1.3 Summary of the Role of Syntax in Compound Relationships............................16
2.3.2 Determining the Relationship in Noun-Noun Compounds..................................... 16
2.3.1.1 Extrinsic Approach to Compounds .......................................................................16
2.3.1.2 Intrinsic Approach to Compounds ........................................................................18
2.4 A Proposed System for Compound Interpretation ............................................................. 22
2.5 Hypotheses and Predictions for Adults’ Interpretation of Compounds .......................... 29
2.6 Summary ..................................................................................................................................... 31
CHAPTER 3: ADULT COMPOUND TASKS ........................................................................... 32
3.0 Introduction ............................................................................................................................... 32
3.1 Adult Study 1: Interpretation of Noun-Noun Compounds and Verbal Compounds .... 32
3.1.1 The Role of Syntax on the Interpretation of Compounds........................................ 32
3.1.2 The Role of Lexical Semantics on the Interpretation of Compounds .................... 34
3.1.3 Method.............................................................................................................................. 36
3.1.3.1 Participants .................................................................................................................36
3.1.3.2 Materials and Procedure ...........................................................................................37
3.1.3.3 Coding and Analysis .................................................................................................37
3.1.4 Results ............................................................................................................................... 40
3.1.4.1 The Role of Syntax on Interpretation ....................................................................40
3.1.4.1.1 The Role of Syntax on Agentive Interpretations ..........................................41
3.1.4.1.2 The Role of Syntax on Constitutive Interpretations.....................................41
3.1.4.1.3 The Role of Syntax on Formal Interpretations .............................................41
3.1.4.1.4 The Role of Syntax on Telic Interpretations .................................................42
3.1.4.1.5 Summary of the Role of Syntax on Interpretation .......................................42
3.1.4.2 The Role of Semantic Natural Class on Interpretation .......................................43
vii

3.1.4.2.1 The Role of Semantic Natural Class on Agentive Interpretations .............46
3.1.4.2.2 The Role of Semantic Natural Class on Constitutive Interpretations .......46
3.1.4.2.3 The Role of Semantic Natural Class on Formal Interpretations ................46
3.1.4.2.4 The Role of Semantic Natural Class on Telic Interpretations ....................47
3.1.4.2.5 Summary of the Role of Semantic Natural Class on Interpretation ..........47
3.1.5 Compound Interpretation Study Discussion .............................................................. 48
3.2 Adult Study 2: Acceptable and Preferred Interpretations of Verbal Compounds .......... 50
3.2.1 The Role of Syntax in the Interpretation of Verbal Compounds ............................ 50
3.2.2 The Role of Telicity in the Interpretation of Verbal Compounds ........................... 52
3.2.3 Method.............................................................................................................................. 53
3.2.3.1 Participants .................................................................................................................53
3.2.3.2 Materials and Procedure ...........................................................................................53
3.2.3.2.1 Acceptability Task ..............................................................................................56
3.2.3.2.2 Preference Task ..................................................................................................56
3.2.3.3 Analysis ...................................................................................................................57
3.2.4 Acceptability Task Results ............................................................................................. 57
3.2.4.1 Acceptability Task Response Results .....................................................................57
3.2.4.2 Acceptability Task Response Time Results ...........................................................59
3.2.5 Preference Task Results ................................................................................................. 61
3.2.5.1 Preference Task Response Results .....................................................................61
3.2.5.2 Preference Task Response Time Results ...........................................................62
3.2.6 Acceptability and Preference Task Discussion ........................................................... 63
3.3 General Discussion ................................................................................................................... 64
CHAPTER 4: ACQUISITION BACKGROUND ...................................................................... 65
4.0 Introduction ............................................................................................................................... 65
4.1 Different Approaches to Language Acquisition ................................................................... 65
4.2 What Children Know about Compounds ............................................................................. 66
4.2.1 Children’s Acquisition of Noun-Noun Compounds ................................................. 66
4.2.1.1 Acquisition of Noun-Noun Compound Syntax ...................................................66
4.2.1.2 Children’s Interpretations of Noun-Noun Compounds .....................................68
4.2.1.3 Acquisition of Noun-Noun Compounds Summary and Questions..................70
4.2.2 Children’s Acquisition of Verbal Compounds............................................................ 71
4.2.2.1 Acquisition of Verbal Compound Syntax .............................................................71
4.2.2.2 Children’s Interpretation of Verbal Compounds .................................................75
4.2.2.3 Children’s Interpretation of Compounds Summary and Questions .................76
4.2.3 Summary of Past Acquisition Research ....................................................................... 76
4.3 Hypotheses and Predictions for Children’s Acquisition of Compounds ......................... 77
4.4 Summary ..................................................................................................................................... 79
CHAPTER 5: CHILD NOUN-NOUN COMPOUND TASKS .............................................. 80
5.0 Introduction ............................................................................................................................... 80
5.1 Child Study 1: Noun-Noun Compounds in Children’s Speech and their Input ............. 81
5.1.1 Method.............................................................................................................................. 83
5.1.1.1 Coding.........................................................................................................................83
5.1.1.2 Analysis .......................................................................................................................85
5.1.2 Results ............................................................................................................................... 86
5.1.2.1 Head Noun Natural Class Analysis Results ..........................................................86
viii

5.1.2.2 Modifier Noun Natural Class Analysis Results ....................................................88
5.1.2.3 Noun Analysis ...........................................................................................................90
5.1.2.3 Summary of Results ..................................................................................................91
5.1.3 Discussion ........................................................................................................................ 92
5.2 Child Study 2: Elicited Production of Noun-Noun Compounds ...................................... 93
5.2.1 Method.............................................................................................................................. 94
5.2.1.1 Participants .................................................................................................................94
5.2.1.2 Procedure ...................................................................................................................94
5.2.1.3 Coding.........................................................................................................................97
5.2.1.4 Analysis .......................................................................................................................97
5.2.2 Results ............................................................................................................................... 97
5.2.2.1 Target Form Analysis ........................................................................................ 100
5.2.2.2 Reversal Form Analysis ..................................................................................... 100
5.2.2.3 Phrasal Form Analysis ....................................................................................... 101
5.2.2.4 Summary of Results ........................................................................................... 101
5.2.3 Discussion ...................................................................................................................... 101
5.3 Child Study 3: Preferred Interpretations of Noun-Noun Compounds .......................... 102
5.3.1 The Role of Lexical Semantics in Children’s Interpretation of Compounds ....... 102
5.3.2 The Role of Intrinsic/Accident Information in Children’s Use of Compounds 104
5.3.3 Method............................................................................................................................ 106
5.3.3.1 Participants .............................................................................................................. 106
5.3.3.2 Materials and Procedure ........................................................................................ 106
5.3.3.3 Analysis .................................................................................................................... 110
5.3.4 Results ............................................................................................................................. 111
5.3.4.1 Overall Analysis ...................................................................................................... 111
5.3.4.2 Analysis of Consistency by Head Noun ............................................................. 112
5.3.4.3 Summary of Results ............................................................................................... 112
5.3.5 Discussion ...................................................................................................................... 113
5.4 General Discussion ................................................................................................................. 113
CHAPTER 6: CHILD VERBAL COMPOUND TASKS........................................................ 116
6.0 Introduction ............................................................................................................................. 116
6.1 Child Study 4: Elicited Production of Verbal Compounds .............................................. 117
6.1.1 Method............................................................................................................................ 117
6.1.1.1 Participants .............................................................................................................. 117
6.1.1.2 Materials and Procedure ........................................................................................ 117
6.1.1.3 Coding and Analysis .............................................................................................. 120
6.1.2 Results ............................................................................................................................. 121
6.1.2.1 Analysis of Verbal Compound Permutations .................................................... 121
6.1.2.3 Analysis of Distribution of Verbal Compound Forms .................................... 123
6.1.2.4 Analysis of Distribution of Overall Forms ........................................................ 124
6.1.2.5 Summary of Results ............................................................................................... 125
6.1.3 Discussion ...................................................................................................................... 126
6.2 Child Study 5: Preferred Interpretations of Verbal Compounds ..................................... 127
6.2.1 Method............................................................................................................................ 128
6.2.1.1 Participants .............................................................................................................. 128
6.2.1.2 Materials and Procedure ........................................................................................ 128
6.2.1.3 Analysis .................................................................................................................... 131
ix

6.2.2 Results ............................................................................................................................. 131
6.2.2.1 Preference Task Analysis ...................................................................................... 131
6.2.2.2 Summary of Results ............................................................................................... 133
6.2.3 The Interaction of Syntax and Interpretation ........................................................... 133
6.3 General Discussion ................................................................................................................. 137
CHAPTER 7: SUMMARY AND CONCLUSIONS ................................................................. 139
7.1 Goal of Dissertation ............................................................................................................... 139
7.2 Conclusions .............................................................................................................................. 144
APPENDICES ................................................................................................................................. 146
APPENDIX A. Full Compound List for Adult Study 1 ......................................................... 147
APPENDIX B. Full Compound List for Corpus Study ......................................................... 151
APPENDIX C. Production Forms for VC Elicited Production Task .................................. 157
APPENDIX D. Responses by Verb for Child Verbal Compound Interpretation Study .. 158
REFERENCES ................................................................................................................................ 159

x

LIST OF TABLES

Table 1. Levi's (1978) Noun-Noun Compound Relationships....................................................17
Table 2. Most Frequent Relationship Types by Head Noun Class .............................................21
Table 3. Hypotheses and Predictions of New Interpretation System ........................................29
Table 4. NNC and VC Interpretation: Noun Classes ...................................................................35
Table 5. NNC and VC Interpretation: Coding Scheme ...............................................................38
Table 6. Count of Items for Syntax X Head Lexical Status Analysis .........................................39
Table 7. Count of Items for Head Natural Class X Modifier Natural Class Analysis .............40
Table 8. Proportion (and Count) of Items by Relationship, Syntax, and Lexical Status .........40
Table 9. Proportion (and Count) of Items by Relationship and Natural Classes .....................45
Table 10. Full Item List for Adult Study 2 .....................................................................................55
Table 11. Type (and token) count of compounds in adult speech in Brown (1973) corpus ..73
Table 12. Possible non-adult forms for NVer, VN, and VingN .................................................74
Table 13. Hypotheses and Predictions for Children's Acquisition of Compounds .................77
Table 14. NNC and VC Interpretation: Noun Categories ...........................................................84
Table 15. Coding Scheme .................................................................................................................85
Table 16. Count of Compound Types by Speaker(s) ....................................................................86
Table 17. χ2 Results on the Distribution of Relationships by Head Natural Class ..................88
Table 18. χ2 Results on the Distribution of Relationships by Modifier Natural Class.............89
Table 19. Count of Individual Noun Types by Position and Speaker(s) ...................................91
Table 20. NNC Elicited Production: Items ....................................................................................95
Table 21. Proportion (and Count) of Forms by Item Type and Age Group ............................98
Table 22. NNC Interpretation: Items .......................................................................................... 107
xi

Table 23. VC Elicited Production: Items..................................................................................... 118
Table 24. VC Elicited Production: Coding Scheme ................................................................... 120
Table 25. Proportion (and Count) of Verbal Compound Permutations ................................ 122
Table 26. VC Interpretation: Items .............................................................................................. 129
Table 27. Groups based on Elicited Production Task Performance ....................................... 135
Table 28. Full Compound List for Adult Study 1 by Head and Modifier Category.............. 147
Table 29. Compound List for Corpus Study by Speaker .......................................................... 151
Table 30. VC Production: Counts for All Production Forms By Age Group ....................... 157

xii

LIST OF FIGURES

Figure 1. Hypotheses for Adults' Interpretation of Compounds.................................................. 4
Figure 2. Hypotheses for Children’s Interpretation of Compounds ............................................ 7
Figure 3. The Qualia Structure of knife ...........................................................................................19
Figure 4. The Qualia Structure of the Unified Type food ..............................................................20
Figure 5. Proposed Compound Interpretation System.................................................................25
Figure 6. Examples of Compound Interpretation in Proposed System .....................................28
Figure 7. Distribution of Relationships by Head Natural Class ..................................................43
Figure 8. Distribution of Relationships by Modifier Natural Class ............................................44
Figure 9. VC Acceptability: Proportion of "1" Responses by Binary Group ............................58
Figure 10. VC Acceptability: Average Acceptability Rating by Likert Group ...........................59
Figure 11. VC Acceptability: log Response Times by Verb Type for Binary Group ...............60
Figure 12. VC Acceptability: log Response Times by Verb Type for Likert Group ................61
Figure 13. VC Preference: Proportion of Argument Responses by Verb Type .......................62
Figure 14. VC Preference: log Response Times by Verb Type ...................................................63
Figure 15. Distribution of Relationships by Head Class and Speaker for Adam Files ............87
Figure 16. Distribution of Relationships by Head Class and Speaker for Sarah Files .............87
Figure 17. Distribution of Relationships by Modifier Class and Speaker for Adam Files ......88
Figure 18. Distribution of Relationships by Modifier Class and Speaker for Sarah Files .......89
Figure 19. Distribution of Relationships by Modifier Noun and Speaker for Adam Files .....90
Figure 20. Distribution of Relationships by Head Noun and Speaker for Adam Files ...........91
Figure 21. NNC Elicited Production: Sample Item ......................................................................96
Figure 22. NNC Production: Proportion of Production Types by Younger Children............99
xiii

Figure 23. NNC Production: Proportion of Production Types by Older Children.................99
Figure 24. NNC Production: Proportion of Production Types by Adults ................................99
Figure 25. NNC Interpretation: Sample Item ............................................................................. 109
Figure 26. Proportion of Telic Responses By Item Type and Age Group ............................. 111
Figure 27. Proportion of Head Nouns with Consistent Telic Interpretation by Age ........... 112
Figure 28. VC Elicited Production: Sample Item ....................................................................... 119
Figure 29. VC Elicited Production: Proportion of Verbal Compound Forms ...................... 123
Figure 30. VC Elicited Production: Proportion of Overall Forms by Age Group ............... 124
Figure 31. VC Interpretation: Sample Item ................................................................................. 130
Figure 32. VC Interpretation: Proportion of Argument Responses by Participant .............. 132
Figure 33. VC Interpretation: Proportion of Argument Responses ........................................ 132
Figure 34. VC Interpretation: Proportion of Argument Responses by Production Group. 136
Figure 35. VC Interpretation: Proportion of Argument Responses by Individual Verb...... 158

xiv

CHAPTER 1: INTRODUCTION
1.0 Introduction
The goal of this dissertation is to argue that children and adults interpret compounds
in the same way. Compounds serve as an interesting window into the acquisition of language
because of their ability to create a rich array of meanings from limited linguistic means. A
candy-machine, for example, could be a machine that makes candy, a machine that sells candy,
a machine that is made out of candy, etc. All of these interpretations are available for candymachine even though the components (candy, machine) and the way in which they are combined
(noun-noun compound syntax) do not change. Other compounds allow a smaller array of
interpretations. A candy-maker, for example, cannot mean a machine that uses candy to make
things. The two overarching questions in this dissertation stem from this simple observation.
(1) Question 1: How do adults arrive at the interpretation of different types of
compounds where the weight of the syntactic and semantic information varies?
Question 2: What process do children go through to arrive at the adult system for
interpreting compounds, given its necessarily flexible nature?
The first part of this dissertation defines a compositional system for interpreting two
different types of compounds: verbal compounds (mouse-catcher, truck-driver) and noun-noun
compounds (mouse-trap, book-shelf). The goal is to provide evidence that adults use the syntax
of compound structures and structured lexical information, including natural class, to
interpret novel compounds. The second part of this dissertation examines how children
learn to produce and interpret compounds. The goal of the acquisition portion of this
dissertation is to provide evidence that children differ in their syntactic and lexical semantic
knowledge, but ultimately use the same system as adults to interpret compounds.

1

1.1 Adults’ Interpretation of Compounds
How might adults interpret compounds? As a starting point, let’s assume that
determining the meaning of a compound minimally requires knowing its syntax and the
meaning of its constituents. Depending on the type of compound, the extent to which these
sources of information are drawn upon varies. The two compounds under investigation in
this dissertation, verbal compounds and noun-noun compounds, are a good example of this
shift in relevant information sources. For verbal compounds, the syntax is argued to largely
constrain the interpretation of the compound down to one possible relationship between its
constituents (Roeper and Siegel 1978, Selkirk 1982, Lieber 1983, Di Sciullo and Williams
1987, Bobaljik 2003). For instance, a mouse-catcher can only be a person or thing that catches
mice, where mouse is the internal argument of the verb catch, and therefore is interpreted as its
theme (Hale and Keyser 1993). A mouse-catcher cannot be a person who catches things using a
mouse, where mouse serves as an instrument to the verb catch. For noun-noun compounds,
on the other hand, syntax is argued to play almost no role in constraining the interpretation
of the relationship in the compound, beyond determining the head of the compound.
If not syntax, what determines the relationship between the nouns in noun-noun
compounds? A majority of the previous work on the interpretation of noun-noun
compounds posits that there is a set number of semantic relationships (‘in’, ‘for’, ‘from’, etc.)
that exist in compounds (Lees 1960, Gleitman and Gleitman 1970, Levi 1978, Warren 1978).
For example, mouse-trap and soup-spoon both have ‘for’ relationships; a mouse-trap is a trap for
(catching) mice, and a soup-spoon is a spoon for (eating) soup. While this type of system allows
for the classification of compounds based on their interpretations, it fails to explain two
things: (i) why a particular set of relationships but not others exist in compounds and (ii) the
mechanism by which a speaker decides which relationship exists in a compound.
2

An improvement on the above system was proposed by Johnston and Busa (1999),
who argued that the interpretation of noun-noun compounds is determined by the internal
lexical structure of its constituents. The internal structure of a lexical item specifies particular
types of information about the lexical item, such as its appearance, function, and part-whole
structure (Pustejovsky 1995). Under this system, a compound like mouse-trap can be
interpreted as a trap for catching mice because trap lists “to catch” under its functional
specification and mice are something than can be caught. Furthermore, because lexical items
are categorized according to the features that they share with other lexical items, the classes
of constituents in a compound determine their most likely interpretation. For example,
mouse-trap is most likely to be interpreted with a functional reading because trap is an artifact
and artifacts are defined by their function. In fact, Downing (1977) noted that the
relationship in a compound was related to the natural class of the constituents in the
compound. In her study, compounds with artifact head nouns often had function-related
interpretations, while animate natural kinds often had appearance- or origin-related
interpretations. This finding is in line with the way that natural classes are generally defined
(Rips 1989). This approach, which employs a generative lexical semantics, helps to explain
both (i) why a particular set of relationships exists and (ii) why a particular relationship is
most likely to be intended.
In this dissertation I explore an interpretation system for compounds that uses
Johnston and Busa’s theory for compound interpretation in conjunction with current
theories of compound syntax. The adult studies in this dissertation aim to test three
hypotheses.

3

Figure 1. Hypotheses for Adults' Interpretation of Compounds
Adult Hypothesis 1: Syntactic structure is deterministic in its assignment of a
relationship to a compound.
Adult Hypothesis 2: Lexical structure supplies the possible relationships in compounds.
Adult Hypothesis 3: Natural class membership is a relevant element to determine the
relationship in a compound.

The first hypothesis is that syntactic structure is deterministic in its assignment of a
relationship to constituents of a compound. This hypothesis predicts that when the syntax of
a compound defines a particular relationship between its constituents, the compound has to
be interpreted using that relationship. Verbal compound syntax, for example, is predicted to
always impose a verb-internal argument relationship on its constituents. That is, mouse in
mouse-catcher must be interpreted as the internal argument of catch when mouse-catcher is parsed
with verbal compound syntax.
The second hypothesis is that the lexical structure supplies the possible relationships
in compounds (Johnston and Busa 1999). This predicts that compounds can only be
assigned a particular relationship if that relationship is specified in the lexical structure of its
constituents. Scissors, for example, has a functional specification (what it is for, to cut), but
salamander does not have an inherent functional specification. Therefore, compounds with
the head noun scissors are likely to receive a functional interpretation while compounds with
the head noun salamander are not likely to receive a functional interpretation.
The third hypothesis is that natural class membership is relevant in determining the
relationship in compounds. This hypothesis predicts that the subclasses of words that make
up the compound, in absence of other information, narrow down the possible relations.
Artifacts, for examples, are likely to lead to functional interpretations, while animals are likely
to lead to an appearance- or origin-related interpretation. Therefore, paper-scissors is likely to
4

receive a functional interpretation (scissors for cutting paper), and paper-salamander is likely to
receive an appearance interpretation (a salamander that looks like paper). This is not to say
that other interpretations are not possible, only that these are the likely interpretations if
category membership is prominent in the absence of context.
1.2 Children’s Interpretation of Compounds
In determining the system that adults use to interpret compounds, we also
distinguish the system that children are targeting during the acquisition process. The goal of
the second part of this dissertation is to provide evidence that children interpret compounds
using the same system as adults. As outlined above, the target adult system has two
components that must be learned: the syntax of compounds and the lexical structure of the
words in a compound including their natural classes. Any deviations from adult
interpretations of compounds should be attributable to lacking syntactic or lexical semantic
knowledge in children’s grammar.
A wealth of work has already set up a good foundation for how children acquire
compounds. We know that children learn noun-noun compound syntax around age two
(Clark 1981, Snyder 1995), but because of the simple morphological structure of noun-noun
compounds and the early age at which its syntax is mastered, little is known about the
acquisition process. Children learn verbal compound syntax around age 5 (Clark, Hecht, and
Mulford 1986) and have been repeatedly shown to only produce a particular subset of
possible non-adult forms (catch-mouse and catcher-mouse, but not mouse-catch for mouse-catcher).
These forms are argued to be a result of particular frequencies in the input (Clark et al. 1986,
Nagpal and Nicoladis 2009) or a reflection of the steps in the adult derivation that children
are acquiring (Gamache and Schmitt to appear).

5

Relatively less work has examined how children interpret compounds. We know that
children understand headedness by age 3 (Clark, Gelman, and Lane 1985). When choosing
between competing relationships in compounds, two recent studies have shown apparent
biases in children’s interpretation of noun-noun compounds. In one study, children were
shown to interpret novel compounds in analogy to known compounds with the same head
noun (Krott, Gagne, and Nicoladis 2009). For instance, if a blueberry-spoon was known to be a
spoon for eating blueberries, then a cake-spoon was interpreted as a spoon for eating cake. In
another study, children were shown to nonce word compounds with a bias towards
interpretations where the head and modifier noun were physically related (Krott, Gagne, and
Nicoladis 2010). For example, a wug-moop was interpreted as a moop that has a wug, or a
moop that is next to a wug, rather than a moop that was used to do something to the wug.
No studies to date have shown how children interpret the relationship between the
verb and the modifier in verbal compounds. Therefore, it is an open question if children
restrict the meaning of mouse-catcher, for example, to a person who catches mice, as adults do,
or if children allow a wider set of interpretations, such as a person who uses mice to catch
other things.
The goal of the acquisition portion of this dissertation is to build on previous
findings and examine (i) how children learn the syntax of compounds and (ii) if children use
the same system as adults to interpret compounds, taking into account their possibly
different syntactic and lexical semantic knowledge. The child studies in this dissertation test
two hypotheses about how children acquire compound structures and meanings.

6

Figure 2. Hypotheses for Children’s Interpretation of Compounds
Child Hypothesis 1: The strong continuity hypothesis is correct for the acquisition of
compound structures.
Child Hypothesis 2: Given principles of compositionality, children’s knowledge of
syntax and lexical semantics go hand in hand with their
interpretation.
The first hypothesis is that the strong continuity hypothesis is correct for the
acquisition of compound structures (Pinker 1984). This predicts that all forms that children
produce are steps in the adult derivation, and that these forms are subject to the same
constraints as the adult structures. For example, the only forms that children should produce
when acquiring verbal compounds are those which are consistent with an adult derivation of
verbal compounds. This hypothesis is in direct opposition to any hypothesis that suggests
that the mechanism that children have at their disposal while learning syntax is different
from the mechanism that adults use in their grammar (e.g, Tomasello 2000).
The second hypothesis is that, given principles of compositionality, children’s
knowledge of syntax and lexical semantics go hand in hand with their interpretation. This
hypothesis predicts that, from the beginning, children use syntax and lexical information to
interpret compounds and any deviations from adult interpretations are attributable to
children’s deficiency in one of these areas. For syntax, for example, if a child is unable to
produce any verbal compound forms that reflect the merging of the verb with the modifier
noun as the internal argument, the child is predicted to allow a much broader range of
interpretations for verbal compounds than adults, who only allow a verb-internal argument
interpretation. For lexical semantics, for example, if a child encounters a new object, but
does not know that it has a function, then the child is not predicted to interpret the
compound in a functional way, even though an adult who is aware of an object’s function
may very well interpret a compound with that object in a functional way.
7

1.3 Relevance
This dissertation informs a number of areas. First, the results of the studies provide
external evidence for the structure of compounds. Assuming a strong version of the
Continuity Hypothesis (Pinker 1984), we can use the non-adult forms that children produce
when acquiring compounds as evidence for steps in the adult derivation. Given the
constrained nature of the non-adult forms that have been found in previous studies
examining the acquisition of compound syntax (Clark et al. 1986, Nagpal and Nicoladis
2009, Gamache and Schmitt to appear), this type of data gives us strong evidence for a
particular derivation of compounds.
The results of the studies in this dissertation also provide evidence for the
components of the system that adults use when interpreting compounds. While syntax and
word meanings are always an assumed part of interpretation, I argue that natural class
information is relevant to determine the most appropriate relationship for different
categories of nouns in the absence of other contextual information. This system for
interpreting compounds improves our ability to predict what novel compounds mean, a
current topic in natural language processing work (Ó Séaghdha 2008).
Finally, the results of this dissertation bear on our understanding of how children
acquire their first language. First, the studies provide evidence suggesting that children are
very systematic in their acquisition of compounds syntax. While many permutations of
verbal compounds are logically possible, children produce a small subset of these, which are
consistent with a particular derivation for verbal compounds. Second, the studies support
the idea that, even in the absence of strong syntactic information guiding interpretation, the
interpretation of novel compounds is not generated by simple analogy. That is, children use
various aspects of what they know about words, as opposed to a single aspect of their
8

knowledge about a particular word, to interpret compounds. These findings support a
generative theory of language acquisition (e.g., Chomsky 1968), where the grammar is
continuous from childhood to adulthood (Pinker 1984).
1.4 Organization of Chapters
This dissertation is organized into two parts. The first part of the dissertation
investigates the system that adults use for interpreting compounds. Chapter 2 begins by
outlining previous research on the syntax and meaning of noun-noun compounds and verbal
compounds in the adult grammar. Finding that the proposed theories are insufficient on
their own to develop a comprehensive system of compound interpretation, I propose a
system for compound interpretation that draws on the strengths of multiple proposals and
lay out the hypotheses and predictions of this system.
Chapter 3 presents the results of two studies that examine the hypotheses and
predictions put forth by the compound interpretation system outlined in Chapter 2. The first
study examines adults’ use of syntactic and lexical semantic information in the interpretation
of noun-noun compounds and verbal compounds. The second study examines adults’
interpretation of verbal compounds, specifically examining their ability to assign
interpretations to verbal compounds that should be ruled out by verbal compound syntax.
The second part of this dissertation investigates how child acquire the system that
adults use to interpret compounds. Chapter 4 outlines previous research on the acquisition
of compounds and the hypotheses and predictions for children’s acquisition of compound
interpretations, based on the adult system that they are targeting and our current knowledge
of children’s acquisition of compounds.
Chapter 5 presents three studies on the acquisition of noun-noun compounds. The
first study, a corpus study, examines noun-noun compounds in children’s speech and child9

directed speech in order to determine at what level children’s compound production is
influenced by their input (item level, natural class level). The second study, an elicited
production task, examines whether or not children attribute noun-noun compound syntax to
the same range of relationships as adults. Finally, the third study, a forced-choice
interpretation task, examines how children interpret noun-noun compounds given the lexical
semantic properties of the nouns involved. Chapter 5 closes by summarizing the acquisition
of noun-noun compounds, given the new data, and discusses the findings in terms of
different theories of language acquisition.
Chapter 6 presents the results of two studies on the acquisition of verbal
compounds. The first study, an elicited production task, examines how children acquire the
syntax of verbal compounds by analyzing the non-adult forms that they produce. The
second study, a forced-choice interpretation task, investigates children’s ability to assign
interpretations to verbal compounds that should be ruled out by the adult verbal compound
syntax. The results of the elicited production task and interpretation task are then compared
to determine if children’s syntactic knowledge predicts their interpretation of verbal
compounds. Chapter 6 closes by summarizing the acquisition of verbal compounds, given
the new data, and discusses the findings in terms of different theories of language
acquisition.
Chapter 7 reviews the findings from this dissertation and closes with the implications
of these findings on both theoretical and acquisition research.

10

CHAPTER 2: LINGUISTICS BACKGROUND
2.0 Introduction
The first part of this dissertation examines adults’ interpretation of noun-noun
compounds and verbal compounds, with the goal of sketching the system that adults use to
determine the meaning of compounds. I begin this chapter by making a couple of
assumptions about the nature of the linguistic system and I also provide some basics about
compounds. Next, I review the literature on adults’ interpretation of compounds. Building
on previous literature, I outline a proposal that makes crucial use of lexical information in
the interpretation of compounds. Finally, I discuss the hypotheses and predictions of this
proposal.
2.1 Basic assumptions
In this dissertation I assume a Distributed Morphology (Halle and Marantz 1993)
approach to syntax. Under this approach, syntax and morphology are a single, generative
component. As such, compounds are generated in the syntax and are subject to all of the
constraints therein (Harley 2009).
The second assumption is that lexical items have internal structure that organizes key
information about their physical properties and the way in which they interact with the world
(Pustejovsky 1995). The lexical structure for a lexical item is based on a speaker’s experience
with that lexical item. Because of this, the lexical structure for the same lexical item may
differ from speaker to speaker, based on their experience with it.
The final assumption is that the grammar and the parser are the same system (Lewis
and Phillips in press). That is, the same set of rules that exists in the grammar is the set of
rules that guides the parser. This runs contrary to a two-system model where the grammar
and sentence production and comprehension operate over two separate, but necessarily
11

similar, sets of rules (e.g., Ferreira and Patson 2007, Trotzke, Bader, and Frazier 2013).
Given the assumption of a one-system model, processing-related findings, such as evidence
of a garden path, can be equated back to the nature of the grammar.
2.2 Compound Basics
Compounds loosely fall into two categories: endocentric compounds and exocentric
compounds. Endocentric compounds have a clear head noun that defines the syntactic
category (noun, verb, adjective, etc.) and semantic category (dog, fish, table) of the compound.
If the head of an endocentric compound is a noun, the entire compound is treated
syntactically as a noun. If the head of an endocentric compound is door, then the entire
compound is a type of door. In English, compounds are mostly right-headed (Williams
1981) and therefore the right-hand element determines the category of the referent. The
modifier, which is the left-hand element, restricts that referent to a particular subcategory
(Downing 1977). For example, dog-house, cat-house, and bird-house all refer to types of houses,
while house-dog, house-cat, and house-bird refer to different types of animals and not types of
houses. In English, noun-noun compounds and synthetic compounds can both be thought
of as endocentric compounds.
Exocentric compounds are defined by their lack of a clear head, due to the
misalignment of their syntactic and semantic properties (Bisetto and Scalise 2005).
Exocentric compounds sometimes have a syntactic head or a semantic head, but these
features are not present on a single lexical unit. Take for example yellow-jacket. Yellow-jacket
has a syntactic head; jacket is a noun and yellow-jacket is treated syntactically as a noun. Yellowjacket, however, does not have a semantic head; a yellow-jacket is not a jacket or a yellow.
Therefore, yellow-jacket, and adjective-noun compounds like it, are considered exocentric. If
we look to a compound like lava-platos (Spanish, lit. clean-dishes, “dishwasher”), we see that it
12

lacks both a syntactic head and a semantic head. Syntactically, platos (“dishes”) is a noun and
lava-platos is a noun, which suggests that platos could serve as the syntactic head, but platos is
marked as plural and lava-platos can be singular. That is, one dishwasher is not *lava-plato and
lava-platos does not necessarily refer to more than one dishwasher. Therefore, platos fails to be
a proper syntactic head. Semantically, lava-platos refers to a machine, but neither lava nor
platos refers, on its own, to a machine and, as such, lava-platos also lacks a semantic head.
Exocentric compounds’ lack of consistent head information makes them an interesting, but
separate1, inquiry for language acquisition, which will be left aside for the purposes of this
dissertation.
2.3 Deriving Interpretations
How do we define the meaning of endocentric compounds like noun-noun
compounds or verbal compounds? If we take as our starting point a basic equation of
compositionality (Frege 1884), the meaning of a complex expression is determined by the
meaning of its parts (word meanings) and the way in which those parts are put together
(syntax), then we must first understand how syntax influences compound interpretations.
2.3.1 The Syntax of Compounds
For the purposes of this dissertation, we are interested in a particular aspect of the
syntax of compounds: the type of relationship that compound structures establish between
their constituents. Given the breadth of work on compounds, I will focus on only the work
that is directly relevant to this aspect of compounds.
2.3.1.1 The Syntax of Noun-Noun Compounds
The proposed structures for noun-noun compounds can be divided into two classes:
structures involving movement and structures involving no movement. In the movement
It may be possible to treat exocentric compounds as cases of metonymy and explain them in a similar manner
to the endocentric compounds explored in this dissertation.
1

13

structure, shown in (2a), the modifier noun mouse merges with the head noun trap and then
moves to the specifier of a functional element that mediates the relation between the two
nouns (Roeper and Snyder 2005, Delfitto, Fábregas, and Melloni 2009, Harley 2009). In the
non-movement structure, shown in (2b), the modifier noun mouse adjoins to the head noun
trap, which is sometimes argued to be mediated by a functional element (Di Sciullo 2005,
Mukai 2008). Evidence for the functional element is apparent in languages such as Dutch,
which have linking elements (LE) that appear between the constituents of a compound
(Delfitto et al. 2009). For example, pan-cake in Dutch is pan-en-koek (lit. pan-LE-cake).
(2) a.

b.
n

n
mouse

n
f

f
Ø

n
mouse
n

n
trap

f
f
Ø

n
trap

n
mouse

While there is a question of whether or not movement is necessary for noun-noun
compounds, the important aspect of these structures for this dissertation is the way in which
the nouns are merged together. It can be seen from both (2a) and (2b) that the structure of
noun-noun compounds itself does not define the relationship between its constituents. That
is to say, neither the noun-noun compound structure with movement nor the structure
without movement explains why we should treat mouse in mouse-trap as the thing that is
caught by the trap, rather than the shape of the trap, or the agent who built the trap, on the
assumption that trap does not assign a role to mouse, at least not obligatorily.

14

2.3.1.2 The Syntax of Verbal Compounds
The proposed structures for verbal compounds also fit into two classes. In one type
of structure, shown in (3a), the verb and internal argument merge first, followed by the
merging of –er into the structure. The verb then incorporates into –er and the internal
argument moves up to the specifier of V-er (Gamache and Schmitt to appear). In the
alternate type of structure, shown in (3b), the verb and internal argument merge first,
followed by the incorporation of the internal argument into the verb, and finally the
incorporation of this complex element into –er (Roeper and Siegel 1978; Harley 2009; Borer
2012).
(3) a.

b.
n

n

n
mouse

n

n

n
v
catch

v
n
er

v
catch

v
n
mouse

n
mouse

n
er
v
catch

v
v

n
mouse

n
mouse
v
catch

What is apparent from both of these structures is that a key aspect of verbal
compound syntax is the merging of the verb with the modifier noun as its internal argument.
Unlike the structure for noun-noun compounds, the structure for verbal compounds does
constrain the interpretation of verbal compounds; since the verb assigns a role to mouse.
Since mouse receives the role from the verb, it must be interpreted as the theme of the verb
(Hale and Keyser 1993). The modifier noun of verbal compounds cannot be interpreted in a
way denoting any other relationship to the verb. For instance, a towel-cleaner must be a person
15

or machine who cleans towels (internal argument), and cannot be a person or machine that
cleans things using a towel or cleans things on a towel (adjunct).
2.3.1.3 Summary of the Role of Syntax in Compound Relationships
The syntax of noun-noun compounds and verbal compounds differ substantially in
the extent to which they determine the relationship in compounds. The syntax of verbal
compounds completely constrains the relationship between its constituents, allowing for
only a verb-internal argument interpretation. The syntax of noun-noun compounds does not
constrain the relationship in noun-noun compounds at all. Therefore, we have to go
elsewhere to determine the relationship that exists in noun-noun compounds.
2.3.2 Determining the Relationship in Noun-Noun Compounds
Since the relationship in noun-noun compounds is not defined by their syntax,
where does the relationship arise from? Previous literature has laid out two possibilities. The
first possibility is that there is a set of relationships that exist in compounds, and this list is
drawn on to produce or interpret compounds. We will call this the extrinsic approach to
compound relationships since the relationships are generated external to the constituents of
individual compounds. The second possibility is that the lexical structure of a compound’s
head noun defines the possible relationships that exist between the head and modifier of the
compound, and the lexical structure of the head and modifier taken together determine the
most likely relationship. We will call this the intrinsic approach since the relationships are
generated internal to the lexical structure of the compound in question. Each approach is
discussed in turn below.
2.3.1.1 Extrinsic Approach to Compounds
Under an extrinsic relationship approach, a predefined set of relationships is posited
to exist between the constituents in a compound (Lees 1960, Levi 1978, Warren 1978). Take,
16

for example, Levi’s (1978) system of possible compound relationships, shown in Table 1. In
this system, compounds are interpreted with one of 12 possible relationships. For example,
mouse-trap would be defined as a trap for mice where ‘for’ is the extrinsic relationship. Levi
argued that these relationships are generated in the base form of a compound, which is
something like a relative clause. These relationships are established by recoverably deletable
predicates which exist in the base form of compounds, something akin to a relative clause,
and are deleted in the transformation process. Levi argued that it is this set of predicates
alone that can be deleted in the transformation process, which restricts the set of possible
compound interpretations.
Table 1. Levi's (1978) Noun-Noun Compound Relationships
Relationship
‘cause’ (first element subject of cause)
‘cause’ (first element object of cause)
‘have’ (first element subject of have)
‘have’ (first element object of have)
‘make’ (first element subject of make)
‘make’ (first element object of make)
‘use’ (instrumental)
‘be’
‘in’
‘for’
‘from’
‘about’

Examples
drug-death, viral-infection
tear-gas, mortal-blow
lemon-peel, feminine-intuition
picture-book, industrial-area
daisy-chain, consonantal-pattern
honey-bee, sebaceous-glands
steam-iron, solar-generator
soldier-ant, consonantal, segment
field-mouse, marital-sex
horse-doctor, avian-sanctuary
olive-oil, solar-energy
abortion-vote, criminal-policy

The extrinsic approach falls short in a couple of ways. First, why should compounds
be restricted to this set of relationships and not others? Levi’s account restricts the possible
interpretations to a particular set, but her account does not explain why compounds exhibit
this specific pattern. Second, the list of relationships, while a good description of the ways in
which people paraphrase compounds, does not explain why people interpret compounds in
the way that they do. That is, how do speakers know, without explicit and directed context,

17

which relationship is intended in a novel compound? It must be associated to some sort of
preference, but Levi does not specify where this preference arises from.
2.3.1.2 Intrinsic Approach to Compounds
Under an intrinsic approach, the relationships that exist in compounds are generated
from the lexical structure of the head noun in compounds (Johnston and Busa 1999).
Johnston and Busa proposed that the machinery in Pustejovsky’s (1995) Generative Lexicon
system be extended to define the possible relationships that exist in compounds.
Pustejovsky’s system defines four levels of lexical representation, listed in (4). Each level of
lexical representation specifies a different type of information about a lexical item.
(4) Levels of Lexical Representation in Pustejovsky’s (1995) System
a. Argument Structure: The behavior of a word as a function, with its arity
specified. This is the predicate argument structure for a word, which indicates
how it maps to syntactic expressions.
b. Event Structure: Identification of the particular event type for a word or phrase:
e.g. as state, process, or transition.
c. Qualia Structure: The essential attributes of an object as defined by the lexical
item.
d. Inheritance Structure: How the word is globally related to other concepts in the
lexicon.
For the purposes of interpreting compounds, Johnston and Busa proposed that the
Qualia Structure of a lexical item, further defined in (5), organizes the different attributes of
an object and how it relates to other lexical items. There are four attributes of a lexical item
that can be modified and therefore four types of relationships that exist in compounds.

18

(5) The Structure of Qualia
a.

Formal Role: that which distinguishes the object within a larger domain -orientation, magnitude, shape, dimensionality, color, position

b.

Constitutive Role: the relation between an object and its constituents, or
proper parts -- material, weight, parts and component elements

c.

Telic Role: purpose and function of the object -- purpose that an agent has in
performing an act, built-in function or aim that specifies certain activities

d.

Agentive Role: factors involved in the origin or “bringing about” of an object
-- creator, artifact, natural kind, causal chain

The Qualia Structure for the lexical item knife is shown in Figure 3. To give an
example of how Johnston and Busa’s system works, consider the types of things that could
be in each of knife’s Qualia Roles. For example, metal-knife is an example of a modification of
the Constitutive Role (a knife made of metal), bread-knife is a modification of the Telic Role
(a knife for cutting bread), and machine-knife is a modification of the Agentive Role (a knife
made by a machine).
Figure 3. The Qualia Structure of knife
knife
QUALIA =

FORMAL = x
CONSTITUTIVE = {blade, handle, metal…}
TELIC = cut_act
AGENTIVE = make_act

Modified from Johnston and Busa (1999, pg 80).

While Johnston and Busa only begin to outline how the Generative Lexicon could be
applied to compounds, there is at least one other aspect of Pustejovky’s system that furthers
our understanding of compound semantics. Types, which categorize lexical items into
19

groups, are created by unifying the qualia structure of lexical items that share particular
features. For example, Pustejovsky gives the example of food as a type. This type is created by
the unification of the formal specification “physical object” and the telic specification “to
eat”. See Figure 4 for the qualia structure of food. Any lexical item that shares the properties
would be considered a member of the food type and would behave similarly to other items in
this group under particular conditions. A single lexical item is categorized according to a
number of types. For example, clementine could be categorized as a food type, a physical object
type, and a natural kind type, and therefore is expected to pattern with each of these types,
depending on the context.
Figure 4. The Qualia Structure of the Unified Type food
food
QUALIA =

FORMAL = physical object
TELIC = eat_act

Modified from Pustejovksy (1995, pg 146).

Types serve to create classes of words that are predicted to behave in similar ways.
We empirically know that adults use natural classes in their interpretation of compounds.
Downing (1977)2 found that participants in her study systematically interpreted compounds
differently based on the class of a compound’s head noun (see Table 2).

2

See also Leonard (1984) and Štekauer (2005).

20

Table 2. Most Frequent Relationship Types by Head Noun Class
Class
Humans

Relationships
Occupational
Sexual Identity
Racial Identity
Animals
Appearance
Habitat
Plants
Appearance
Habitat
Natural objects
Composition
Origin
Location
Synthetic objects
Purpose
Modified from Downing (1977, pg. 831).

Examples
police demonstrators
women officers
Negro women
giraffe bird
Salt Creek coyotes
trumpet plants
Texas roadside flowers
granite out-croppings
cow hair
Montana beach
banana fork

This finding closely aligns with related research on speakers’ categorization of natural classes
and the properties that define them (e.g., Gelman 1988, Rips 1989). Artifacts, for example,
tend to be defined in terms of their function and natural kinds are generally defined in terms
of their appearance or origin.
The intrinsic approach is an improvement on the extrinsic approach for three
reasons. First, the intrinsic approach explains why there is a particular set of relationships in
compounds; particular relationships exist in compounds because these relationships are
specified by the Qualia Structure of compounds’ head nouns. Moreover, the intrinsic
approach explains the existence of the set of relationships in compounds without adding any
machinery to our notion of compositionality. This approach to compound interpretation
suggests that, rather than simple word meanings, we need to consider a generative lexical
semantics with structured lexical information that defines the way in which words interact
with one another. Finally, the intrinsic approach explains why speakers choose to interpret
compounds using particular relationships instead of others, given that multiple relationships

21

are possible. Lexical item are categorized into many types, or classes, and these classes are
interpreted in particular ways in particular contexts.
2.4 A Proposed System for Compound Interpretation
The goal of the first half of this dissertation is to develop a comprehensive system of
compound interpretation. Considering syntax alone, we were able to determine the sole
relationship that exists between the verb and modifier in a verbal compound; the syntax of
verbal compounds requires the verb and modifier noun to be interpreted with a verbinternal argument relationship. Syntax does not determine the relationship that exists in
noun-noun compounds, however, aside from assignment of the head. We then considered
previous proposals to generate the relationships that exist in noun-noun compounds. The
intrinsic approach (Johnston and Busa 1999), which extends Pustejovsky’s Generative
Lexicon to compounds, provided a more defined notion of word meanings, which
prescribes the possible relationships that are available for modification and the way in which
words are organized into various natural classes.
In the production and comprehension of compounds, there are three pieces of
knowledge that must be in place. First, a speaker must know the syntax of a compound. This
part of the system determines any syntactic constraints that must be imposed on the
interpretation of the compound. For example, the syntax of verbal compounds constrains
the interpretation of their constituents to a verb-internal argument relationship.
Second, a speaker must have a structured lexical entry for the head noun of a
compound that defines the ways in which the noun interacts with others concepts. This part
of the system determines the range of relationships that are possible for a compound with a
specific head noun. For a given relationship to be possible the Qualia Role corresponding to
that relationship must be specified on the head noun. For example, the Telic Role of scissors
22

is specified as “to cut” while the Telic Role of salamander is likely unspecified as salamanders
are not functional.
Finally, from the Qualia Structure, a speaker must know which information is
privileged according to the natural classes of the constituents in a compound and the
tendencies of these natural classes to behave in particular ways. This part of the system
determines which Qualia Role is the most likely to be used in interpretation of the
compound. For example, artifacts are most likely to have a purpose and therefore access the
Telic Role and animate natural kinds are most likely to be defined by their appearance and
therefore access the Formal Role. For this reason a strawberry-bowl is most likely to be a type
of bowl that is used for strawberries, while a strawberry-fish is most likely to be a fish that
looks like a strawberry.
Figure 5 lays out the way in which a speaker would interpret a novel compound according to
this system. In Step 1, a speaker identifies the morphemes that make up a compound. In
parallel, speakers begin constructing interpretations based on the syntax and semantics. In
Steps 2 and 3 in the syntax, speakers parse the morphemes into a compound structure and
assign any relationships to those morphemes that are imposed by the syntax. I make an
assumption here that given a transitive verb, a speaker always tries to first parse the string
noun-verb-er with verbal compound syntax. In Step 2 of the semantics speakers access the
lexical structure of the head noun and determine the set of possible relationships. For nounnoun compounds this is the second noun. For verbal compounds this is most simply the
verb-er noun, but more specifically it could be considered the morpheme –er which is
specified by the verb for its Telic Role. In Step 3 speakers determine the semantic natural
class of the constituents in order to determine which of the possible Qualia Roles is the most
likely to be picked up. This Qualia Role is then checked to determine if the modifier noun is
23

an appropriate fit for that Role. In Step 4, the speaker determines if the relationships
generated by the syntax and semantics are compatible. In the case of noun-noun
compounds, the syntax only generates an unspecified, abstract relationship between the
constituents that is compatible with any interpretation generated by the lexical semantics. In
the case of verbal compounds, the relationships generated by the syntax and lexical
semantics can either be compatible, when a Telic relationship is generated by the lexical
semantics, or incompatible, when any other relationship (Agentive, Constitutive, Formal) is
generated by the lexical semantics. In cases of incompatible relationships, the compound
must be reanalyzed by using a different part of the head nouns’ Qualia Structure.

24

Figure 5. Proposed Compound Interpretation System

Parse
morphemes
into syntactic
structure.
Identify
morphemes.

Step 1

Assign
syntactic
relationships.

Match.
Interpret and
move on.
Reconcile.

Access
qualia
structure of
head noun.

Assign
semantic
relationships
by natural
classes.

Step 2

Step 3

25

Mismatch.
Reanalyze.
Step 4

Step 5

Figure 6 shows how a speaker would determine the meaning of four different
compounds in this system. The first example, table-soap, is a noun-noun compound with an
artifact head noun that has a canonical Telic interpretation. The morphemes table and soap
are parsed into noun-noun compound syntax. The head noun soap is specified for all four
Qualia Roles, but its status as an artifact favors the Telic relationship in compounds. The
Telic role of soap is checked first to see if table can reasonably modify soap in a functional
way. As a function of soap is to clean, and a table is a reasonable thing to clean, a Telic
relationship is satisfied in the lexical semantics. The relationships proposed by the syntax and
semantics are found to be compatible and the speaker successfully interprets table-soap as
soap for cleaning tables.
The second example in Figure 6, bleach-soap, is an example of a noun-noun
compound with an artifact head noun that has a non-canonical, Constitutive interpretation.
The morphemes bleach and soap are parsed into noun-noun compound syntax. The head
noun soap is specified for all four Qualia Roles, but its status as an artifact favors the
functional, Telic relationship in compounds. The Telic role of soap is checked first to see if
bleach can reasonably modify soap in a functional way. As bleach is not a reasonable thing to
clean, the Telic Role is not satisfied and a different Qualia Role is considered. Finding that
bleach can reasonably modify soap in a part/whole way, a Constitutive relationship is satisfied
in the lexical semantics. The relationships proposed by the syntax and semantics are found
to be compatible and the speaker successfully interprets bleach-soap as soap that contains
bleach.
The third example in Figure 6, table-cleaner, is an example of a verbal compound that
converges on the first attempt to determine its meaning. The morphemes in table-cleaner are
parsed into verbal compound syntax and a verb-internal argument relationship is assigned to
26

clean and table by the syntax. As the function of a cleaner is to clean, and a table is a
reasonable thing to clean, a Telic semantic relationship is satisfied in the lexical semantics.
The relationships proposed by the syntax and semantics are found to be compatible and the
speaker successfully interprets table-cleaner as a person who cleans table.
The final example in Figure 6, bleach-cleaner, is an example of a compound that is first
parsed as a verbal compound, but is then needs to be reanalyzed because, without context, a
verb-internal argument relationship is an unlikely interpretation. On the first pass, the
morphemes in bleach-cleaner are parsed into verbal compound syntax and a verb-internal
argument relationship is assigned by the syntax to clean and bleach. Semantically, the function
of a cleaner is to clean, but bleach is not a reasonable thing to clean. Rather, bleach is a
reasonable thing to use to clean and bleach-cleaner is assigned a formal relationship by the
semantics. The relationships proposed by the syntax and semantics are found to be
incompatible and bleach-cleaner must be reanalyzed. On the reanalysis, the morphemes are
parsed into a noun-noun compound structure, leading to an unspecified, abstract
relationship by the syntax. The lexical semantics again is satisfied by the Formal Role and the
speaker successfully interprets bleach-cleaner as a cleaner that cleans using bleach. Note that a
different reanalysis is possible for bleach-cleaner. Rather than reanalyzing the syntax, the lexical
semantics could have been reanalyzed to force a telic relationship between bleach and cleaner,
leading to the interpretation of someone or something that cleans bleach.

27

syn.
sem.

bleach
clean
-er

syn.

bleach
clean
-er

Telic (functional) relationship
(soap that cleans tables)

[bleach-soap]
noun-noun compound syntax

modifier-head relationship

sem.

noun-verb-er
bleach-cleaner with “bad”
and reanalysis internal
argument noun

table
clean
-er

table = artifact (furniture)
soap = artifact that cleans

bleach = artifact (chemical)
soap = artifact that cleans

Constitutive relationship
(soap that contains bleach)

[[table-clean]-er]
verbal compound syntax
table = artifact (furniture)
clean = verb
-er = human/machine
[[bleach-clean]-er]
verbal compound syntax
bleach = artifact (chemical)
clean = verb
-er = human/machine
[bleach-[clean-er]]
noun-noun compound syntax
bleach = artifact (chemical)
clean = verb
-er = human/machine

table = internal argument,
-er = external argument

sem.

table-cleaner

noun-verb-er
with “good”
internal
argument noun

bleach
soap

Step 3
modifier-head relationship

syn.

bleach-soap

noun-noun
with non-typical
artifact Qualia
Role

table
soap

Step 2
[table-soap]
noun-noun compound syntax

sem.

table-soap

noun-noun
with typical
artifact Qualia
Role

Step 1

syn.

Example Type

sem.

Example

syn.

Figure 6. Examples of Compound Interpretation in Proposed System

28

Telic (functional) relationship
(table-cleaner cleans tables)
bleach = internal argument,
-er = external argument
Formal relationship
(cleaner uses bleach to do
cleaning)

Step 4

Step 5

Match.

Interpret.

Match

Interpret.

Match.

Interpret.

Mismatch.

Reanalyze.
→

Match.

Interpret.

modifier-head relationship
Formal relationship
(cleaner uses bleach to do
cleaning)

2.5 Hypotheses and Predictions for Adults’ Interpretation of Compounds
This system relies on four hypotheses, shown in Table 3, about the role of syntax
and lexical semantics on the interpretation of noun-noun compounds and verbal
compounds. Each hypothesis is discussed in turn, along with any predictions that the
hypothesis makes about adults’ behavior.
Table 3. Hypotheses and Predictions of New Interpretation System
Hypothesis
Hypothesis 1: Syntactic
structure is deterministic in its
assignment of a relationship to
a compound.
Hypothesis 2: The Qualia
Structure of the head noun in
a compound supplies the
range of possible relationships
for that compound.
Hypothesis 3: Semantic
natural classes strongly
influence the most likely
aspect of a lexical structure to
be accessed for compound
interpretation.
Hypothesis 4: The
relationships supplied by the
syntax and semantics must be
reconcilable.

Predictions
Prediction 1: Morphemes that are parsed with verbal
compound syntax are always assigned a verb-internal
argument relationship by the syntax and morphemes that
are parsed with noun-noun compound syntax are always
assigned an unspecified, abstract relationship by the syntax.
Prediction 2: Compounds can only be interpreted with
relationships that are specified in the Qualia Structure of
their head noun. Compounds whose head nouns are nonce
or novel should defer to the lexical structure of the
modifier noun.
Prediction 3: Compounds are interpreted in line with the
tendencies of their natural classes (e.g., artifacts as
functional, animate natural kinds as appearance- or originrelated).
Prediction 4: Compounds composed of the string nounverb-er garden path when the semantics of the string
generates a a non-telic relationship. Compounds composed
of the string noun-noun never garden path because of the
syntactic structure imposed on them.

The first hypothesis is that syntactic structure is deterministic in its assignment of a
relationship to constituents of a compound. This hypothesis predicts that, given a transitive
verb, verbal compound syntax always imposes a verb-internal argument relationship on its
constituents and noun-noun compound syntax always proposes an unspecified, abstract
relationship on its constituents. Verbs that do not require an internal argument, either
because of their transitivity or because of their telic properties (Borer 1994), should be
29

parsed with noun-noun compound syntax and allow a wider range of interpretations than
verbs that do require an internal argument.
The second hypothesis is that the Qualia Structure of the head noun supplies the
possible relationships in compounds (Johnston and Busa 1999). This predicts that the
possible relationships for compounds differ based on the lexical structure their constituents.
Scissors, for example, has a functional specification (to cut), but salamander does not usually
have a functional specification. Therefore, paper-scissors are likely to be scissors that cut paper,
while a paper-salamander is not likely to be a salamander that does something to paper.
Another prediction made by this hypothesis is that words without lexical structure (novel or
nonce words) will defer to the other constituent in the compound, as the lexical structure of
a novel or nonce word is almost or completely unspecified.
The third hypothesis is that semantic natural class membership strongly influences
the most likely aspect of a lexical structure to be accessed for a compound’s interpretation
(Downing 1977). This hypothesis predicts that the mostly likely relationship in a compound
is determined by the natural classes of its constituents. Compounds with artifact head nouns,
for examples, are likely to lead to functional, Telic interpretations, while compounds with
animate head nouns are likely to lead to appearance-related, Formal interpretations. Given
that a single lexical item can belong to multiple classes, for example, scissors refers to both an
artifact and a physical object, no single class can definitely predict the relationship.
Finally, the fourth hypothesis is that the relationships supplied by the syntax and
semantics must be reconcilable. This predicts that the string noun-verb-er causes a speaker to
garden path when the relationship generated by the lexical semantics is anything other than a
Telic interpretation. Recalling the bleach-cleaner example, the syntax assigns a verb-internal

30

argument relationship to clean and bleach, but the lexical semantics assigns a Formal
relationship, forcing a reanalysis of the structure.
2.6 Summary
In this chapter I reviewed previous work on the syntax and interpretation of nounnoun compounds and verbal compounds. A basic notion of compositionality was found to
be sufficient for interpreting verbal compounds, whose syntax defines the relationship
between their constituents. In order to interpret noun-noun compounds, however, we
showed that we must assume a dynamic semantics that considers class membership in the
ordering of possible interpretations. Under this system, different relationships are privileged
for different noun-noun combinations, based on the class membership of those nouns. In
the next chapter I present the results of two studies that test the predictions made by this
system for interpreting novel compounds.

31

CHAPTER 3: ADULT COMPOUND TASKS
3.0 Introduction
In Chapter 2 I sketched a system for how adults use syntactic and lexical semantic
knowledge to interpret compounds. This chapter presents the results of two studies that test
the predictions made by this system. The first study examined how adults interpret
compounds based on their syntactic structure (verbal compounds and noun-noun
compounds), lexical status (known words and nonce words), and natural class (animate
natural kinds, inanimate natural kinds, locative/temporal nouns, humans, artifacts). The
second study examined adults’ ability to assign interpretations to verbal compounds that are
either predicted to be allowed by verbal compound syntax (modifier as internal argument) or
not allowed by verbal compound syntax (modifier as adjunct). The goal of these two studies
was to better understand the way in which adults interpret compounds, such that we can
define the target state for children.
3.1 Adult Study 1: Interpretation of Noun-Noun Compounds and Verbal Compounds
The first study examined adults’ interpretation of noun-noun compounds and verbal
compounds using a free interpretation task. There were two goals of this study. The first
goal was to examine the role of syntax on the interpretation of compounds. The second goal
was to examine the role of lexical semantics on the interpretation of compounds, namely the
role of the Qualia Structure (Johnston and Busa 1999) and natural class tendencies
(Downing 1977). I address the background and predictions for each of these goals in turn.
3.1.1 The Role of Syntax on the Interpretation of Compounds
One goal of this study was to examine the role of syntax in the interpretation of
compounds. More specifically, the goal was to provide evidence that, even in the absence of
lexical information, the syntax of verbal compounds restricts the range of possible
32

interpretations, while the syntax of noun-noun compounds does not constrain the
interpretations. As discussed in Chapter 2, the syntax of verbal compounds defines the
relationship that exists between the constituents; the modifier noun merges as the internal
argument of the verb and, as such, the modifier is interpreted as the theme of the verb. The
syntax of noun-noun compounds, however, does not establish a particular relationship
between its constituents. Rather, the syntax of noun-noun compounds establishes that there
is an unspecified, abstract relationship between the constituents and it is left to another
system, namely lexical semantics, to define the particular relationship.
This study was designed to examine the impact of syntax on the interpretation of
compounds in two ways. First, the stimuli used in the experiment were either noun-noun
strings or noun-verb-er strings, which were predicted to be parsed as noun-noun compounds
and verbal compounds, respectively. This allowed us to compare the range of interpretations
that these two structures lead to. Second, the stimuli were designed to lessen any bias from
prior exposure and lexical structure by (i) randomly pairing the head and modifier nouns and
(ii) including a set of nonce words. This allowed us to abstract away from the behavior of
individual items as much as possible and instead see the behavior of these compound classes
as a whole.
The hypotheses put forth in this dissertation made two predictions about the role of
syntax on the interpretation of compounds. First, noun-verb-er strings were predicted to be
parsed as verbal compounds and interpreted with verb-internal argument relationships. This
was predicted to happen even when the verb in the verbal compound was a nonce verb. For
example, both towel-cleaner and towel-wugger were predicted to be interpreted as a person who
cleans/wugs towels. Second, noun-noun strings were predicted to be parsed as noun-noun

33

compounds and, as such, be interpreted with any of the four relationships specified by
Qualia Structure of the head noun (Agentive, Constitutive, Formal, Telic).
3.1.2 The Role of Lexical Semantics on the Interpretation of Compounds
The second goal of this study was to examine the role of lexical semantics on the
interpretation of compounds. The proposed system uses lexical semantics in two ways. First,
following Johnston and Busa (1999), the Qualia Structure of the head noun in a compound
defines the possible interpretations for the compound. In order for a compound to be
interpreted as associated with a particular relationship made available in the Qualia Structure
(Pustejovsky 1995). Second, following Downing (1977), the tendencies of the natural classes
in a compound determine the compound’s most likely interpretation. For example, artifacts,
which are defined in terms of their function, are likely to lead to Telic interpretations, while
animate natural kinds, which are defined in terms of their appearance or origin, are most
likely to lead to Formal interpretations.
This study was designed to examine the role of lexical semantics in two ways. First,
nonce words were included as a class of nouns in the study. By pairing the nonce class of
words with real natural classes of words, we could determine a baseline for the contribution
of different natural classes in modifier and head noun positions without influence from the
other word in a compound. For example, in the compounds wug-bike or bike-wug, the nonce
word wug allowed us to determine the most likely contribution of the artifact bike as both a
head noun and a modifier noun, without interference from wug. Second, nouns were chosen
for inclusion on their membership in one of five natural classes: animate natural kinds,
inanimate natural kinds, locative/temporal nouns, human nouns, and artifacts. Therefore, we
could systematically examine the impact of different natural classes on the relationship in a
compound. For example, we could determine how likely a compound with an artifact head
34

noun is to receive a functional, Telic interpretation as compared to how likely a compound
with an animate natural kind head noun is to receive the same type of interpretation. For the
full list of noun classes used in the study, see Table 4.
Table 4. NNC and VC Interpretation: Noun Classes
Class
Verb-er
Nonce-er
Nonce
Animate Natural Kind
Artifact
Human
Inanimate Natural Kind
Locative/Temporal

Defining Features
transitive verb with –er affix
English nonce word with –er affix
English nonce word
naturally occurring, sentient
functional, manmade objects
human
naturally occurring, non-sentient
defined by physical or temporal space

Example
drinker, carrier
daxer, lainter
wug, moop
cow, dolphin
spoon, box
boy, lawyer
rock, wind
circus, summer

Notice that this is just one set of classes. On top of these classes there are
simultaneous classifications that unify some lexical items that are members of different
classes under this system, and also simultaneous classifications that divide lexical items that
are members of the same class under this system. For example, stone and scissors exist in
different classes under the system used here, animate natural kind and artifact, respectively,
but under a class of physical objects, they would be grouped together. It should be kept in
mind that this classification system is by no means the only system at work.
The hypotheses put forth in this dissertation made three predictions about the role
of lexical semantics on the interpretation of compounds. First, known and nonce head
nouns of noun-noun compounds were predicted to behave in different manners. Known
head nouns were predicted to be interpreted in line with their inherent lexical properties.
Compounds whose head nouns do not have certain specifications were predicted to not
receive those types of interpretations. For example, if salamander did not have a functional
specification for a speaker, then a compound with the head noun salamander should not have

35

been interpreted in a functional way, in the absence of a special context. Nonce head nouns,
on the other hand, were predicted to be less constrained in their interpretations since they
can only be interpreted with respect to the modifier noun that they are paired with. For
instance, since wug is a nonce word, when it is paired with bike in bike-wug, it could lead to any
interpretation that bike can reasonably fit into. For example, a bike-wug could be a wug made
by a bike (agentive), a wug that is a part of a bike (constitutive), a wug that looks like a bike
(formal), or a wug that makes bikes (telic).
The second prediction was that compounds composed of different natural classes
should be interpreted using different relationships, in line with the tendencies of those
natural classes. For example, compounds with artifact head nouns such as scissors, spoon, and
blanket were predicted to lead to Telic interpretations. Compounds with animate natural
kinds such as dog, horse, and chipmunk were predicted to lead to Formal interpretations.
The third prediction was that interpretations were more likely to appear in
contrastive cases, since by their nature, compounds introduce a restriction on the
interpretation of the head noun. For example, bike-lock contrasts with other types of locks.
Certain Qualia properties do not restrict the interpretation of the head because without
context, a default interpretation is the only option.
3.1.3 Method
3.1.3.1 Participants
25 adults (20 F, 5 M) were recruited from an undergraduate linguistics course at
Michigan State University. The subjects received extra credit in their course for their
participation.

36

3.1.3.2 Materials and Procedure
Lists of words were generated based on the categories in Table 4. The categories
were fully crossed to create 64 item types. Each participant received two of each item type
for a total of 128 items. The compounds were generated by randomly pairing words from
each category. For any given combination of words, one set of participants saw the
compound as word1-word2, and another set of participants saw the compound as word2word1. For example, some participants saw rock-circus and some circus-rock. A total of three
pairings were generated and also reversed (turning heads into modifiers and modifiers into
heads) for a total of six lists that participants were randomly assigned to and 768
compounds. See Appendix A for the full list of nouns used in this study.
Participants were tested individually on a PC using E-Prime experimental software.
Participants were presented with novel compounds and instructed to give short descriptions.
Dog-house and book-shelf were given as examples of compounds, with “a house that dogs live
in” given as an example description for dog-house. All compounds were presented in the frame
“What’s a ___?” with a box below the prompt for the participants to give their descriptions.
3.1.3.3 Coding and Analysis
Prior to coding, all the items that had either a nonce modifier or a nonce-er modifier
(N = 800) were removed from the analysis due to a high proportion of adjectival
interpretations (for example, wug-summer as “a summer that is wug”).
All other items (N = 2400) were coded according to the scheme in Table 5, which
was loosely based on the scheme in Johnston and Busa (1999) for Pustejovky’s (1995) Qualia
Structure. Any descriptions that were coded as unusable (21.7%, N = 521), reversal (9.1%, N
= 218), or no response (<1%, N = 5) were excluded from the analysis. A total of 69% (N =
1656) of the coded items were used in the analysis.
37

All responses were coded by the author. Additionally, 10% of the responses were
coded by a second trained coder. Cohen’s kappa (Cohen 1960), a measure of inter-rater
reliability, was found to be .64, which is considered to be a good level of agreement (Altman
1990).
Table 5. NNC and VC Interpretation: Coding Scheme
Relation
Agentive
Constitutive
Formal
Telic
Reversal
Unusable

No
Response

Defining Features
relation that involves the
creation of an object
part/whole relation, including
material, parts, and contents
descriptive relation, including
age, location/origin, shape
function, capability

Example Description
“a hole created by a bullet” for bullet-hole

head and modifier switched in
description
ambiguous – cannot reliably be
placed in a single category

“a couch that boys sit on” for couch-boy

“a book about cowboys” for cowboy-book
“a lion that lives in the mountains” for
mountain-lion
“a knife for cutting bread” for bread-knife

“a vacuum for birds” is ambiguous
between Telic (used to suck up birds) and
Formal (used by birds) for bird-vacuum
metaphorical – does not directly “a person who spends too much time in
relate the two lexical items
the sun” for lobster-eater
wrong word(s) – one or more
“a coach of baseball” for baseball-couch
real words was misread
unclear – description given by
“likes to eat coconuts” for ocean-hugger
participant is undecipherable
no response entered by
--participant

The results of the study were analyzed in two ways. First, to examine the role of
syntax on the interpretation of compounds, all items were collapsed into four categories:
verbal compounds with known head nouns, verbal compounds with nonce head nouns,
noun-noun compounds with known head nouns, and verbal compounds with nonce head
nouns. See Table 6 for the number of items in each category. The items were then recoded
to isolate each relationship (Agentive, Constitutive, Formal, Telic) as the dependent variable

38

for four separate ANOVAs3. For example, for the Agentive ANOVA, items with Agentive
interpretations were coded as “1” and all other items (Constitutive, Formal, Telic) were
coded as “0”. For each of the four analyses, the proportion of the relationship in question
(Agentive for the Agentive ANOVA) served as the dependent measure for a 2 Syntax
(Noun-Verb-er, Noun-Noun) X 2 Head Lexical Status (Known, Nonce) repeated-measures
ANOVA. Note that the Syntax variable refers to likely syntax that is applied to these strings:
noun-noun compounds syntax for noun-noun strings and verbal compound syntax for
noun-verb-er strings. This is not necessarily the syntax that participants use to interpret these
strings. For example, noun-verb-er can be interpreted as either a verbal compound or a
noun-noun compound (see Experiment 2).
Table 6. Count of Items for Syntax X Head Lexical Status Analysis

Known Head
Nonce Head
Total

Noun-Noun
String
1038
184
1222

Noun-Verb-er
String
227
207
434

Total
1265
391
1656

Second, to examine the role of semantic natural class on the interpretation of
compounds, all noun-noun compounds constructed of known head and modifier nouns
(animate, inanimate, locative/temporal, human, artifact) were analyzed. See Table 7 for the
number of items in each natural class. As in the above analysis, the items were recoded to
isolate each relationship (Agentive, Constitutive, Formal, Telic) as the dependent variable for
four separate ANOVAs. For each of the four analyses, the proportion of the relationship in
question served as the dependent measure for a 5 Head Class (Animate, Inanimate,
Locative/Temporal, Human, Artifact) X 5 Modifier Class (Animate, Inanimate,

The author recognizes that mixed effects modeling may be a more appropriate way to analyze this type of
data. This will be left for future research. This comment holds for all ANOVAs in this dissertation.
3

39

Locative/Temporal, Human, Artifact) repeated-measures ANOVA. A Bonferroni correction
was used for all post-hoc tests to adjust for multiple comparisons.
Table 7. Count of Items for Head Natural Class X Modifier Natural Class Analysis
Modifier Natural Class
Locative/
Animate Inanimate Temporal Human Artifact Total
Head
Animate
47
42
42
33
30
194
Natural Inanimate
31
37
32
22
23
145
Class
Loc/Temp
33
32
28
39
32
164
Human
40
29
37
39
40
185
Artifact
40
48
41
40
43
212
Total
191
188
180
173
168
900
3.1.4 Results
3.1.4.1 The Role of Syntax on Interpretation
Table 8 show the distribution of relationships for noun-noun and noun-verb-er
strings with known and novel head nouns.
Table 8. Proportion (and Count) of Items by Relationship, Syntax, and Lexical Status

Known Head
Agentive
Nonce Head
Agentive Overall
Known Head
Constitutive Nonce Head
Constitutive Overall
Known Head
Formal
Nonce Head
Formal Overall
Known Head
Telic
Nonce Head
Telic Overall

Noun-Noun
String
.02 (22)
.06 (11)
.03 (33)
.05 (50)
.11 (20)
.06 (70)
.74 (770)
.63 (115)
.72 (885)
.19 (196)
.21 (38)
.19 (234)

40

Noun-Verb-er
String
.00 (0)
.00 (0)
.00 (0)
.00 (0)
.01 (2)
.00 (2)
.28 (64)
.36 (75)
.32 (139)
.72 (163)
.63 (130)
.68 (293)

Overall
.02 (22)
.03 (11)
.02 (33)
.04 (50)
.06 (22)
.04 (72)
.66 (834)
.49 (190)
.62 (1024)
.28 (359)
.43 (168)
.32 (527)

3.1.4.1.1 The Role of Syntax on Agentive Interpretations
A 2 Syntax (Noun-Noun, Noun-Verb-er) X 2 Head Lexical Status (Known, Nonce)
ANOVA was carried out on the proportion of Agentive responses. The main effect of
Syntax was significant (F(1, 24) = 22.117, p < .001); Noun-Noun received Agentive
interpretations significantly more often than Noun-Verb-er. The main effect of Lexical Status
was marginally significant (F(1, 24) = 4.121, p = .054); compounds with Nonce heads
received Agentive interpretations marginally more often than compounds with Known
heads. The interaction between Syntax and Lexical Status was also marginally significant
(F(1, 24) = 4.121, p = .054); Noun-Verb-er never received Agentive interpretations, whether
the head was Known or Nonce, while Noun-Noun received slightly more Agentive
interpretations when the head was Nonce as opposed to Known.
3.1.4.1.2 The Role of Syntax on Constitutive Interpretations
A 2 Syntax (Noun-Noun, Noun-Verb-er) X 2 Head Lexical Status (Known, Nonce)
ANOVA was carried out on the proportion of Constitutive responses. The main effect of
Syntax was significant (F(1, 24) = 26.197, p < .001); Noun-Noun received Constitutive
interpretations significantly more often than Noun-Verb-er. The main effect of Lexical Status
was also significant (F(1, 24) = 5.337, p = .030); compounds with Nonce heads received
Constitutive interpretations significantly more often than compounds with Known heads.
The interaction between Syntax and Lexical Status was not significant (p = .219).
3.1.4.1.3 The Role of Syntax on Formal Interpretations
A 2 Syntax (Noun-Noun, Noun-Verb-er) X 2 Head Lexical Status (Known, Nonce)
ANOVA was carried out on the proportion of Formal responses. The main effect of Syntax
was significant (F(1, 24) = 163.749, p < .001); Noun-Noun received Formal interpretations
41

significantly more often than Noun-Verb-er. The interaction between Syntax and Lexical
Status was also significant (F(1, 24) = 13.799, p = .001); Noun-Noun strings with Known
heads received significantly more Formal interpretations than Noun-Noun strings with
Nonce heads, while Noun-Verb-er strings with Known heads received significantly less
Formal interpretations than Noun-Verb-er strings with Nonce heads. The main effect of
Lexical Status was not significant (p = .829).
3.1.4.1.4 The Role of Syntax on Telic Interpretations
A 2 Syntax (Noun-Noun, Noun-Verb-er) X 2 Head Lexical Status (Known, Nonce)
ANOVA was carried out on the proportion of Telic responses. The main effect of Syntax
was significant (F(1, 24) = 222.756, p < .001); Noun-Verb-er received Telic interpretations
significantly more often than Noun-Noun. The interaction between Syntax and Lexical
Status was also significant (F(1, 24) = 4.880, p = .037); Noun-Noun strings with Known
heads and Noun-Noun strings with Novel heads received Telic interpretations to the same
degree, while Noun-Verb-er strings with Known heads received more Telic interpretations
than Noun-Verb-er strings with Nonce heads. The main effect of Lexical Status was not
significant (p = .214).
3.1.4.1.5 Summary of the Role of Syntax on Interpretation
For each type of relationship, Agentive, Constitutive, Formal, and Telic, the syntax
of the compound was a significant predictor of how often it would be used. Agentive,
Constitutive, and Formal relationships were used more often for noun-noun strings, while
Telic relationships were used more often for noun-verb-er strings. Importantly, this
relationship held regardless of whether or not the head of the compounds was a familiar,
known word or an unfamiliar, nonce word, suggesting that the syntax alone drove this

42

pattern, rather than lexical structure or prior exposure to compounds containing the same
lexical items.
The results of these analyses also provide evidence that particular relationships are
more likely when there is no constraining lexical information. Both Agentive and
Constitutive relationships were more frequent for noun-noun strings with nonce heads than
for noun-noun strings with known heads. This likely reflects the ability for nonce words to
be interpreted with respect to the modifier noun in any way that the modifier noun prefers.
Compounds with known head nouns, on the other hand, are subject to their Qualia
Structure and must reconcile the preferences of the head noun with the modifier noun.
3.1.4.2 The Role of Semantic Natural Class on Interpretation
See Figures 7 and 8 for the distribution of relationships by head natural class and
modifier natural class, respectively. Table 9 shows the distribution of relationships by both
head and modifier natural class.

Proportion of Relationship

Figure 7. Distribution of Relationships by Head Natural Class
1.0
.9
.8
.7
.6
.5
.4
.3
.2
.1
.0

Agentive
Constitutive
Telic
Formal

Animate

Inanimate Locative/
Temporal

43

Human

Artifact

Proportion of Relationship

Figure 8. Distribution of Relationships by Modifier Natural Class
1.0
.9
.8
.7
.6
.5
.4
.3
.2
.1
.0

Agentive
Constitutive
Telic
Formal

Animate

Inanimate Locative/
Temporal

44

Human

Artifact

Telic

Formal

Constitutive

Agentive

Table 9. Proportion (and Count) of Items by Relationship and Natural Classes

Animate Modifier
Inanimate Modifier
Location/Temporal Modifier
Human Modifier
Artifact Modifier
Agentive Overall
Animate Modifier
Inanimate Modifier
Location/Temporal Modifier
Human Modifier
Artifact Modifier
Constitutive Overall
Animate Modifier
Inanimate Modifier
Location/Temporal Modifier
Human Modifier
Artifact Modifier
Formal Overall
Animate Modifier
Inanimate Modifier
Location/Temporal Modifier
Human Modifier
Artifact Modifier
Telic Overall

Animate
Head
.00 (0)
.02 (1)
.00 (0)
.00 (0)
.00 (0)
<.01 (1)
.00 (0)
.10 (4)
.00 (0)
.00 (0)
.00 (0)
.02 (4)
1.00 (47)
.88 (37)
1.00 (42)
1.00 (33)
.90 (27)
.96 (186)
.00 (0)
.00 (0)
.00 (0)
.00 (0)
.10 (3)
.02 (3)

Inanimate Loc/Temp
Head
Head
.29 (9)
.00 (0)
.05 (2)
.06 (3)
.00 (0)
.00 (0)
.18 (4)
.00 (0)
.09 (2)
.00 (0)
.12 (17)
.01 (3)
.03 (1)
.03 (1)
.11 (4)
.22 (7)
.00 (0)
.00 (0)
.05 (1)
.05 (2)
.04 (1)
.25 (8)
.05 (7)
.11 (18)
.68 (21)
.76 (25)
.84 (31)
.59 (19)
.97 (31)
.93 (26)
.59 (13)
.85 (33)
.83 (19)
.72 (23)
.79 (115) .77 (126)
.00 (0)
.21 (7)
.00 (0)
.19 (6)
.03 (1)
.07 (2)
.18 (4)
.10 (4)
.04 (1)
.03 (1)
.04 (6)
.12 (20)

45

Human
Head
.00 (0)
.00 (0)
.00 (0)
.00 (0)
.00 (0)
.00 (0)
.03 (1)
.00 (0)
.00 (0)
.00 (0)
.03 (1)
.01 (2)
.75 (30)
.34 (10)
.78 (29)
.87 (34)
.78 (31)
.72 (134)
.23 (9)
.66 (19)
.22 (8)
.13 (5)
.20 (8)
.26 (49)

Artifact
Head
.00 (0)
.00 (0)
.00 (0)
.00 (0)
.00 (0)
.00 (0)
.15 (6)
.17 (8)
.05 (2)
.03 (1)
.00 (0)
.08 (17)
.58 (23)
.52 (25)
.85 (35)
.63 (25)
.51 (22)
.61 (130)
.28 (11)
.25 (12)
.10 (4)
.35 (14)
.49 (21)
.29 (62)

Overall
.05 (9)
.03 (6)
.00 (0)
.02 (4)
.01 (2)
.02 (21)
.05 (9)
.12 (23)
.01 (2)
.02 (4)
.06 (10)
.05 (48)
.76 (146)
.65 (122)
.91 (163)
.80 (138)
.73 (122)
.77 (691)
.14 (27)
.20 (37)
.08 (15)
.16 (27)
.20 (34)
.16 (140)

3.1.4.2.1 The Role of Semantic Natural Class on Agentive Interpretations
A 5 Head Natural Class (Animate, Inanimate, Locative/Temporal, Human, Artifact)
X 5 Modifier Natural Class (Animate, Inanimate, Locative/Temporal, Human, Artifact)
ANOVA was carried out on the proportion of Agentive responses. The main effect of Head
Natural Class was significant (F(4, 32) = 9.071, p < .001). Post-hoc comparisons, however,
revealed no differences between groups after the Bonferroni correction.
The main effect of Modifier Natural Class was also significant (F(4, 32) = 4.037, p =
.009). Post-hoc comparisons revealed that Animate modifiers were marginally more likely
than Artifact modifiers (p = .085) and Locative Temporal/Temporal modifiers (p = .071) to
receive Agentive interpretations.
The interaction between Head Natural Class and Modifier Natural Class was also
significant (F(16, 128) = 5.555, p < .001).
3.1.4.2.2 The Role of Semantic Natural Class on Constitutive Interpretations
A 5 Head Natural Class (Animate, Inanimate, Locative/Temporal, Human, Artifact)
X 5 Modifier Natural Class (Animate, Inanimate, Locative/Temporal, Human, Artifact)
ANOVA was carried out on the proportion of Constitutive responses. The interaction
between Head Natural Class and Modifier Natural class was significant (F(16, 128) = 1.807,
p = .037). The main effects of Head Natural Class (p = .606) and Modifier Natural Class (p =
.162) were not significant.
3.1.4.2.3 The Role of Semantic Natural Class on Formal Interpretations
A 5 Head Natural Class (Animate, Inanimate, Locative/Temporal, Human, Artifact)
X 5 Modifier Natural Class (Animate, Inanimate, Locative/Temporal, Human, Artifact)
ANOVA was carried out on the proportion of Formal responses. The main effect of Head
Natural Class was significant (F(4, 32) = 3.132, p = .028). Post-hoc comparisons revealed
46

that Animate heads were more likely than Artifact heads (p = .018), Human heads (p = .033)
and Inanimate heads (p = .044) to receive Formal interpretations
The main effect of Modifier Natural Class was also significant (F(4, 32) = 4.049, p =
.009). Post-hoc comparisons revealed that Locative/Temporal modifiers were significantly
more likely than Inanimate modifiers (p = .018) and marginally more likely than Animate
modifiers (p = .058) to receive Formal interpretations.
The interaction between Head Natural Class and Modifier Natural Class was also
significant (F(16, 128) = 2.335, p = .005).
3.1.4.2.4 The Role of Semantic Natural Class on Telic Interpretations
A 5 Head Natural Class (Animate, Inanimate, Locative/Temporal, Human, Artifact)
X 5 Modifier Natural Class (Animate, Inanimate, Locative/Temporal, Human, Artifact)
ANOVA was carried out on the proportion of Telic responses. The main effect of Head
Natural Class was significant (F(3, 24) = 12.315, p < .001). Post-hoc comparisons revealed
that Artifact heads were more likely than Animate heads p = .006) and Inanimate heads (p =
.017) to receive Telic interpretations. Human heads were more likely than Animate heads (p
= .002) and Inanimate heads (p = .003) to receive Telic interpretations. Finally,
Locative/Temporal heads were marginally more likely than Animate heads to receive Telic
interpretations (p = .072).
The interaction between Head Natural Class and Modifier Natural Class was
significant (F(16, 128) = 1. 734, p = .048).
The main effect of Modifier Natural Class was not significant (p = .195).
3.1.4.2.5 Summary of the Role of Semantic Natural Class on Interpretation
For almost all relationship types, the semantic natural class of the head and modifier
nouns was a significant predictor of how often that relationship would be used. Importantly,
47

the use of different relationships with different natural classes reflected the way in which we
naturally categorize nouns. For example, as head nouns, artifacts and humans, which can be
thought of as functional, were interpreted as Telic more often than inanimate natural kinds,
animate natural kinds, and locative/temporal nouns, which are thought of less, if at all, as
functional. As modifier nouns, locative/temporal nouns, which can serve to define the origin
of another noun, were most often interpreted as Formal. These results suggest that
determining the natural class of the constituents in a compound can aid in their
interpretation, especially in the absence of other relational cues, such as syntax.
3.1.5 Compound Interpretation Study Discussion
The results of the compound interpretation task supported all of our predictions for
the interpretation of compounds. For syntax, verbal compounds with both known and
nonce verbs were mainly interpreted as Telic, while noun-noun compounds with both
known and nonce heads were interpreted with all four possible relationships. This result
supports the claim that the modifier of a verbal compound must be interpreted as its internal
argument by virtue of merging as the complement of the verb in the derivation (Roeper and
Siegel 1978, Selkirk 1982, Lieber 1983, Di Sciullo and Williams 1987, Bobaljik 2003).
The results also supported the prediction that a noun must be specified for a
particular qualia role in order to be interpreted using that type of relationship, in the absence
of other contextual information. First, when an entire natural class was unlikely to have a
particular qualia role specified, that entire natural class did not receive an interpretation
based on that qualia role. For example, animate natural kinds are not thought of as
functional and compounds with animate natural kind heads only received 2% (N = 3) Telic
interpretations. Second, nonce head nouns were more likely than known head nouns to
receive Agentive or Constitutive interpretations, which are interpretations that require very
48

specific conditions to be met between the head and modifier noun. Familiarity with known
words makes these conditions very hard to meet, reflected in the 2% (N = 22) of known
noun-noun compounds that received Agentive interpretations and 5% (N = 50) of known
noun-noun compounds that received Constitutive interpretations. The rate of these
interpretations nearly doubled, however, when the head noun was nonce. Noun-noun
compounds with nonce head nouns received 5% (N =11) Agentive interpretations and 11%
(N = 20) Constitutive interpretations.
Finally, the natural class of the head and modifier nouns was shown to play a
significant role in the interpretation of the novel compounds (Downing 1977). Head noun
classes that are generally thought of as functional, such as artifacts and humans, were more
often interpreted as functional compared to head noun classes that are not thought of as
functional, such as animate natural kinds, inanimate natural kinds, or locative/temporal
nouns. For example, blueberry-boy and blueberry-bowl were both likely to be interpreted with
functional interpretations, a boy who sells blueberries and a bowl that holds blueberries.
Modifier noun classes also patterned according to our organization of the world. When the
modifier of a compound was a locative/temporal noun, it was almost always interpreted
with a Formal interpretation, where the location/time served as the origin of the head noun.
For example, a lake-pig was likely to be interpreted as a pig that lives near a lake. One
question for further research concerns the relative strength of the head and modifier natural
classes in determining the overall interpretation of the compound.
In sum, the results of this study support the proposed system for interpreting
compounds. Syntax strongly influenced the interpretation of the different strings in the
study; noun-noun strings were parsed as noun-noun compounds by the syntax and were
interpreted with many different relationships, while noun-verb-er strings were most often
49

parsed as verbal compounds with a verb-internal argument relationship. The lexical structure
of the constituents in the compounds constrained the relationships allowed by a compound
to only those specified in their lexical structure. Finally, the natural class of the head and
modifier nouns strongly influenced the compounds to be interpreted much in the same way
that we define those natural classes outside of compounds.
One question that this study was not able to answer was what types of
interpretations are disallowed by the grammar. By its nature, the current task relies on the
preferences of the participants’ grammars to determine an interpretation, but it only provides
evidence for particular interpretations are dispreferred, not disallowed. The next study
extends the findings of this study with verbal compounds to determine if non-verb-internal
argument interpretations are, in fact, disallowed by the grammars of English speakers.
3.2 Adult Study 2: Acceptable and Preferred Interpretations of Verbal Compounds
This study examined adults’ interpretations of verbal compounds using both an
acceptability task, which shows a grammar’s ability to produce a particular interpretation, and
a preference task, which shows a grammar’s preference for a particular interpretation. The
main goal of this study was to further explore the claim that verbal compounds can only
receive a verb-internal argument interpretation (Roeper and Siegel 1978, Selkirk 1982, Lieber
1983, Di Sciullo and Williams 1987, Bobaljik 2003). A secondary goal of this study was to
explore the claim that telicity interacts with argument realization (Borer 1994) in the
compound domain. I address the background and predictions for each of these goals in turn.
3.2.1 The Role of Syntax in the Interpretation of Verbal Compounds
The main goal of this study was to examine the interpretations that are allowed by
the syntax of verbal compounds. The structure of verbal compounds is argued to have a
verbal core much like that of a full verb phrase (Harley 2009, Gamache and Schmitt to
50

appear), where the modifier noun is merged as the complement of the verb and therefore
interpreted as its internal argument rather than its adjunct. In Adult Study 1 we saw that
adults prefer verb-internal argument interpretations of noun-verb-er strings, but we were
unable to test whether or not adult grammars disallow other types of readings altogether.
This study was designed to examine this claim by pairing nouns and verbs together
where the noun could plausibly serve as either the internal argument of the verb or an
adjunct to the verb. For example, clean and towel were paired since a towel is both something
that can be cleaned and can be used to clean. For both the acceptability task and the
preference task, two types of descriptions of compounds were given, one that reflects a
verb-internal argument relationship and one that reflects a verb-adjunct relationship. For
example, a towel-cleaner was described as either a person who cleans towels (internal
argument) or a person who cleans things using a towel (adjunct).
The hypotheses put forth in this dissertation predicted that speakers should both
prefer and allow only verb-internal argument interpretations of verbal compounds. For the
acceptability task, I predicted that argument descriptions of verbal compounds would be
rated as significantly more acceptable than adjunct descriptions of verbal compounds. As the
adjunct interpretation of a verbal compound is ruled out by its syntax, participants should
not allow this type of interpretation. Moreover, I predicted that argument descriptions of
verbal compounds trigger faster responses than adjunct descriptions of verbal compounds.
This delay in response time was predicted to be the product of a syntactic garden path, due
to a second possible analysis of the string noun-verb-er. If noun-verb-er were reanalyzed as a
noun-noun compound, the modifier noun would no longer the internal argument of the
verb and would not be subject to the argument structure of the verb. In other words, towel, in

51

towel-cleaner could be interpreted as the location or manner in which the cleaning happens
with noun-noun compound syntax.
For the preference task, the hypotheses put forth in this dissertation predicted that
the argument description will be overwhelmingly preferred to the adjunct description. Since
the noun-noun compound reanalysis was proposed as a repair strategy, and the argument
description was always present for the preference task, non-argument interpretations should
not have been considered.
3.2.2 The Role of Telicity in the Interpretation of Verbal Compounds
The secondary goal of this study was to examine the interaction of telicity with the
realization of arguments. Borer (1994) proposed that the difference in argument structure
between atelic and telic predicates involves the presence or absence of ASPQ, a low,
functional position that is occupied by the internal argument of the verb. In telic events,
ASPQ is present and the internal argument must be pronounced. In atelic events, ASPQ is
absent and the internal argument may be omitted. If atelic predicates can omit their internal
argument, it follows that verbal compounds representing atelic events can omit their internal
argument, which is normally realized as the modifier noun. Without the requirement for the
modifier noun to be interpreted as the internal argument of the verb, it is possible that the
modifier noun could be interpreted as an adjunct to the verb, an interpretation that is
possible with noun-noun compound syntax.
This study was designed to examine this claim by systematically choosing verbs that
fall into three categories: atelic, telic, and particle. Particle verbs are generally interpreted as
strongly telic, creating a range of telicity for the study.
The hypotheses put forth in this dissertation predicted that the extent to which
speakers allow and prefer adjunct readings of verbal compounds would be mediated by their
52

telicity. For the acceptability task, I predicted that adjunct interpretations of verbal
compounds would be more acceptable for atelic items as compared to telic and particle
items. Given the hypothesized ability for atelic predicates to omit their internal arguments,
atelic items should have been more acceptable than telic items and particle items with
descriptions where the modifier noun is not the internal argument. Moreover, atelic items as
a whole should have been responded to more quickly since they are hypothesized to allow
either an internal argument interpretation or an adjunct interpretation.
For the preference task, I predicted that the argument interpretation would be
preferred most often for the telic items and particle items, and to a lesser degree for the
atelic items. Because atelic items can allow both interpretations, however, they should have
been responded to more slowly in the preference task as compared to the telic items and
particle items, given the competition between the interpretations for atelic verbal
compounds that does not exist for telic verbal compounds and particle verbal compounds.
3.2.3 Method
3.2.3.1 Participants
32 adults (20 F, 12 M) were recruited from an undergraduate linguistics course at
Michigan State University. The subjects received extra credit in their course for their
participation.
3.2.3.2 Materials and Procedure
Each participant participated in an acceptability task and a preference task. Target
items were created for both tasks by pairing verbs with nouns that could reasonably be
interpreted as either an internal argument or an adjunct. For example, towel in towel-cleaner was
either what was cleaned (argument) or what was used to clean (adjunct). Eight verbs were
chosen for each of the three verb types that were tested, atelic, telic, and particle verbs. The
53

eight verbs were split across the two tasks with 12 target items (four atelic verbs, four telic
verbs, four particle verbs) in each task. See Table 10 for the full list of target items. The
verbs were balanced such that all verbs showed up in both the acceptability and the
preference task for different participants.
Participants were tested individually on PCs using E-Prime experimental software
and Sennheiser headphones. The acceptability task was always carried out first as the
preference task calls direct attention to the contrast under examination, the use of argument
vs. adjunct interpretations for verbal compounds. All participants participated in both tasks
in one session with a two-minute break between tasks.

54

Table 10. Full Item List for Adult Study 2
Verb
Type

Atelic

Telic

Particle

Verb

Compound

Argument Description

Adjunct Description

burn
carry
fight
protect
push
pull
tickle
wipe
build
buy
clean
cut
deliver
fix
paint
wash
check-out
drop-off
fill-up
hand-out
pass-out
pick-up
set-up
take-away

candle-burner
helicopter-carrier
robot-fighter
sword-protector
head-pusher
truck-puller
toe-tickler
tongue-wiper
tool-builder
phone-buyer
towel-cleaner
diamond-cutter
bike-deliverer
microscope-fixer
hand-painter
sponge-washer
computer-checker-outter
wagon-dropper-offer
bucket-filler-upper
basket-hander-outter
spoon-passer-outter
stick-picker-upper
ladder-setter-upper
car-taker-awayer

burns candles
carries helicopters
fights robots
protects swords
pushes heads
pulls trucks
tickles toes
wipes tongues
builds tools
buys phones
cleans towels
cuts diamonds
delivers bikes
fixes microscopes
paints hands
washes sponges
checks out computers
drops off wagons
fills up buckets
hands out baskets
passes out spoons
picks up sticks
sets up ladders
takes away cars

burns things using a candle
carries things using a helicopter
fights things using a robot
protects things using a sword
pushes things using their head
pulls things using a truck
tickles things using their toes
wipes things using a tongue
builds things using a tool
buys things using a phone
cleans things using a towel
cuts things using a diamond
delivers things using a bike
fixes things using a microscope
paints things using their hands
washes things using a sponge
checks out things using a computer
drops off things using a wagon
fills up things using a bucket
hands out things using a basket
passes out things using a spoon
picks up things using a stick
sets up things using a ladder
takes away things using a car

55

3.2.3.2.1 Acceptability Task
On each trial an audio recording first presented a person (“This is a person who
cleans towels.”) and then gave a potential name for that person (“a towel-cleaner”). The
description type was either an Argument description where the modifier of the compound
was treated as an internal Argument of the verb (“cleans towels”), or an Adjunct description
where the modifier of the compound was treated as an Adjunct to the verb (“cleans things
using a towel”). There were a total of 80 trials, 12 target trials and 68 filler trials, presented in
a uniquely randomized order for each participant. Description types were counterbalanced
across subjects such that for any compound, half of the subjects heard an Argument
description and half of the subjects heard an Adjunct description. Participants responded in
one of two methods. The Binary Response Group (N = 16) responded with either a “1” if
the compound and description matched perfectly, or “0” if the compound meant something
slightly or completely different than the description. The Likert Scale Group (N = 16)
responded on a Likert scale from 1 to 7, “1” being “That name means something slightly or
completely different.” and “7” being “I would definitely call them by this name.”
3.2.3.2.2 Preference Task
On each trial an audio recording first named a person using a verbal compound (“a
towel-cleaner”) followed by a 500 ms pause and a screen that showed two possible
descriptions (“This is a person who cleans towels.” or “This is a person who cleans things
using a towel.”). There were a total of 48 trials, 12 target trials and 36 filler trials, presented
in a uniquely randomized order for each participant. Participants responded by choosing
which description better first their interpretation of the compound.

56

3.2.3.3 Analysis
Responses and responses times were analyzed for both the acceptability and
preference tasks. A log transform was conducted on all response times to remove the
positive skew associated with the distribution of response times.
For the acceptability task four separate 3 Verb Type (Atelic, Telic, Particle) X 2
Description Type (Argument, Adjunct) ANOVAs were carried out. For the binary response
group, one ANOVA was carried out on the dependent measure of proportion of acceptable
(“1”) responses and one ANOVA was carried out on the dependent measure of logtransformed response times, originally measured in milliseconds. For the Likert response
group, one ANOVA was carried out on the dependent measure of average acceptability and
one ANOVA was carried out on the dependent measure of log-transformed response times,
originally measured in milliseconds.
For the preference task two separate 3 Verb Type (Atelic, Telic, Particle) ANOVAs
were carried out, one on the dependent measure of proportion of Argument responses and
one of the dependent measure of log-transformed response times, originally measured in
milliseconds.
3.2.4 Acceptability Task Results
3.2.4.1 Acceptability Task Response Results
Figure 9 shows the proportion of “1” responses by the Binary Response Group by
Verb Type and Description Type. A 3 Verb Type X 2 Description Type ANOVA on the
proportion of “1” responses revealed a main effect of Description Type (F(1, 30) = 77.039, p
< .001); Argument descriptions were rated as significantly more acceptable than Adjunct
Descriptions. Paired-sample t-tests revealed that this pattern reliably held for each Verb Type
(Atelic t(15) = 3.955, p =. 001; Telic t(15) = 6.455, p < .001; Particle t(15) = 4.392, p = .001).
57

The main effect of Verb Type (p = .683) was not significant. The interaction between
Description Type and Verb Type (p = .201) was also not significant.
Figure 9. VC Acceptability: Proportion of "1" Responses by Binary Group

Proportion of "1" Responses

1.0

Argument

.5

Adjunct

.0
Atelic

Telic

Particle

Figure 10 shows the average rating by the Likert Response Group by Verb Type and
Description Type. The 3 Verb Type X 2 Description Type ANOVA on the proportion of
“1” responses revealed a main effect of Description Type (F(1, 30) = 36.348, p < .001);
Argument descriptions were rated as more acceptable than Adjunct Descriptions. Pairedsample t-tests revealed that this pattern reliably held for each Verb Type (Atelic t(15) =
3.746, p =. 002; Telic t(15) = 4.772, p < .001; Particle t(15) = 4.109, p = .001). The main
effect of Verb Type (p = .634) was not significant. The interaction between Description
Type and Verb Type (p = .871) was also not significant.

58

Figure 10. VC Acceptability: Average Acceptability Rating by Likert Group
7

Acceptability Rating

6
5
Argument

4

Adjunct

3
2

1
Atelic

Telic

Particle

3.2.4.2 Acceptability Task Response Time Results
Figure 11 shows the log-transformed response times for the Binary Response Group
by Verb Type and Description Type. The 3 Verb Type X 2 Description Type ANOVA
performed on the log-transformed response times for the binary group revealed a marginally
significant main effect of Description Type (F(1, 30) = 4.085, p = .061); Argument
descriptions were responded to slightly faster than Adjunct descriptions. The main effect of
Verb Type was not significant (p = .148). The interaction between Verb Type and
Description Type was also not significant (p = .361).

59

Figure 11. VC Acceptability: log Response Times by Verb Type for Binary Group
3.4
3.3

log RT

3.2
3.1

Argument

3

Adjunct

2.9
2.8

2.7
Atelic

Telic

Particle

Figure 12 shows the log-transformed response times for the Likert Response Group
by Verb Type and Description Type. The 3 Verb Type X 2 Description Type ANOVA
performed on the log-transformed response times for the binary group revealed a main
effect of Description Type (F(1, 30) = 12.009, p = .003); Argument descriptions were
responded to significantly faster than Adjunct descriptions. Paired-sample t-tests revealed
that Argument description response times were significantly shorter than Adjunct
description response times for Atelic items (t(15) = -3.694, p = .002), and Particle items
(t(15) = -2.185, p = .045), and marginally shorter for Telic items (t(15) = -1.780, p = .095).
The main effect of Verb Type (p = .334) was not significant. The interaction between Verb
Type and Description Type was also not significant (p = .588).

60

Figure 12. VC Acceptability: log Response Times by Verb Type for Likert Group
3.5
3.4

log RT

3.3
3.2

Argument

3.1

Adjunct

3
2.9

2.8
Atelic

Telic

Particle

3.2.5 Preference Task Results
3.2.5.1 Preference Task Response Results
Figure 13 shows the proportion of Argument responses by Verb Type and
Description Type. The 3 Verb Type ANOVA on the proportion of Argument responses
revealed a main effect of Verb Type (F(2, 62) = 15.500, p < .001). Post-hoc comparisons
revealed that Argument descriptions were chosen significantly more often for Particle items
than Atelic items (p < .001) and significantly more often for Particle items than Telic items (p
< .001). There was no difference in how often Argument descriptions were chosen for Telic
and Atelic items (p = .441).

61

Figure 13. VC Preference: Proportion of Argument Responses by Verb Type

Proportion of Argument
Responses

1.0

.5

.0
Atelic

Telic

Particle

3.2.5.2 Preference Task Response Time Results
Figure 14 shows the log-transformed response times by Verb Type. The 3 Verb
Type ANOVA on the log-transformed response times revealed a main effect of Verb Type
(F(2, 62) = 6.924, p = .002). Post-hoc comparisons revealed that responses were significantly
faster for Particle items than Atelic items (p < .001) and marginally faster for Telic items than
Atelic items (p = .057). There was no difference in how quickly responses were made for
Telic items and Particle items (p = 1.000).

62

Figure 14. VC Preference: log Response Times by Verb Type

log RTs

3.7

3.6

3.5

3.4
Atelic

Telic

Particle

3.2.6 Acceptability and Preference Task Discussion
The results of the acceptability and preference tasks supported both predictions of
this study. First, both tasks overwhelmingly supported the prediction that the structure for
verbal compounds does not allow adjunct interpretations. This was shown in two ways.
First, both response groups for the acceptability task failed to rate the adjunct readings as
acceptable as compared to the argument readings. For the preference task, participants
overwhelmingly preferred the internal argument reading for verbal compounds. Second,
when presented with an adjunct interpretation of a verbal compound, participants’ responses
times increased significantly, suggesting that they initially analyzed the compounds as verbal
compounds, which do not allow adjunct interpretations, and were forced to reanalyze the
compounds as noun-noun compounds, which do allow adjunct interpretations.
The results also supported the second prediction, that atelic predicates, even in
compounds, can omit their internal arguments (Borer 1994). In the preference task, both the
response patterns and the response time patterns support a theory where atelic predicates in
verbal compounds can omit their internal arguments. For responses, adjunct interpretations
were chosen most often for atelic verbal compounds. For response times, atelic verbal
63

compounds had the longest response times, reflective of the availability of either an
argument interpretation or an adjunct interpretation, a choice that was not present for telic
verbal compound and particle verbal compounds.
3.3 General Discussion
The goal of this chapter was to test the predictions made by the interpretation
system outlined in Chapter 2. The system proposed in Chapter 2 suggests that adults use two
types of information when interpreting compounds: syntax and lexical structure, which
critically includes natural class information. The two studies presented in this chapter
suggested that both of these information sources are, in fact, used in conjunction with one
another to arrive on a compound’s interpretation. Syntax was shown to systematically
determine a relationship for verbal compounds, but not for noun-noun compounds.
Importantly, the syntax of verbal compounds only required this relationship when the verb
in the verbal compound obligatorily required an internal argument; atelic verbs, which are
able to omit their internal arguments, were able to be parsed as noun-noun compounds with
adjunct-type readings. For noun-noun compounds, the relationship was determined by the
information in the head noun’s lexical structure. The set of possible relationships was
determined by the qualia roles that were specified for on the head noun, and the most likely
relationship was determined by the natural classes of the head and modifier in conjunction
with one another. In the second half of this dissertation this system will be treated as the
target state for children who are learning the English system of compounds.

64

CHAPTER 4: ACQUISITION BACKGROUND
4.0 Introduction
The second half of this dissertation examines how children acquire noun-noun
compounds and verbal compounds. In Chapters 2 and 3 we found evidence that adults use
two aspects of compounds to determine their meaning: their syntax and the lexical structure
of their constituents. In the second half of this dissertation I take the position that children
use the same compositional system as adults to interpret compounds and any deviations
from adult-like interpretations arise from children having different syntactic representations
or different lexical knowledge during development. In this chapter I review the work that
has been done on children’s acquisition of compounds and summarize some questions that
remain about this acquisition process.
4.1 Different Approaches to Language Acquisition
Any acquisition research has to consider the existence of an innate component and
the influence from the language the learner is exposed to. Current approaches to language
acquisition diverge in both qualitative and quantitative terms. On one side we have an
approach that assumes an innate component which filters the input in specific ways and on
the other hand we have an approach that considers domain general learning mechanisms to
be sufficient for language acquisition. Under the latter, children do not come with prespecified categories and develop their linguistic system by analogy with items in the input.
We will call this the usage-based approach (Tomasello 2000). Under the former approach,
data from the input is filtered and choices are made based on the input. In other words, in
both cases, information from the input is crucial, but in the usage-based case we do not
expect deviations from the input but rather a conservative build-up of the system. In the
input-filtering mode children may, in the acquisition of the grammar, produce non-adult
65

forms absent in the input that are related to a particular analysis of the structure. In the next
section I review the research on children’s acquisition of compounds, bearing in mind these
different approaches to language acquisition and the types of data that support each.
4.2 What Children Know about Compounds
This section reviews the acquisition work that has been done on children’s
acquisition of compound syntax and children’s interpretation of compounds. The first
section reviews past literature on children’s acquisition of noun-noun compound syntax and
children’s interpretation of noun-noun compounds. The second section reviews past
literature on children’s acquisition of verbal compound syntax and children’s interpretation
of verbal compounds.
4.2.1 Children’s Acquisition of Noun-Noun Compounds
4.2.1.1 Acquisition of Noun-Noun Compound Syntax
The relatively simple syntax of noun-noun compounds and the early age at which
noun-noun compounds become productive makes their acquisition somewhat opaque.
Children produce adult-like novel noun-noun compounds by age 3 (Clark 1981, Snyder
1995), and for the most part, produce them without apparent syntactic errors, such as order
reversals like trap-mouse for mouse-trap (Clark et al. 1985, Clark & Berman 1987, Hiramatsu et
al. 2000).
In addition to learning the syntax of noun-noun compounds very early, Clark et al.
(1985) found that young children also pick up on the distribution with which adults use
noun-noun compounds versus other structures. In a production study Clark et al. prompted
children to name objects in two images that different by only one feature, for example, a
truck that carries a horse and a truck that carries a bike in another. They found that children,
like adults, produced compounds in cases of contrast (a horse-truck vs. a bike-truck for two
66

trucks whose beds contained those objects). Children and adults also produced compounds
more often when the modifier-head relationship defined an intrinsic property of the referent
rather than a temporary one. For instance, both children and adults were more likely to use
pumpkin-house for a house made of pumpkin (intrinsic) than for a house with a pumpkin near
it (temporary).
Previous literature has pointed to one aspect of children’s production of noun-noun
compounds that may give insight into the acquisition process. Hiramatsu et al. (2000)
showed that children produce order errors when attempting to produce compounds for
particular types of descriptions. In their production study, 3- to 4-year old English-speaking
children often reversed the order of the nouns in shape/appearance descriptions such as “a
chair shaped like a hand”, producing chair-hand rather than hand-chair. The children in this
study did not produce order errors for any of the other tested description types (material,
place of origin). Hiramatsu et al. suggest that this pattern may be due to a compound subset
principle, where children are using the compound system of a language such as Hungarian,
where shape/appearance compounds are not allowed. This pattern, however, failed to hold
up in their corpus study, where young children were shown to produce novel adult-like
shape/appearance compounds, leaving open the question of whether or not children have
non-adult restrictions on their use of compound syntax.
To summarize, by most measures, children learn the syntax of noun-noun
compounds very early and use noun-noun compounds in the same scenarios as adults. One
area for further research is to determine if children use compound syntax to represent the
same range of meanings as adults, or if they use compound syntax in a more restricted way.

67

4.2.1.2 Children’s Interpretations of Noun-Noun Compounds
Turning to children’s interpretation of noun-noun compounds, children understand
compound headedness by age 3 (Clark et al. 1985). For example, three-year-old children
know that a blueberry-spoon is a spoon and not a blueberry. Moreover, children understand
that compounds create a subset relationship between a basic term and its compound
subordinate members (Gelman, Wilcox, and Clark 1989). For instance, young children
understand that a car-fap is a member of the type fap.
Children begin to look less adult-like in studies that ask children to define the
meaning of an entire compound. In a production study, Clark and Berman (1987) showed
that younger children described compounds only using the head noun and only later in
development identified the modifier and the relationship between the two. In their study,
three-year-olds identified the relationship between the head and the modifier noun on only
13% of trials, while four-year-olds identified this relationship on 74% of trials. In a similar
study, Krott and Nicoladis (2005) showed that the likelihood of a child to include the
modifier in their description of a compound was related to the number of compounds in the
children’s input that shared the same modifier. For example, if children knew many
compounds with chocolate as the modifier, they were likely to say that a chocolate-cake is a cake
made of chocolate. If children were not familiar with many compounds with chocolate as the
modifier, they were more likely to say that a chocolate-cake was just a cake, without mention of
the modifier chocolate. Interestingly, in this study, this pattern did not change over time. The
five-year-olds in this study, who presumably had a significantly larger vocabulary than the
four-year-olds, were as likely as the four year olds to exclude either the head or modifier
nouns from their compound descriptions.

68

While the Clark and Berman (1987) and Krott and Nicoladis (2005) studies report on
children’s ability to give a description using both the head and modifier nouns, these studies
do not report what type of relationship children ascribed to the compound. In fact, relatively
little is known about how children interpret the relationship between the nouns in a nounnoun compound. Recently, Krott et al. (2009) found that the frequency of specific head
noun-relationship pairings in the input plays a role in children’s interpretations of nounnoun compounds. For example, in their task, a child was better at interpreting mouse-trap as a
trap that catches mice if the child was familiar with ant-trap, bear-trap, etc., all of which have
trap as their head noun and a ‘for’ relationship. The authors suggest that this ease of
interpretation is due to item-specific templates that children develop based on their input.
Two things are of note for this Krott et al. study. First, the children in this study
were on average five years old and were still argued to be using templates for compounds.
Second, a brief inspection of the items used in the Krott et al. (2009) study reveals that,
while it is true that each noun is used most often with a particular relationship, it is also true
that different natural classes are each used most often with a particular relationship. For
example, the artifacts basket, table, shoes, rack, suit, foil, helmet, etc. were all used most frequently
with ‘for’ relationships as head noun. Therefore, while it may appear that children use itemspecific templates in their interpretation, it is also plausible that children use natural class
level information to interpret novel compounds. Independently of compounds, we have
good reason to believe that children would employ natural class information in their
interpretation of novel compounds. It has been shown in numerous studies that children use
natural class membership to induce properties of novel nouns that are not necessarily
apparently in the learning environment (e.g., Gelman 1988, Gelman and Markman 1987,
Davidson and Gelman 1990). For example, children are able to instantaneously associate
69

properties such as “needs lungs to breathe” from a known animal to a novel animal without
direct evidence that the novel animal has that property. The same mechanism, capitalizing
on commonalities among classes, is used by at least adults when interpreting compounds.
The final finding of note is that in comprehension children appear to prioritize
physically apparent information over intrinsic, but possibly not apparent, information. In an
interpretation task, Krott et al. (2010) used nonce word compounds to determine how
children interpreted compounds without prior experience with the constituents in the
compound. Their results suggest a bias towards ‘has’- or ‘located’-type relations in children’s
interpretation of compounds, that is not predicted by the relative frequencies in the input.
That is, in their study, children were more likely to pick the wug with a moop near it, as
opposed to a wug made of woop for wug-moop. It is of note, however, that using nonce
words does not allow for children to access any of the rich lexical knowledge that we would
argue is used in compound interpretation, such as physical appearance, function, etc.
Therefore, a safe bet for the child participants, who were unfamiliar with the words, would
be to go with the physically apparent relationship, which is easily seen in the images that
represent each examples.
To summarize, by some measures, children interpret noun-noun compounds in an
adult-like way; children understand headedness and that compounds create subsets of the
head noun. Based on the findings to date, however, it is not clear if children are adult-like or
not in establishing the relationship between the nouns in a noun-noun compound.
4.2.1.3 Acquisition of Noun-Noun Compounds Summary and Questions
Children begin producing novel noun-noun compounds by age three, and appear to
produce them under largely the same circumstances as adults. Less is known, however, about
how children interpret novel noun-noun compounds. The only work done to date on how
70

children determine relationship that exists between the head and modifier in a noun-noun
compound suggest that frequencies in the input may play a role in children’s interpretation.
The previous literature on the acquisition of noun-noun compounds raises three questions
for further research:
(6) Questions raised by previous compound acquisition literature
a. Do young children systematically avoid using noun-noun compounds for
particular types of descriptions?
b. Are children using templates or a more general mechanism to interpret nounnoun compounds?
c. Why is there a disparity between children’s preference for intrinsic relationships
when producing novel compounds, but temporary relationships when
interpreting novel compounds?
4.2.2 Children’s Acquisition of Verbal Compounds
4.2.2.1 Acquisition of Verbal Compound Syntax
Unlike noun-noun compounds, the syntactic complexity of verbal compounds makes
their acquisition much more transparent. Children produce adult-like novel verbal
compounds around age five (Clark et al. 1986) and adult-like recursive verbal compounds
(mouse-catcher-maker) around age seven (Hiraga 2010). Clark et al. (1986) were the first to show
that children produce a specific set of non-adult forms during the acquisition of verbal
compounds; children reliably produce verb-internal argument catch-mouse forms and verb-erinternal argument catcher-mouse forms, but fail to produce logically possible internal argumentverb mouse-catch forms. This pattern has been replicated in production studies with English
children (Nagpal and Nicoladis 2009, Gamache and Schmitt to appear) and Dutch children
(Brisard et al. 2008). Moreover, Clark and Barron (1988) found that children’s acceptability
71

of verbal compound forms largely matched the current state of their own grammar. While all
children accepted grammatical verbal compounds (internal argument-verb-er), younger
children often also accepted non-adult forms and “repaired” non-adult forms to other nonadult forms. For example, a child whose current produces catcher-mouse forms might say that
catch-mouse is bad, but may only repair catch-mouse to catcher-mouse, instead of the adult mousecatcher.
The locus of the pattern of forms that children produce (catch-mouse, catcher-mouse, and
mouse-catcher but not catch-mouse) has been attributed to (i) interference from other frequent
nominal constructions in the input (Nagpal and Nicoladis 2009), (ii) interference from verb
phrase verb-internal argument order (Clark et al. 1986), and (iii) adherence to the steps in the
adult derivation (Gamache and Schmitt to appear).
To address the first claim, that children’s non-adult forms are due to influence from
verb-internal argument order in phrasal syntax, we turn to Dutch data. In Dutch, the target
adult form is the same and children have been shown to produce the same set of non-adult
forms (Brisard et al. 2008). However, Dutch children’s input is 45% internal argument-verb
order (Evers and van Kampen 2001) and they produce no internal argument-verb verbal
compound forms, suggesting that it is unlikely simple frequency of phrasal word order drives
the pattern of non-adult forms.
The second claim is that catch-mouse and catcher-mouse forms surface due to
interference from other nominal constructions that have verb-noun order, such as running
man phrasal constructions, or dump-truck or frying-pan compound constructions (Nagpal and
Nicoladis 2009). There are two reasons why this argument is likely to be incorrect. First, one
of the tenets of construction grammar is that children are mapping word strings to
interpretations. However, the relationship between catch and mouse in mouse-catcher is quite
72

different than the relationship between the noun and verb in any of the other constructions.
In mouse-catcher the noun is the object of the verb, while in running-man the noun is the agent
of the verb, and in dump-truck and frying-pan the noun is the instrument by which the verb’s
action is done. If children do not assign thematic relations to these constructions in the same
way as adults, then children should be more likely to allow a range of meanings for verbal
compounds than adults, which it is not clear that they do. For example, do children allow
catch-mouse to be a mouse used for catching, as it would if it were assigned the same
interpretation as a dump-truck-type compound?
The second reason why this argument is unlikely to be correct is because it predicts a
wide range of errors that children are not attested to produce. Table 11 shows the type and
token count of four different compound constructions in the input to children in the Brown
(1973) corpus on CHILDES (MacWhinney 2000). The first thing to note is that, relative to
noun-noun compounds, all of the compounds containing verbs are infrequent. The second
thing to note is that the compounds contain verbs are roughly equally frequent, relative to
one another. Given their equal frequency, they should all affect each other in the acquisition
process.
Table 11. Type (and token) count of compounds in adult speech in Brown (1973) corpus
Compound
NVer Compound
VN Compound
VingN Compound
NN Compound

Example
mouse-catcher
dump-truck
reading-lamp
book-trap

Input to Adam
15 (103)
18 (58)
6(51)
200 (~1000)

Input to Sarah
8 (34)
18 (77)
8 (24)
175 (~700)

Table 12 shows the forms that should be attested in children’s speech if all of these
compound types are able to affect each other. In these forms, the word order of each
compound type is crossed with the morphology of each compound type. Looking first at the
predicted forms for the mouse-catcher target, we know from the production studies discussed
73

above that the greyed forms are attested in children’s speech. However, two logical
possibilities mouse-catching and mouse-catch are not attested in any of the production studies to
date, critically to mean mouse-catcher. Unfortunately there are no production studies of VN
and VingN compounds, but the non-adult form types for dump-truck and reading-lamp to our
knowledge have not been discussed in the literature, despite being quite salient (for example,
lamp-reader in the place of reading-lamp). In other words, there appears to be a unidirectionality
of the VN and VingN compounds affecting NVer compounds, but NVer compounds not
affecting VN and VingN compounds. This unidirectionality is not predicted by the usagebased account.
Table 12. Possible non-adult forms for NVer, VN, and VingN
Morphology
Target word order
and interpretation
mouse-catcher
dump-truck
reading-lamp

Word
Order
VN
NV
VN
NV
VN
NV

-er
catcher-mouse
mouse-catcher
dumper-truck
truck-dumper
reader-lamp
lamp-reader

-ing
catching-mouse
mouse-catching
dumping-truck
truck-dumping
reading-lamp
lamp-reading

-Ø
catch-mouse
mouse-catch
dump-truck
truck-dump
read-lamp
lamp-read

A generative approach, such as the one we have proposed in Gamache and Schmitt
(to appear), straightforwardly explains why children produce this particular set of forms and
not others. In line with a strong version of the continuity hypothesis (Pinker 1984),
children’s non-adult forms are steps in the adult derivation, and therefore only forms that are
reflective of steps in the derivation are those produced by children. Shown in (7) – (10),
children learn the first step of the derivation, merging the verb and its internal argument,
producing catch-mouse. In the next step, children merge –er in the structure incorporate the
verb into –er, producing catcher-mouse. Finally, children move the internal argument to the
specifier of Ver, producing mouse-catcher.
74

(7) Step 1 catch-mouse
v
v
catch

n
mouse

(8) Step 2 catcher-mouse
n
n
v
catch

v
n
er

v
catch

n
mouse

(9) Step 3 mouse-catcher
n
n
mouse

n
n

v
catch

v
n
er

v
catch

n
mouse

4.2.2.2 Children’s Interpretation of Verbal Compounds
While we know of no systematic study of how children interpret the modifier in
verbal compounds, there is some evidence that children’s interpretation of certain aspects of
verbal compounds take a while to develop. Gelman and Heyman (1999) found that older
children and adults were more likely than younger children to interpret a property as being
75

stable when referred to with a verbal compound. For example, when called a person was
called a carrot-eater older children and adults were more likely than younger children to infer
that eating carrots was a regularly occurring activity of the named person.
4.2.2.3 Children’s Interpretation of Compounds Summary and Questions
Children begin produce adult-like verbal compounds around age five, and prior to
target forms, produce a particular set of non-adult forms (catch-mouse, catcher-mouse, but not
mouse-catch). There is no work to date on how children interpret the relationship in verbal
compounds. The previous literature on the acquisition of noun-noun compounds raises two
questions for further research:
(10) Questions raised by previous child compound interpretation literature
a. Why do children produce the set of forms that they do?
b. What type of relationship do children attribute to verbal compounds and does
the relationship change relative to their syntactic knowledge?
4.2.3 Summary of Past Acquisition Research
There are a number of findings that bear on our goal of determining if children use
the same system as adults to interpret compounds. For the syntax of noun-noun
compounds, children appear to avoid associating the noun-noun compound structure with
particular descriptions. This avoidance could be due to a restriction that children have on the
compound structure, which might support a generative approach to language acquisition, or
it could be due to the compound structure being infrequently associated to this description
in the input, which might support a usage-based approach. For the interpretation of nounnoun compounds, children appear to interpret compounds in line with related compounds
in their input, which supports a usage-based theory of language acquisition. For the syntax of
verbal compounds, children only produce a particular set of forms (catch-mouse and catcher76

mouse, but not mouse-catch for mouse-catcher). This pattern has been explained as interference
from other constructions in the input, in line with a usage-based account, and as children
building the steps in the adult derivation, in line with a generative account. For the
interpretation of verbal compounds, the question of how children learn to interpret the
modifier in verbal compounds as the internal argument of the verb is still completely open.
4.3 Hypotheses and Predictions for Children’s Acquisition of Compounds
In this section I present the hypotheses and predictions for how children acquire
compounds. The hypotheses, shown in Table 13, are discussed in turn, along with any
predictions that the hypothesis makes about children’s behavior.
Table 13. Hypotheses and Predictions for Children's Acquisition of Compounds
Hypothesis
Hypothesis 1: The strong
continuity hypothesis is
correct for the acquisition of
compounds.
Hypothesis 2: Given
principles of compositionality,
syntax and interpretation go
hand in hand provided that
children know the meaning of
the words in compounds.

Predictions
Prediction 1: The non-adult compound forms that
children produce are steps in the adult derivation of that
compound form.
Prediction 2: Even when producing non-adult forms,
children interpret compounds in line with the constraints of
their knowledge of compound syntax.
Prediction 3: Non-adult compound structures lead to nonadult interpretations of compounds.
Prediction 4: Incomplete lexical semantic knowledge leads
to non-adult interpretations of compounds.
Prediction 5: Children’s use of noun-noun compounds
reflects natural class-, not item-, level knowledge.

The first hypothesis is that the strong continuity hypothesis (Pinker 1984) is correct
for the acquisition of compounds. This hypothesis predicts that the only non-adult forms
that children produce are forms that are steps in the adult derivation. For example, if we
assume that the adult derivation begins with the merging of the verb with a noun as its
internal argument, then one form that children are predicted to produce is verb-internal
argument catch-mouse. Moreover, this hypothesis predicts that children’s interpretations of
compounds are in line with the constraints of the current state of their grammars. For
77

example, if a child is producing verb-internal argument catch-mouse forms, then their grammar
has already learned that the first step in the adult derivation of verbal compounds is to merge
the verb with the internal argument. Therefore, already at this point in the acquisition
process, the child should only allow interpretations where the modifier noun in a verbal
compound is interpreted as the theme of the verb.
The second hypothesis is that, given principles of compositionality, syntax and
interpretation go hand in hand provided that children know the meaning of the words in
compounds. This hypothesis suggests that the mechanism that children are using is the same
that adults are using and predicts that any deficiency in the parts of that mechanism, whether
syntactic or lexical semantic knowledge, lead to non-adult interpretations. This hypothesis
leads to three predictions.
The first prediction is that a child who has not yet acquired the syntax of a structure
should allow a different set of interpretations than a child who has acquired the syntax. For
example, a child who has not yet acquired any part of the verbal compound structure has
also not acquired any of the constraints that are part of that structure. Therefore, this child is
predicted to allow interpretations of verbal compounds that would not be allowed by adult
speakers, such as an instrument reading of towel in towel-cleaner.
The second prediction is that a child who does not know the lexical semantic
features of a word in a compound should interpret the compound differently than a child
who does know those features of the word. For example, if a child knows that an object is
an artifact, but does not know what the function of the object is, the child will not be able to
have a functional interpretation of the object.
The third prediction is that children’s use of noun-noun compounds should reflect
natural class-, not item-, level knowledge. Therefore, children are predicted to be able to use
78

words in compounds that they have previously not heard in compounds. For example, even
if a child has not heard the words hamburger or spatula used in a compound before, they can
generate or interpret a novel compound such as hamburger-spatula based on their knowledge
of how hamburger and spatula should behave. Children can also use words that they have
heard before in compounds, but they can use them in different ways, given the tendencies of
the natural classes of the head and the modifier. For example, even if a child has only heard
spatula used in a functional way, which is likely given its status as an artifact, children would
still be able to produce and interpret beach-spatula, as a spatula that is used when at the beach,
not a spatula that is used to flip the beach.
4.4 Summary
In this chapter I reviewed previous acquisition work pertaining to children’s
acquisition of compound syntax and children’s interpretation of compounds. I then
discussed which questions remain for the study of children’s acquisition of compounds.
Finally, I outlined a series of hypotheses and predictions for children’s acquisition of
compounds. In the next two chapters I present the results of five studies that test the
predictions made by this system for children’s acquisition of noun-noun compounds and
verbal compounds.

79

CHAPTER 5: CHILD NOUN-NOUN COMPOUND TASKS
5.0 Introduction
In the first half of this dissertation I outlined a system for how adults interpret
compounds. In this system, adults reconcile syntactic and lexical semantic information to
arrive at common interpretations of compounds. This chapter presents the results of three
studies that examine how children acquire the components of the adult system necessary to
interpret noun-noun compounds. The aim of these studies was to provide evidence that
children use the same system as adults when interpreting noun-noun compounds, by
addressing the questions in (11).
(11) Questions addressed in Chapter 5 Studies
a. Do children use the tendencies of natural classes when producing compounds?
b. Do children use the full range of relationships available from lexical structure
when producing compounds?
c. Do children use the tendencies of natural classes when interpreting compounds?
d. Do children use the full range of relationships available from lexical structure
when interpreting compounds?
The first study addressed the question in (11a) by examining the behavior of
individual nouns and natural classes of nouns in children’s production of novel noun-noun
compounds in light of their input. The goals of this study were (i) to determine if children
produce noun-noun compounds in line with the tendencies of natural classes and (ii) to
determine at what level (individual nouns or natural classes of nouns) children’s novel nounnoun compound productions mirror their input.
The second study addressed the question in (11b) by examining children’s
production of noun-noun compounds dependent on which Qualia Role (Pustejvosky 1995)
80

is being accessed by the compound. The goal of this study was to determine if children
produce noun-noun compounds with the full range of meanings that adults do, or if they
focus initially on only a subset of meanings.
The third study addressed the questions in (11c) and (11d) by examining children’s
interpretation of noun-noun compounds. The goal of this study was to determine if children
are flexible in their assignment of interpretation of noun-noun compounds based on the
lexical semantic features of both constituents in the compound, or if one constituent
determines the type of relationship that is assigned to the entire compound.
Taken together, the ultimate goal of this chapter is to provide evidence that children
use the same mechanism as adults to produce and interpret noun-noun compounds, and any
deviations in previous literature can be explained by the availability of lexical information.
5.1 Child Study 1: Noun-Noun Compounds in Children’s Speech and their Input
The first study examined the behavior of individual nouns and natural classes of
nouns in children’s production of novel noun-noun compounds in light of the behavior of
individual nouns and natural classes in their input. Krott et al. (2009) found in an
interpretation task that children interpret novel compounds in analogy to known compounds
with the same head noun. For instance, if a blueberry-spoon was known to be a spoon for
eating blueberries, then a cake-spoon was interpreted as a spoon for eating cake. They suggest
that this ease of interpretation is due to word-specific templates that supply the head noun
and the relationship that exists between the head and the modifier noun.
The goal of our first study was to provide evidence that, in fact, children use the
natural class information that has been shown to be used by adults (Downing 1977, Chapter
3 of this dissertation), rather than item-specific information, as might be predicted by Krott
et al (2009).
81

To examine the impact of the input on children’s production of novel compounds, a
corpus study was conducted. We analyzed all noun-noun compounds in the children’s
speech and all noun-noun compounds in their input. The compounds were analyzed
according to (i) the relationship in the compound and (ii) the head and modifier noun in the
compound. First, the compounds were analyzed according to the individual nouns. For
example, mouse-trap was analyzed as a compound with a mouse modifier and a trap head. The
purpose of this analysis was to compare the behavior of particular nouns in the input to the
behavior of these same nouns in children’s production of novel noun-noun compounds.
Second, compounds were analyzed according to the natural classes of the head and modifier
nouns. For example, mouse-trap was analyzed as a compound with an animate natural kind
head and an artifact head. The purpose of this analysis was to compare the behavior of
natural classes in the input to the behavior of natural classes in children’s production of
novel noun-noun compounds.
The hypotheses put forth in this dissertation made two predictions about the
relationship between noun-noun compounds in children’s input and the novel noun-noun
compounds that children produce. The first prediction was that, at the natural class level,
children would match their input in terms of the relationships that were used with different
head natural classes and modifier natural classes. For example, compounds with artifact head
nouns in the input and in children’s novel production were predicted to reflect a high level
of Telic relationships. The second prediction was that, at the individual noun level, children
should not necessarily have matched their input in terms of the relationships that were used
with different head nouns and modifier nouns. For example, compounds with the head
noun trap may have had a different distribution of relationships in the input compared to
children’s novel productions.
82

5.1.1 Method
All compounds were extracted from the transcripts of parent/child interactions for
two children, Adam (2;3-4;8, 55 sessions) and Sarah (2;3-4;6, 130 sessions), in the Brown
(1973) corpus of the CHILDES database (MacWhinney 2000). For this study a compound
was operationally defined as two consecutive nouns that denote a single entity. A total of
703 different compounds were extracted from the transcripts. See Appendix B for this list,
broken down by the speaker who each compound is attributed to (Adam, Adam’s input,
Sarah, Sarah’s input). Note that all references to counts of compounds refer to compound
types rather than tokens.
5.1.1.1 Coding
Compounds and the nouns that they contained were coded in four ways: (i) speaker,
(ii) position of noun, (iii) natural class of noun, and (iv) relationship.
For speaker, all compounds that were uttered by an adult, even if after a child had
said them in the corpus, were considered as part of the input, and therefore tagged as Adult.
All compounds that were only ever uttered by the child were tagged as Child. Speaker was
tagged in this way to arrive on the strongest set of compounds that are children’s novel
productions.
For position, the first noun in the compound was tagged as Modifier and the second
compound was tagged as Head.
For natural class, each noun was analyzed according to the natural class that it falls
into. The coding of nouns was based on their prototypical properties, rather than meanings
that have evolved. For example, star was coded as being an inanimate natural kind, referring
to the object in space, rather than a concept, referring to the five-pointed shape. To ensure
an unbiased analysis of the nouns in each compound, all nouns were extracted from their
83

context and analyzed in isolation, with the exception of homophones or otherwise unclear
examples (for example, a child’s made up word). The coding guidelines used are summarized
in Table 14.
Table 14. NNC and VC Interpretation: Noun Categories
Category
Animate Natural Kind
Artifact
Concept
Human
Inanimate Natural Kind
Locative
Temporal
Unknown

Defining Features
naturally occurring, sentient
functional, manmade objects
non-physical, abstract, rule-governed
human, profession
naturally occurring, non-sentient
defined by its physical space
defined by its temporal space
unsure of classification
homophone representing different classes

Example
cow, dolphin
spoon, box
happiness, safety
boy, lawyer
rock, wind
school, lake
summer,
moo
bank

Finally, for relationship, the compounds were coded according to the type of relation
that held between the two nouns in the compound. Guidelines were created that were
loosely based on Johnston and Busa’s (1996) application of Pustejovsky’s (1995) Generative
Lexicon theory to compounds. According to Johnston and Busa’s analysis, compounds can
be interpreted according to the internal lexical properties of the two nouns, which are
organized according in their Qualia Structure. Since the meaning of a compound is
dependent on the context, the coders had access to the context the compounds appeared in
when making decisions. The coding guidelines used are summarized in Table 15. Note that
the Agentive role is not part of this coding system because of the ambiguity, even in context,
of items to be coded as Agentive or one of the other roles.

84

Table 15. Coding Scheme
Relation
Constitutive
Formal
Telic
Lexicalized
Unusable

Defining Features
part/whole relation, including material, parts, and contents
descriptive relation, including age, location/origin, shape
function
one or both nouns no longer contribute transparent meaning
no sensible relationship can be determined from context

Example
cowboy-book
mountain-lion
bread-knife
butter-fly
butter-birdie

5.1.1.2 Analysis
The data were analyzed at two levels. First, a series of χ2 analyses were used to
compare the distribution of relationships that adults and children used by head and modifier
natural classes4. For this analysis, all compounds whose head or modifiers were classified as
either animate, inanimate, locative/temporal, human, or artifact were included to make the
dataset parallel to that done in Adult Study 1. The goal of the first analysis was to determine
if children abide by the same tendencies as adults for different natural classes. For example,
do children produce more Telic compounds with artifact head nouns than animate head
nouns? Yates’ correction was used when necessary to adjust for low numbers of items.
Next, the distribution of relationships that adults and children used were compared
at the individual noun level. All nouns that were used in at least three compounds by both
the child and adult were included. The goal of the second analysis was to determine if
children produce novel noun-noun compounds with particular nouns based on the behavior
of those same nouns in the input. For example, if adults only used Telic compounds with the
head noun bowl, do children also only use Telic bowl compounds?

χ2 analyses were used (i) because the categorical nature of the data and (ii) because only four speakers total
were being compared. The author recognizes that another option for future research would be to run the same
data through a mixed effects modeling analysis.
4

85

5.1.2 Results
To summarize the compounds in the corpus, Table 16 shows the count of
compound types by speaker(s). Both children used many noun-noun compounds that were
not also used by the adults in the corpus (Adam 270 and Sarah 83). These compounds were
considered to be the children’s novel productions. The other compounds, those used by the
adults only or by both the adults and child, were considered to be in the adult input for all
analyses, even if not used in the same file. Of course we cannot really say which compounds
the children never have heard before, only approximate this set of compounds by ruling out
all of those that the adults ever say.
Table 16. Count of Compound Types by Speaker(s)
Corpus
Adam
Sarah

Adult only
171
192

Child
and adult
282
89

Child only
270
83

5.1.2.1 Head Noun Natural Class Analysis Results
Considering first the natural class of the heads, Figures 15 and 16 show the
distribution of relationships by head natural class and speaker for the Adam and Sarah files,
respectively. Considering the overall patterns, the distribution of relationships differs
systematically based on the head natural class. For example, compounds with animate natural
kind heads were most often formal compounds, while compounds with artifact natural kind
heads were often telic compounds.
A series of χ2 analyses were used to compare the distribution of relationships used
for each head natural class in the children’s speech and in their input. First, the distribution
of relationships was compared for each natural class for the child and the input. See Table 17
for the results of the χ2 analyses. For both Adam and Sarah, the distribution of relationships

86

used by the child matched the distribution of relationships used by the adults in the same
files.

Proportion of Relationship

Figure 15. Distribution of Relationships by Head Class and Speaker for Adam Files

1.0
.9
.8
.7
.6
.5
.4
.3
.2
.1
.0

Constitutive
Telic
Formal

Adam Adult Adam Adult Adam Adult Adam Adult Adam Adult
Animate

Inanimate Locative/
Temporal

Human

Artifact

Proportion of Relationship

Figure 16. Distribution of Relationships by Head Class and Speaker for Sarah Files
1.0
.9
.8
.7
.6
.5
.4
.3
.2
.1
.0

Constitutive
Telic
Formal

Sarah Adult Sarah Adult Sarah Adult Sarah Adult Sarah Adult
Animate

Inanimate Locative/
Temporal

87

Human

Artifact

Table 17. χ2 Results on the Distribution of Relationships by Head Natural Class
Head
Natural Class
Animate
Inanimate
Loc/Temp
Human
Artifact

Adam vs. Input
χ (2, N = 89) = 0.053, p = .974
χ2(2, N = 39) = 0.616, p = .735
χ2(2, N = 78) = 0.034, p = .983
χ2(2, N = 66) = 2.877, p = .237
χ2(2, N = 348) = 2.365, p = .307
2

Sarah vs. Input
χ (1, N = 30) = 0.019, p = .890
χ2(2, N = 18) = 1.897, p = .387
χ2(2, N = 49) = 0.063, p = .969
χ2(1, N = 23) = 0.276, p = .599
χ2(2, N = 192) = 0.598, p = .603
2

5.1.2.2 Modifier Noun Natural Class Analysis Results
Considering now the natural class of the modifiers, Figures 17 and 18 show the
distribution of relationships by modifier natural class and speaker for the Adam and Sarah
files, respectively. Considering the overall patterns, the distribution of relationships differs
systematically based on the modifier natural class. For example, compounds with human
modifiers were most often formal compounds, while compounds with artifact modifiers
were often telic compounds.

Proportion of Relationship

Figure 17. Distribution of Relationships by Modifier Class and Speaker for Adam Files
1.0
.9
.8
.7
.6
.5
.4
.3
.2
.1
.0

Constitutive
Telic
Formal
Adam Adult Adam Adult Adam Adult Adam Adult Adam Adult
Animate

Inanimate

Locative/
Temporal

88

Human

Artifact

Proportion of Relationship

Figure 18. Distribution of Relationships by Modifier Class and Speaker for Sarah Files
1.0
.9
.8
.7
.6
.5
.4
.3
.2
.1
.0

Constitutive
Telic
Formal

Sarah Adult Sarah Adult Sarah Adult Sarah Adult Sarah Adult
Animate

Inanimate

Locative/
Temporal

Human

Artifact

A series of χ2 analyses were used to compare the distribution of relationships used
for each modifier natural class in the children’s speech and in their input. See Table 18 for
the results of the individual χ2 analyses. For the Sarah files, the distribution of relationships
used by the child matched the distribution of relationships used by the adults for all modifier
natural classes. For the Adam files, the distribution of relationships used by the child
matched the distribution of relationships used by adults for all modifier classes except
Artifact.
Table 18. χ2 Results on the Distribution of Relationships by Modifier Natural Class
Modifier
Natural Class
Animate
Inanimate
Loc/Temp
Human
Artifact

Adam vs. Input
χ (2, N = 90) = 3.531, p = .171
χ2(2, N = 134) = 0.003, p = .999
χ2(2, N = 66) = 2.539, p = .281
χ2(2, N = 100) = 0.613, p = .736
χ2(2, N = 223) = 11.378, p = .003
2

Sarah vs. Input
χ (2, N = 73) = 2.349, p = .309
χ2(2, N = 59) = 0.236, p = .889
χ2(2, N = 32) = 0.366, p = .832
χ2(2, N = 42) = 0.351, p = .839
χ2(2, N = 98) = 0.943, p = .624
2

To summarize, at the natural class level, both Adam and Sarah closely matched their
input in terms of the distribution of relationships for heads and modifiers. The distribution
of relationships used by the adults and children mirrors the findings from Adult Study 1,
89

which showed that different natural classes behave in different ways. For example, in the
current dataset, animate natural kind head nouns were most often part of Formal
compounds, while artifact head nouns were often part of Telic compounds.
5.1.2.3 Noun Analysis
Figures 19 and 20 show the distribution of relationships for individual nouns in
Adam’s speech and in his input. It can be seen that for both head and modifier nouns, Adam
used a slightly or completely different distribution of relationships than was used in his
input. Sarah’s files did not have enough compounds with the same heads and modifiers to
look at this comparison.

1.0
.9
.8
.7
.6
.5
.4
.3
.2
.1
.0

Constitutive
Telic

baby cow-boy

fire

circus

90

fish

toy

Adult

Adam

Adult

Adam

Adult

Adam

Adult

Adam

Adult

Adam

Adult

Adam

Adult

Formal

Adam

Proportion of Relationship

Figure 19. Distribution of Relationships by Modifier Noun and Speaker for Adam Files

paper

1.0
.9
.8
.7
.6
.5
.4
.3
.2
.1
.0

Constitutive
Telic
Formal

Adam
Adult
Adam
Adult
Adam
Adult
Adam
Adult
Adam
Adult
Adam
Adult
Adam
Adult
Adam
Adult
Adam
Adult
Adam
Adult
Adam
Adult

Proprtion of Relationship

Figure 20. Distribution of Relationships by Head Noun and Speaker for Adam Files

man boy house car time box ball hat truck book train
Additionally, Adam and Sarah used many nouns in both head and modifier position
that were not used in the input at all (see Table 19). Excluding those also used by the adults,
Adam used 89 nouns as heads and 112 nouns as modifiers while Sarah used 75 nouns as
heads and 64 nouns as modifiers. For these nouns in these positions, children have no prior
word-specific input on which they can base their compound productions.
Table 19. Count of Individual Noun Types by Position and Speaker(s)

Noun Position
Head
Modifier

Adult
only
169
190

Adam
Sarah
and Adam Adult and Sarah
adult only
only Adult only
69
89
116
31
75
67
112
114
35
64

5.1.2.3 Summary of Results
At the natural class level, both Adam and Sarah were very well-matched to their
input in terms of the distribution of relationship used by natural class. Like adults, children’s
productions followed our expectations of natural classes. For instance, children were more

91

likely to use compounds with artifact head nouns to denote Telic relationships than
compounds with animate head nouns.
At the noun level, however, Adam used head and modifier nouns with a different
distribution of relationships than the distribution of relationships used with those same
nouns in his input. Moreover, both Adam and Sarah used a large number of nouns in
compounds for which the corpus reflected no adult forms in the input.
5.1.3 Discussion
The goal of this study was to provide evidence that children use natural class-level
information, rather than item-specific information, to guide their novel noun-noun
compound productions. The results of the study supported both of our predictions for
children’s production of novel noun-noun compounds. First, the results of the natural class
analyses showed that children, like adults, systematically produce novel compounds based on
natural class information. Children were shown to match their parents, almost completely, in
terms of their relationship use for both head and modifier nouns. These patterns reflected
the fact that particular natural classes are more likely to lead to particular types of
relationships (Downing 1977).
The results also supported the prediction that children do not necessarily match their
input in terms of the distribution of relationships used for individual nouns. This prediction
was supported by the nouns that Adam and Sarah used in their novel compounds but did
not hear in their input. This finding runs counter to a claim that children are forming wordspecific compound templates (Krott and Nicoladis 2005, Krott et al. 2009).
In summary, the results suggest that, at least when producing novel noun-noun
compounds, children use the same information as adults (natural classes) to guide their

92

productions. In the next study we use an elicited production task to further test children’s
ability to produce novel noun-noun compounds across a range of meanings.
5.2 Child Study 2: Elicited Production of Noun-Noun Compounds
This study examines the likelihood of children to produce noun-noun compounds
based on the proposed relationship between two nouns. The goal of this study was to
provide evidence that children produce noun-noun compounds for the same set of
relationships that adults produce noun-noun compounds. In an elicited production task
Hiramatsu et al. (2000) found that children were unlikely to use compound forms for
descriptions of the appearance of an object. For example, children were unlikely to use handchair for a chair that looks like a hand. The authors argued that this was due to a subset
principle whereby English-speaking children begin by speaking a language like Hungarian,
which does not have shape/appearance compounds. Alternatively, a usage-based theory
might predict that children do not produce novel compounds with relationships for which
they have not been given sufficient evidence in the input. For example, a child may not
produce hand-chair for a chair shaped like a hand either because they have not previously
heard a compound with the head noun chair that is defined by its shape or because, in
general, a child has not heard a sufficient number of compounds with shape descriptions to
have a productive template.
To examine the impact of the relationship on children’s production of novel
compounds, an elicited production task was conducted that contained items with
descriptions from all four Qualia Roles (Agentive, Constitutive, Formal, Telic). Children’s
productions were compared to adults’ productions for the same items to determine if
children avoid using compound forms in situations where adults are likely to use them.

93

The hypothesis put forward here predicted that, given knowledge of the constituents
in the target compound and noun-noun compound syntax, children should use noun-noun
compounds in the same situations where adults use them.
5.2.1 Method
5.2.1.1 Participants
29 children (17F, 12M) and 15 adults (12F, 3M) participated in the study. 13 children
made up the Younger Children group (3;0 – 5;0: Mean 4;2) and 16 children the Older
Children group (5;7 – 6;7: Mean 6;2). Children were recruited from local schools in Western
Massachusetts and adults from an undergraduate linguistics course at Michigan State
University. The adult participants received extra credit in their course for their participation.
5.2.1.2 Procedure
There were a total of 16 test items comprised of four descriptions types: agentive,
constitutive, formal, and telic. Each description type represents a different way that the
lexical structure of a noun can be modified by another noun. For each description type, two
head nouns were selected, and for each head noun, two modifiers were selected, for a total
of four items for each description type. See Table 20 for a breakdown of the items by
description type.

94

Table 20. NNC Elicited Production: Items
Description
Type

Head
Noun
puddle

Agentive
shadow
book
Constitutive
table
chair
Formal
van
box
Telic
blanket

Description
This is a puddle made by rain.
This is a puddle made by paint.
This is a shadow made by a house.
This is a shadow made by a tree.
This is a book about cowboys.
This is a book about princesses.
This is a table made of books.
This is a table made of blocks.
This is a chair that is shaped like a flower.
This is a chair that is shaped like a hand.
This is a van with a bunny on it.
This is a van with a monkey on it.
This is a box to put jackets in.
This is a box to put shirts in.
This is a blanket for cats to lay on.
This is a blanket for dogs to lay on.

Target
Compound
rain-puddle
paint-puddle
house-shadow
tree-shadow
cowboy-book
princess-book
book-table
block-table
flower-chair
hand-chair
bunny-van
monkey-van
jacket-box
shirt-box
cat-blanket
dog-blanket

To begin the task, the child was introduced to Monkey, a curious animal who liked to
learn things only from children. The following instructions were read to the child:
(12) Noun-Noun Compound Elicited Production Instructions
“This is my friend Monkey. There’s something that you need to know about this
monkey. She’s a funny monkey who only likes listening to children. She has a
problem, though, because she sees all of these new things, but she’s not sure what
to call them. Do you think you can help her? I will help with you the first couple of
names and then you can help Monkey with the rest, okay?”
Two practice items were introduced by Monkey and named by the experimenter using nounnoun compound forms. For example, one of the images was a cage for birds, a bird-cage, and
the other was a cage for hamsters, a hamster-cage. The names were told to the children, who
were then instructed to repeat the names to Monkey. After the child succeeded in doing this,
the experimenter moved on to the test items.
95

Test items were presented in pairs, with each slide showing the two items a common
head noun (e.g., book-table and block-table). The contrast created by introducing the set of
items together was intended to encourage participants to used compounds. See Figure 21 for
a sample item. Children were given two chances to come up with a noun-noun compound
form for each item. If they did not give a noun-noun compound by their second form, the
experimenter moved on to the next item.
Figure 21. NNC Elicited Production: Sample Item

“Look! This is a chair that’s shaped like a
flower, and this is a chair that’s shaped like a
hand! What could we call a chair that’s shaped
like a flower? (Child responds.) And what
could we call a chair that’s shaped like a chair?
(Child responds.)”
Children were tested individually either in a quiet room or in a quiet part of their
classroom. Each child was seated in front of a 12” laptop, approximately 24 inches from the
screen. All sessions were audio-recorded. The stimuli were displayed using Microsoft
Powerpoint. All images were on a white background and the audio was recorded by the
author, with a minor pitch and formant manipulation to create the Monkey’s voice.
96

Generally the stimuli were played over the laptop speakers, but in cases where sound from
the laptop speakers was not sufficient, the children wore headphones. The adults were tested
in a group, with the visual stimuli displayed on an overhead projector and the audio stimuli
played over classroom speakers.
5.2.1.3 Coding
Production forms were coded based on the target compound. If participants
produced the target compound in modifier-head order the response was coded as Target. If
participants produced the target words in head-modifier order the response was coded as
Reversal. If participants produced a full phrase, for example, “a box for shirts” for shirt-box
the response was coded as Phrasal. If participants produced any other form, for example just
a noun (a box) or a possessive (shirt’s box), the response was coded as Other.
5.2.1.4 Analysis
The items were recoded to isolate each major production type (Target, Reversal,
Phrasal) as the dependent variable for four separate ANOVAs. For example, for the Target
ANOVA, items with Target productions (modifier-head) were coded as “1” and all other
items (Reversal, Phrasal, Other) were coded as “0”. For each of the four analyses, the
proportion of the production type in question (Target for the Target ANOVA) served as the
dependent measure for a 3 Age Group (Younger Children, Older Children, Adults) X 4
Relationship (Agentive, Constitutive, Formal, Telic) ANOVA. A Bonferroni correction was
used for all post-hoc tests to adjust for multiple comparisons.
5.2.2 Results
Table 21 shows the proportion (and count) of forms produced by Item Type and
Age Group. Figures 22, 23, and 24 show the proportion of forms used by Description Type
for Younger Children, Older Children, and Adults, respectively.
97

Table 21. Proportion (and Count) of Forms by Item Type and Age Group

Target

Reversal

Phrasal

Other

Agentive
Constitutive
Formal
Telic
Target Overall
Agentive
Constitutive
Formal
Telic
Reversal Overall
Agentive
Constitutive
Formal
Telic
Phrasal Overall
Agentive
Constitutive
Formal
Telic
Other Overall

Younger
Children
.42 (22)
.71 (37)
.69 (36)
.63 (33)
.62 (128)
.17 (9)
.00 (0)
.08 (4)
.04 (2)
.07 (15)
.13 (7)
.15 (8)
.12 (6)
.13 (7)
.13 (28)
.27 (14)
.13 (7)
.12 (6)
.19 (10)
.18 (37)

98

Older
Children
.70 (39)
.70 (39)
.73 (41)
.82 (46)
.74 (165)
.09 (5)
.07 (4)
.09 (5)
.04 (2)
.07 (16)
.14 (8)
.16 (9)
.14 (8)
.11 (6)
.14 (31)
.07 (4)
.07 (4)
.04 (2)
.04 (2)
.05 (12)

Adults
.88 (53)
.90 (54)
.90 (54)
.97 (58)
.91 (219)
.00 (0)
.00 (0)
.00 (0)
.00 (0)
.00 (0)
.03 (2)
.07 (4)
.10 (6)
.03 (2)
.05 (14)
.08 (5)
.03 (2)
.00 (0)
.00 (0)
.04 (7)

Overall
.68 (114)
.77 (130)
.78 (131)
.82 (137)
.76 (512)
.08 (14)
.02 (4)
.05 (9)
.02 (4)
.05 (31)
.10 (17)
.13 (21)
.12 (20)
.09 (15)
.11 (73)
.14 (23)
.08 (13)
.05 (8)
.07 (12)
.08 (56)

Figure 22. NNC Production: Proportion of Production Types by Younger Children

Proportion of Form

1.0
.8
Other

.6

Phrasal
.4

Reversal
Target

.2
.0
Agentive Constitutive

Formal

Telic

Figure 23. NNC Production: Proportion of Production Types by Older Children

Proportion of Form

1.0
.8
Other

.6

Phrasal
.4

Reversal
Target

.2
.0
Agentive

Constitutive

Formal

Telic

Figure 24. NNC Production: Proportion of Production Types by Adults

Proportion of Form

1.0
.8
Other

.6

Phrasal

.4

Reversal

.2

Target

.0
Agentive Constitutive

Formal
99

Telic

5.2.2.1 Target Form Analysis
A 4 Description Type (Agentive, Constitutive, Formal, Telic) X 3 Age Group
(Younger Children, Older Children, Adults) ANOVA was carried out on the proportion of
Target forms. The main effect of Description Type was significant (F(1, 39) = 6.009, p =
.019). Post-hoc comparisons revealed that, overall, participants produced target forms with
Agentive Description Types marginally less often than Telic Description Types (p = .088).
There were no other differences in proportion of target productions by Description Type.
The main effect of Age Group was also significant (F(2, 39) = 3.648, p = .035). Posthoc comparisons revealed that Adults produced significantly more target forms than
Younger Children (p = .033) There was no difference in the degree of target productions
between Younger Children and Older Children (p = .870) or between Older Children and
Adults (p = .344).
The interaction between Description Type and Age Group was not significant (p =
.137).
5.2.2.2 Reversal Form Analysis
A 4 Description Type (Agentive, Constitutive, Formal, Telic) X 3 Age Group
(Younger Children, Older Children, Adults) ANOVA was carried out on the proportion of
Reversal forms. The main effect of Age Group was significant (F(2, 39) = 3.572, p = .038).
Post-hoc comparisons revealed that Adults produced marginally less reversal forms than
Younger Children (p = .080) and Older Children (p = .084). There was no difference in the
degree of reversal forms between Younger Children and Older Children (p = 1.000).
The main effect of Description Type was marginally significant (F(3, 117) = 2.316, p
= .079) and the interaction between Description Type and Age Group was not significant (p
= .216).
100

5.2.2.3 Phrasal Form Analysis
A 4 Description Type (Agentive, Constitutive, Formal, Telic) X 3 Age Group
(Younger Children, Older Children, Adults) ANOVA was carried out on the proportion of
Phrasal forms. The main effects of Description Type (p = .696) and Age Group (p = .533)
were not significant. The interaction between Description Type and Age Group was also not
significant (p = .922).
5.2.2.4 Summary of Results
The likelihood of a participant producing a target form increased over developmental
time, while the likelihood of a participant producing any other type of form decreased.
Importantly, however, there were no significant interactions between Description Type and
Age Group, suggesting that children and adults produce noun-noun compounds in a similar
manner for each Description Type. Younger Children, Older Children, and Adults were all
likely to produce target forms for Constitutive, Formal, and Telic descriptions, but less likely
for Agentive descriptions.
5.2.3 Discussion
The goal of this study was to provide evidence that children production noun-noun
compounds for the same set of relationships that adults do. The results of the study support
the prediction made by this dissertation. While it was true that children were less likely to
produce target forms for certain description types (agentive, telic), adults were also less likely
to produce target forms for these same description types. The results of the study fail to
support the claim by Hiramatsu et al. (2000) that children avoid shape/appearance
compounds. In fact, the shape/appearance formal descriptions received the most target
forms overall of the four description types.

101

Taken together, the results of the first two studies suggest that children are quite
adult-like in their production of noun-noun compounds. The corpus study showed that (i)
children are adult-like in their use of natural classes and (ii) children are not dependent on
word-specific items in the input to produce novel compounds. The current study provides
evidence that children use the full range of relationships made available by the Qualia
Structure of words when producing novel compounds. In the final study for this chapter we
examine how children use this same lexical semantic information when interpreting nounnoun compounds.
5.3 Child Study 3: Preferred Interpretations of Noun-Noun Compounds
This study examined children’s preferred interpretations of noun-noun compounds.
The first goal of this study was to provide evidence that children use natural class
information, and not individual noun information, to generate interpretations for novel
compounds. The second goal of this study was to provide evidence that, given familiar
lexical information, children interpret compounds as having intrinsic relationships, rather
than accidental/temporary relationships, as was found by Krott et al. 2010.
5.3.1 The Role of Lexical Semantics in Children’s Interpretation of Compounds
The first goal of this study was to provide evidence that children use natural class
information, and not individual noun information, to generate interpretations for novel
compounds. More specifically, the goal was to provide evidence that children are using a
generative mechanism, like that proposed in the first half or this dissertation, rather than a
usage-based mechanism to interpret compounds. Krott et al. (2009) found in an
interpretation task that children interpret novel compounds by analogy to known
compounds with the same head noun. For instance, if a blueberry-spoon was known to be a
spoon for eating blueberries, then a cake-spoon was interpreted as a spoon for eating cake.
102

They suggest that this ease of interpretation is due to word-specific templates that supply the
head noun and the relationship for the compound with a slot for the modifier noun. Their
findings, however, can also be explained by a different mechanism. Spoon is an artifact, and
as we have shown in both an adult and a child study now, artifacts are most likely to be
interpreted with a functional reading (a spoon that scoops up X). Therefore, the findings by
Krott et al. may actually point to the same bias that we are suggesting is in the adult system:
depending on the natural classes of the constituents in a compound, speakers are inclined to
interpret compounds using particular parts of their lexical structure (Downing 1977).
To examine if children use word-specific or natural class knowledge in their
interpretation of compounds, the study was designed such that participants interpreted a
number of compounds that all share the same artifact head noun. For example, the head
noun bowl was used in three items: orange-bowl, cake-bowl, and beach-bowl. For each item a
functional, telic interpretation was available (a bowl that oranges go in, a bowl that cake goes
in, a bowl that beach goes in) in addition to a second, possible interpretation (a bowl that is
next to an orange, a bowl that is shaped like a cake, a bowl that is used on the beach). This
design allowed us to pit the expectations of a generative compound interpretation system,
like that proposed in the first half of this dissertation, against a item-based interpretation
system, like that proposed in Krott et al. (2009). Given that bowl is an artifact, a functional,
telic relationship is the preferred relationship based on the behavior of natural classes. Given
that bowl is likely usually defined in a functional way, a template with ___-bowl with a
functional relationship is also the most likely template for children to develop. Where these
two theories make different predictions is when the normal tendency of artifacts as
functional is overruled by the behavior of the modifier noun.

103

The hypothesis put forward here predicted that children would change the
relationship in their interpretation of a compound based both on its head lexical information
and modifier lexical information, in the same way that adults do. In other words, even in the
absence of examples in their input of a particular artifact head noun behaving in a nonfunctional way, children should be able to choose a non-functional interpretation.
5.3.2 The Role of Intrinsic/Accident Information in Children’s Use of Compounds
The second goal of this study was to provide evidence that, given familiar lexical
information, children interpret compounds as having intrinsic relationships, rather than
accidental/temporary relationships, as was found by Krott et al. 2010. In a production study,
Clark et al. showed that children, like adults, were more likely to produce noun-noun
compounds for intrinsic properties than accidental or temporary properties. For instance,
both children and adults were more likely to use pumpkin-house for a house made of pumpkin
(intrinsic) than for a house with a pumpkin near it (accidental). Krott et al. (2010), however,
found that children were more likely to interpret novel objects with accidental or temporary
relationships. For instance, children, unlike adults, were more likely to interpret a wug-moop as
a moop that has a wug on it than a moop that has a functional relationship to wugs.
Critically, a major difference between these two study designs was the familiarity of the
words used. In Clark et al.’s study, where children were adult-like in their preference to use
noun-noun compounds to name constituents related by an intrinsic property, all items were
known words. For these words, children had access to all of the information that they had
learned up to that point about these words. In Krott et al.’s study, where children were nonadult-like in their preference to interpret noun-noun compounds as representing accidental
or temporary relationships, all items were nonce words. For these words, children had access
to only the information that they could ascertain from the study. While Krott et al. claim that
104

this shows a bias that young children have for physically apparent relationships (Smith et al.
1996), the system proposed in this dissertation presents a simpler solution: children
preferred physically apparent relationships for nonce word compounds because their system
for interpreting compounds relies on lexical knowledge to determine a likely relationship,
and no lexical information was present, leaving physical information as the strongest
information source.
This study was designed to examine this claim in three ways. First, the head and
modifier nouns used in the study were well-known by children in the tested age group (3- to
6-year-olds) according to the MacArthur’s CDI (Fenson et al. 2006). Therefore, any bias
towards a physically apparent interpretation could not be due to a lack of lexical information
to provide any other possible interpretation. Second, one type of item was specifically
included that pinned a telic reading of a compound against a physically apparent reading of a
compound. For example, children had to decide if the better interpretation of an orange-bowl
was “a bowl that is used to put oranges in” (telic) or “a bowl that is next to an orange”
(physically apparent). Finally, while the head nouns and modifier nouns were introduced
using pictures, the two interpretations of the compound that those nouns form were not
pictured. By not presenting pictures of the two options, children could not be biased by the
pictures towards a referent where both the head noun and modifier noun were physically
present.
The hypothesis put forward here predicted that, given sufficient lexical information,
children should have preferred telic interpretations of compounds to physically apparent
interpretations. That is, in the orange-bowl example above, children should have preferred the
telic reading (a bowl that is used to put oranges in) to the location reading (a bowl that is
next to an orange).
105

5.3.3 Method
5.3.3.1 Participants
31 children (19F, 12M) and 15 adults (12F, 3M) participated in the study. 15 children
made up the Younger Children group (3;0 – 5;0: Mean 4;2) and 16 children the Older
Children group (5;7 – 6;7: Mean 6;2). Children were recruited from local schools in Western
Massachusetts and adults from an undergraduate linguistics course at Michigan State
University. The adult participants received extra credit in their course for their participation.
5.3.3.2 Materials and Procedure
The noun-noun compound interpretation task was a 3 Item Type (Location of Use,
Appearance, Temporary Position) X 3 Age Group (Younger Children, Older Children,
Adults) design where Item Type was a within-subjects variable. Six artifact nouns were
chosen as head nouns for this task. Each head noun was then paired with three different
nouns, one for each of the different Item Types, for a total of 18 items. See Table 22 for the
full list of items.

106

Table 22. NNC Interpretation: Items
Head Noun
bed
bowl
oven
pillow
soap
spoon

Compound
mountain-bed
lion-bed
girl-bed
beach-bowl
cake-bowl
orange-bowl
forest-oven
apple-oven
muffin-oven
playground-pillow
monkey-pillow
boy-pillow
hospital-soap
penguin-soap
foot-soap
school-spoon
strawberry-spoon
peanut-spoon

Item Type
Location
Appearance
Temporary
Location
Appearance
Temporary
Location
Appearance
Temporary
Location
Appearance
Temporary
Location
Appearance
Temporary
Location
Appearance
Temporary

Telic Description
Bed that mountains sleep in
Bed that lions sleep in
Bed that girls sleep in
Bowl that beach goes in
Bowl that cake goes in
Bowl that oranges go in
Oven that bakes forests
Oven that bakes apples
Oven that bakes muffins
Pillow that playgrounds lay on
Pillow that monkeys lay on
Pillow that boys lay on
Soap that cleans hospitals
Soap that cleans penguins
Soap that cleans feet
Spoon that scoops up schools
Spoon that scoops up strawberries
Spoon that scoops up peanuts

107

Other Description
Bed that is used in the mountains
Bed that is decorated with lions
Bed that a girl is laying on
Bowl that is used on the beach
Bowl that is shaped like a cake
Bowl that is next to an orange
Oven that is used in the forest
Oven that is decorated with apples
Oven that is next to a muffin
Pillow that is used on the playground
Pillow that is decorated with monkeys
Pillow that a boy is laying on
Soap that is used at a hospital
Soap that is shaped like a penguin
Soap that is next to a foot
Spoon that is used at school
spoon that is shaped like a strawberry
Spoon that is next to peanuts

The Item Types manipulated the likelihood of a functional, telic interpretation, the
type of interpretation that is most likely for a compound with an artifact head noun. The
first type of item, Location of Use, used a locative noun as a modifier. The descriptions that
were given either used the modifier noun as the location of use (likely interpretation, formal)
or as that which the artifact acted upon (unlikely interpretation, telic). For example, a
mountain-bed is more likely to be a bed that is used in the mountains, rather than a bed that is
used by mountains. The Location of Use item was designed to as a low-telic interpretation
item type.
The second type of item, Temporary Position, used an artifact or a natural kind as a
modifier. The descriptions given either used the modifier as an object that the head noun
was temporarily positioned next to (unlikely interpretation, formal) or as that which the
artifact acted upon (likely interpretation, telic). For example, a boy-bed is more likely to be a
bed that is used by boys, than a bed that is a boy is laying on. The Temporary Position item
was designed as a high-telic interpretation item type.
The final type of item, Appearance, used an artifact or natural kind as a modifier.
The descriptions given either used the modifier as something that modified the head noun’s
appearance (likely interpretation, formal) or as that which the artifact acted upon (likely
interpretation, telic). For example, a lion-bed is equally likely to be a bed for lions to sleep on a
bed that is decorated with lions. The Appearance item was a designed as a neural item type.
To begin the task, the child was introduced to Monkey, a curious animal who liked to
learn things only from children. The following instructions were read to the child:

108

(13) Noun-Noun Compound Interpretation Instructions
“This is my friend Monkey. There’s something that you need to know about this
monkey. She’s a funny monkey who only likes listening to children. She has a
problem, though, because she knows some names for things, but she’s not quite
sure what the names belong to. Do you think you can help her? She is going to start
by telling us about a couple of things that we already know about, and then she’s
going to ask us to figure out what a third thing is. She is going to give us a name and
then two possible descriptions and we have to pick which one it is. Does that sound
okay?”
On each trial the modifier and head nouns were introduced in that order, one at a time with
an image and an audio recording naming the object. On the third slide, Monkey appeared
and the children heard the two possible descriptions. No images were given to match the
two descriptions. See Figure 25 for a sample item. Children responded by choosing the
description that best fit the name under question.
Figure 25. NNC Interpretation: Sample Item

Slide 1

Slide 2

Slide 3

“This is a penguin.”

“And this is soap.”

“But what is penguin soap?
Is it soap that is used to
clean penguins or soap that
is shaped like a penguin?”

109

The general procedure for this task is the same as for the noun-noun compound
elicited production task. If children participated in both the noun-noun compound elicited
production task and the noun-noun compound interpretation task, the interpretation task
was administered on a second day of testing. All adults participated in both the elicited
production task and the interpretation task in the same testing session.
5.3.3.3 Analysis
The data were analyzed in two ways. The purpose of the first analysis was to
determine if participants’ response patterns changed based on the item type and their age.
For the first analysis, a 3 Item Type (Location of Use, Appearance, Temporary Position) X 3
Age Group (Younger Children, Older Children, Adults) ANOVA was computed on the
proportion of Telic responses.
The purpose of the second analysis was to determine if any of the age groups were
more likely to use only Telic interpretations across the board for a given head noun, as they
might if they had formed word-specific templates for these nouns. To determine if this was
the case, each of the six head nouns were considered for each participant to see if it was ever
the case that a participant used telic interpretations for all three instances of that head noun.
If a participant did use telic interpretations across the board, that head noun was coded as a
“1” and if not, that head noun was coded as a “0”. For the second analysis, a 3 Age Group
(Younger Children, Older Children, Adults) ANOVA was computed on the proportion of
head nouns with consistently telic head nouns.
A Bonferroni correction was used for all post-hoc tests to adjust for multiple
comparisons.

110

5.3.4 Results
5.3.4.1 Overall Analysis
Figure 26 shows the proportion of Telic responses by Age Group and Item Type.
The results of the 3 Item Type X 3 Age Group ANOVA revealed a significant main effect of
Item Type (F(2, 84) = 210.327, p < .001). To unpack the significant main effect of Item
Type, pairwise comparisons were carried out on all Item Types, collapsing the data by Age
Group. Telic responses were reported significantly more often for Temporary items than for
Appearance items (p < .001), significantly more often for Temporary items than for Location
of Use items (p < .001), and significantly more often for Appearance items than Location of
Use items (p < .001).
The interaction between Item Type and Age Group was also significant (F(4, 84) =
3.525, p = .010) were not significant. This interaction was likely driven by the tendency for
Younger Children to have more Telic responses than Adults for Location of Use items, but
Adults to have more Telic responses than children for Appearance items.
The main effect of Age Group was not significant (p = .822).

Proportion of Telic Responses

Figure 26. Proportion of Telic Responses By Item Type and Age Group
1.0

Younger Children

.5

Older Children
Adults

.0
Location of Apperance Temporary
Use
Position
111

5.3.4.2 Analysis of Consistency by Head Noun
Figure 27 shows the average proportion of compounds with head nouns that
received consistently Telic responses. The 3 Age Group ANOVA on the proportion of
compounds with consistently telic head nouns revealed that Age Group was not significant
(p = .251). This finding suggests that both younger children and older children were no more
likely than adults to assign only Telic interpretations to any given head noun.

Proportion of Head Nouns

Figure 27. Proportion of Head Nouns with Consistent Telic Interpretation by Age
1.0
.9
.8
.7
.6
.5
.4
.3
.2
.1
.0
Younger
Children

Older
Children

Adults

5.3.4.3 Summary of Results
Children behaved very adult-like in their interpretation of novel compounds. Like
adults, children in both age groups strongly preferred Telic interpretations for the
Temporary Position Items and strongly dispreferred Telic interpretations in the Location of
Use items. Moreover, children were no more likely than adults to prefer Telic interpretations
across the board for a given head noun.

112

5.3.5 Discussion
The goal of this study was to provide evidence that children, like adults, use lexical
semantic information to interpret noun-noun compounds. The results of the study
supported the two predictions made by this dissertation. First, children were adult-like in
their ability to change their interpretation of compounds with the same head noun based on
the lexical semantics of the modifier noun. For example, both children and adults preferred
telic interpretations of compounds with an inanimate or artifact modifier noun, but
dispreferred telic interpretations of compounds with a locative modifier noun. The children
in the study were no more likely than adults to interpret compounds with the same head
noun consistently with the same telic interpretation, suggesting that children rely on more
than a template with head noun and relationship information to interpret novel compounds,
contra Krott et al. (2009).
The results of the study also support the second prediction, that children do not
have a general preference for temporary/accidental interpretation compounds. Rather, when
children have sufficient lexical information they prefer intrinsic relationships, suggesting that
the findings of Krott et al. (2010) were an artifact of the nonce word study design.
5.4 General Discussion
The goal of the studies in this chapter was to provide evidence that children use the
same system as adults to produce and interpret noun-noun compounds. In the first half of
this dissertation we showed that adults use the tendencies of natural classes in conjunction
with the parts of the lexical structure available on the head noun to determine the meaning
of a noun-noun compound. From all three studies, we found results that support the
hypothesis that children interpret compounds using the same mechanism as adults.

113

The first study, a corpus study, compared children’s production of novel noun-noun
compounds to the noun-noun compounds in the children’s input. The results showed that
children were sensitive to the behavior of natural classes, but were not sensitive to behavior
of particular words. For instance, children matched adults in their high production of
compounds with artifact heads and telic relationships. On the other hand, children did not
necessarily match adults in the relationships used with individual head nouns. These findings
are the first piece of evidence that runs contrary to a template-based noun-noun compound
system for children (Krott and Nicoladis 2005, Krott et al. 2009).
The second study, an elicited production study, compared children’s production of
novel noun-noun compounds to adults’ production of novel noun-noun compounds across
the full span of relationships. While previous research had suggested that young children do
not use compounds for particular types of relationships (Hiramatsu et al. 2000), the results
of this study found no evidence for that claim. Across the four relationships, including the
specific examples used in the Hiramatsu study, children’s preference to use a noun-noun
compound form closely matched adults’ preference to use a noun-noun compound form.
For example, both adults and children were very likely to use noun-noun compounds for
formal relationships (a chair shaped like a hand is a hand-chair). Both adults and children were
less likely to use noun-noun compounds for agentive relationships (a shadow caused by a
tree is a tree-shadow). This study provides evidence that children, like adults, are able to use all
parts of a word’s lexical structure to produce compounds.
The last study, a forced-choice interpretation task, compared children’s interpretation
of noun-noun compounds to adults’ interpretation of noun-noun compounds. The results
showed that children were adult-like in two manners in terms of their interpretations. First,
children used the lexical information coming from both the head and modifier nouns to
114

determine the compounds meaning. On the items where a functional, telic interpretation was
available and plausible, children and adults chose this interpretation. On the items where a
functional interpretation was available, but not plausible, children and adults chose a
different interpretation. For example, both children and adults interpreted lion-bed as a bed
that lions sleep in, but did not interpret mountain-bed as a bed that mountains sleep in. Rather,
children and adults preferred a location of use interpretation for mountain bed, a bed that is
used in the mountains.
Taken together, the results of these studies suggest that children use the same
generative mechanism as adults in their production and interpretation of compounds. Rather
than relying on templates, children are able to weigh multiple sources of lexical information
at once to determine which interpretation is most likely for noun-noun compounds. In the
next chapter we explore the syntactic side of the proposed interpretation system to
determine if children are adult-like in their interpretation when syntax, rather than lexical
semantics, determines the relationship in a compound.

115

CHAPTER 6: CHILD VERBAL COMPOUND TASKS
6.0 Introduction
In Chapter 5 I showed that children adhere to the adult system to interpret nounnoun compounds, using lexical information from the head and modifier in noun-noun
compounds to determine the appropriate interpretation. The goal of this chapter is to
provide evidence that children use their knowledge of verbal compound syntax to constrain
their interpretation of verbal compounds in the same way that adults do.
This chapter presents the results of two studies examining children’s acquisition of
verbal compounds. The aim of these studies is to address the questions in (14).
(14) Questions addressed in Chapter 6 Studies
a. How do children learn verbal compound syntax?
b. Does children’s knowledge of verbal compound syntax restrict their
interpretation of verbal compounds in an adult-like way?
The first study, which addressed the question in (14a), examined the set of non-adult verbal
compound forms that children produce, in order to determine how children acquire verbal
compound syntax and what type of syntactic knowledge children should have as they
progress through the acquisition process. The second study, which addressed the question in
(14b), examined the way in which children interpret verbal compounds, given their syntactic
knowledge as measured by their verbal compound productions in the first study.
Taken together, the ultimate goal of this chapter was to provide evidence that
children use the same mechanism as adults to produce and interpret verbal compounds, and
any deviations from adult-like interpretations can be explained by lacking syntactic
knowledge.

116

6.1 Child Study 4: Elicited Production of Verbal Compounds
This study examined how children acquire verbal compound syntax. It is wellattested that children produce a particular set of forms, verb-internal argument catch-mouse
and verb-er-internal argument catcher-mouse for the target mouse-catcher (Clark et al. 1986,
Nagpal and Nicoladis 2009, Gamache and Schmitt to appear). The goals of this study were
(i) to replicate previous findings that children produce this set of non-adult forms (catch-mouse
and catcher-mouse, but not mouse-catch for mouse-catcher) and (ii) to determine a baseline for
individual children’s syntactic knowledge. In the second study in this chapter, we used this
baseline to determine if children’s interpretation of verbal compounds is predicted by their
knowledge of verbal compound syntax.
The hypothesis put forward in this dissertation predicted that children would
produce the same pattern of forms that have been found in previous studies, namely catchmouse and catcher-mouse, but not mouse-catch, as the non-adult forms are intermediate steps in
the adult derivation.
6.1.1 Method
6.1.1.1 Participants
31 children (19F, 12M) and 15 adults (12F, 3M) participated in the study. 15 children
made up the Younger Children group (3;0 – 5;0: Mean 4;2) and 16 children the Older
Children group (5;7 – 6;7: Mean 6;2). Children were recruited from local schools in Western
Massachusetts and adults from an undergraduate linguistics course at Michigan State
University. The adult participants received extra credit in their course for their participation.
6.1.1.2 Materials and Procedure
Four verbs were chosen for each of three verb types (atelic, telic, particle) for a total
of 12 items. See Table 23 for the full list of items. Note that three verb types were used in
117

this study (atelic, telic, particle) to mirror the verb types used in the second study in this
chapter, but for the purposes of determining a baseline for children’s syntactic knowledge,
these verb types have been collapsed.
Table 23. VC Elicited Production: Items
Verb
Type

Verb

wipe
burn
Atelic
push
fight
fix
cut
Telic
buy
wash
set up
drop off
Particle
take away
check out

Prompt

Target Form

This is a woman who wipes tables.
This is a man who burns candles.
This is a man who pushes carts.
This is a boy who fights robots.
This is a man who fixes chairs.
This is a man who cuts onions.
This is a man who buys food.
This is a woman who washes plates.
This is a man who sets up ladders.
This is a woman who drops off kids.
This is a man who takes away cars.
This is a man who checks out computers.

table-wiper
candle-burner
cart-pusher
robot-fighter
chair-fixer
onion-cutter
food-buyer
plate-washer
ladder-setter-upper
kid-dropper-offer
car-taker-awayer
computer-checker-outter

To begin the task, the child was introduced to Monkey, a curious animal who liked to
learn things only from children. The following instructions were read to the child:
(15) Verbal Compound Elicited Production Instructions
“This is my friend Monkey. There’s something that you need to know about this
monkey. She’s a funny monkey who only likes listening to children. She has a
problem, though, because she meets all of these new people, but she’s not sure what
to call them. Do you think you can help her? I will help with you the first couple of
names and then you can help Monkey with the rest, okay?”
Two practice items were introduced by Monkey and named by the experimenter using verbal
compound forms in full sentences. For example, children were shown a picture of a girl
reading a book and were told “This is a girl who reads books. We could call a girl who reads
books a book-reader. Can you tell Monkey that she is a book-reader?” After the child
118

repeated the compounds from the practice items to Monkey, the experimenter moved on to
the test items.
On each trial an image was presented along with an audio recording that described
the person in the picture. See Figure 28 for a sample item. Children were given two chances
to come up with a verbal compound form for each item. If they did not give a verbal
compound by their second attempt, the experimenter moved on to the next item. For each
response, children were prompted to respond in a full sentence (“He is a X.”/”She is a X.”)
to ensure that the forms that children were producing were in fact nominal forms and not
simply repetitions of the final words of the prompts.
Figure 28. VC Elicited Production: Sample Item

“Look! This is a man who cuts onions. What
could we call a man who cuts onions?”
Children were tested individually either in a quiet room or in a quiet part of their
classroom. Each child was seated in front of a laptop with a 12 inch screen approximately 24
inches from the computer. All sessions were audio-recorded. The stimuli were displayed
using Microsoft Powerpoint. All images were on a white background and the audio was
recorded by the author, with a minor pitch and formant manipulation to create the Monkey’s
voice. Generally the stimuli were played over the laptop speakers, but in cases where the
119

speakers were not sufficiently loud, the children wore headphones. The adults were tested in
a group, with the visual stimuli displayed on an overhead projector and the audio stimuli
played over classroom speakers.
6.1.1.3 Coding and Analysis
Production forms were coded by according to the position of each morpheme.
Verbs were coded as V, internal arguments as O (object), agents as A, particles as P, and
morphemes as reflected in their English orthography (e.g., er, ing). Simple nouns (N) and
noun-noun compounds (NNC) were coded accordingly. Phrasal responses, for example “a
man who drives trucks”, were coded as Phrasal. A sample of this coding scheme is shown in
Table 24.
Table 24. VC Elicited Production: Coding Scheme
Production Form
truck-driver
driver-man
pick-up-car
chauffeur
car-man

Code
OVer
VerA
VPO
N
NNC

For analysis, coding was simplified in a few ways. First, while they were coded separately, any
form that included both the agent (A) and –er morphology, for example, driver-man, was
collapsed with those forms that included only the –er morphology. Second, forms that
included a particle were coded with the particle, for example OVerPer for car-picker-upper, but
were collapsed with non-particle forms for the purposes of the general analyses. That is, a
form of OVerPer and OVer were both coded as OVer. Finally, any forms that did not meet a
threshold of appearing more than 5% of the time for any age group were collapsed into an
Other category. Note that the Other category made up less than 6% of forms for each age
group. See Appendix C for the full set of forms produced in the study.
120

The results were analyzed in three ways. First, the forms that children and adults did
produce were examined in light of the full range of possible forms that the children and
adults could have produced. That is, of all of the possible permutations of the morphemes in
a verbal compound, which permutations were produced and which failed to be produced?
For a form to be considered reliably produced, it had to be produced by more than one child
and by each child more than one time.
Second, the distribution of the forms that were deemed representative of children’s
learning path by the first analysis (OVer, VerO, VO) were examined to determine if the
forms proceed in a way that is reflective of steps, as they are proposed to be reflective of the
steps in the adult derivation. A series of one-way Age Group (Younger Children, Older
Children, Adults) were carried out on the proportion of each form.
Finally, the overall distribution of production forms was examined to determine if
children and adults use verbal compound forms to the same degree. For this analysis, forms
were collapsed into four groups: verbal compound forms (OVer, VerO, and VO), other
nominal forms (nouns, noun-noun compounds, and Ver), phrasal forms, and other forms. A
series of one-way Age Group (Younger Children, Older Children, Adults) ANOVAs were
carried out on the proportion of verbal compound forms, the proportion of other nominal
forms, and the number of phrasal forms.
A Bonferroni correction was used for all post-hoc tests to adjust for multiple
comparisons.
6.1.2 Results
6.1.2.1 Analysis of Verbal Compound Permutations
Table 25 shows the proportion and counts of possible permutations of verbal
compounds. Only four permutations reached the reliability threshold: OVer (the target),
121

VerO, VO, and Ver. Of these permutations, OVer and Ver, were used by both child and adult
participants and are attested in the adult grammar. VerO and VO were not used by adult
participants, but VerO was used by nine children, and VO by ten children.
Table 25. Proportion (and Count) of Verbal Compound Permutations
Form

Example

Younger
Older
Children Children
OVer
truck-driver .22 (35) .47 (91)
VerO
driver-truck .06 (10) .04 (8)
VOer
drive-trucker
.01 (1)
.00
(0)
OerV
trucker-drive
.00 (0)
.00
(0)
VO
drive-truck .17 (26) .04 (7)
OV
truck-drive
.01 (1)
.01
(1)
Ver
driver
.15 (24) .20 (38)
Oer
trucker
.00 (0)
.00
(0)
O
truck
.01 (1)
.01
(1)
V
drive
.01 (1)
.00
(0)
Other
.37 (57) .24 (46)
Grey = attested adult forms.
Bolded = forms that surpassed reliability threshold.

Adults
.85 (153)
.00 (0)
.00 (0)
.00 (0)
.00 (0)
.00 (0)
.02 (3)
.00 (0)
.00 (0)
.00 (0)
.13 (24)

Six permutations of the verbal compound morphemes were not reliably produced:
VOer, OerV, OV, Oer, O, and V. The absence of three of these forms, VOer, OerV, and Oer,
suggests that children understand that –er merges with verbs, not nouns. The absence of just
O or just V forms suggests that children are minimally able to merge the verb and internal
argument when they begin to produce verbal compound forms. Finally, what does the
absence of OV suggest about children’s knowledge of verbal compound syntax? Or, more to
the point, what does the presence of VO and VerO and the absence of OV suggest? As I
have proposed in previous research (Gamache and Schmitt to appear), this pattern reflects
the steps in the adult verbal compound derivation, which children are learning step-by-step.
In other words, VO and VerO are partial steps in the verbal compound derivation, while OV
is an impossible set in the derivation, since –er must merge into the structure prior to the

122

internal argument (O) moving to the left of the verb. For the next two analyses VO and
VerO, in addition to the target OVer, are all considered verbal compound forms.
6.1.2.3 Analysis of Distribution of Verbal Compound Forms
Figure 29 shows the proportion of verbal compound forms by age group. A series of
one-way 3 Age Group (Younger Children, Older Children, Adults) ANOVAs were
performed on the proportion of each form type (VO, VerO, OVer).

Proportion of Form

Figure 29. VC Elicited Production: Proportion of Verbal Compound Forms
1.0
.9
.8
.7
.6
.5
.4
.3
.2
.1
.0

VO
VerO
OVer

Younger
Children

Older
Children

Adults

The ANOVA performed on the proportion of OVer forms revealed a main effect of
Age Group (F(2, 41) = 19.022, p < .001). Post-hoc comparisons revealed that both Adults (p
< .001) and Older Children (p = .001) were more likely than Younger Children to use target
OVer forms. Older Children were marginally more likely than Younger Children to use target
forms (p = .054).
The ANOVA performed on the proportion of VerO forms revealed no main effect
of Age Group (p < .136).
Finally, the ANOVA performed on the proportion of VO forms revealed a
significant main effect of Age Group (F(2, 41) = 4.037, p = .025). Post-hoc comparisons
123

revealed that Younger Children were significantly more likely than Adults to use VO forms
(p = .028). There were no differences between Younger Children and Older Children (p =
.110) or Older Children and Adults (p = 1.000).
In summary, the proportion of VO and VerO forms decreased over time, while the
proportion of OVer increased.
6.1.2.4 Analysis of Distribution of Overall Forms
Figure 30 shows the distribution of form types by age group. A series of one-way 3
Age Group (Younger Children, Older Children, Adults) ANOVAs were performed on the
proportion of each form type (VO, VerO, OVer).

Proportion of Form

Figure 30. VC Elicited Production: Proportion of Overall Forms by Age Group
1.0
.9
.8
.7
.6
.5
.4
.3
.2
.1
.0

Other
Phrasal
Other Nominal Form

Verbal Compound Form

Younger
Children

Older
Children

Adults

The ANOVA performed on the proportion of Verbal Compound Forms revealed a
main effect of Age Group (F(2, 41) = 7.634, p = .002). Post-hoc comparisons showed that
Adults used significantly more Other Forms than Younger Children (p = .002) and Older
Children (p = .016). There was no difference between Younger Children and Older Children
in the use of Verbal Compound Forms (p = 1.000).

124

The ANOVA performed on the proportion of Other Nominal Forms revealed a
main effect of Age Group (F(2, 41) = 3.940, p = .027). Post-hoc comparisons showed that
Younger Children used significantly more Other Nominal Forms than Adults (p = .035).
There were no differences between Younger Children and Older Children (p = 1.000) or
Older Children and Adults (p = .188) in the use of Other Nominal Forms.
The ANOVA performed on the proportion of Phrasal Forms revealed a main effect
of Age Group (F(2, 41) = 5.841, p = .020). However, post-hoc comparisons showed no
significant between-group differences after Bonferroni’s correction.
Finally, the ANOVA performed on the proportion of Other Forms revealed a main
effect of Age Group (F(2, 41) = 5.139, p = .010). Post-hoc comparisons showed that Adults
used significantly more Other Forms than Younger Children (p = .011) and marginally more
Other Forms that Older Children (p = .077). There was no difference between Younger
Children and Older Children in the use of Other Forms (p = 1.000).
In summary, children used less verbal compound forms than adults, and instead
frequently produced other adult-like, but non-OVer nominal and phrasal forms.
6.1.2.5 Summary of Results
Children produced only four possible permutations of the morphemes in verbal
compounds: OVer, VerO, Ver, and VO. Ver represents a different adult target form. The
other three forms were shown to progress as expected in development: VerO and VO were
more prevalent in the children’s productions and decreased with age, while OVer was more
prevalent in adults’ productions and increased in children’s productions with age. Moreover,
in the overall distribution of forms produced, adults used predominantly target verbal
compound forms, while children, who produced less target verbal compound forms, tended
to use a wider range of other adult forms, such as Ver or noun-noun compounds.
125

6.1.3 Discussion
The results of this study replicated the findings of previous studies: children produce
catch-mouse and catcher-mouse but not mouse-catch for mouse-catcher (Clark et al. 1986, Nagpal and
Nicoladis 2009, Gamache and Schmitt to appear). As I have argued in previous research, this
pattern of forms reflects children building a particular derivation of verbal compounds,
repeated from (3a) in (16) below.
(16)
n
n
mouse

n
n

v
catch

v
n
er

v
catch

n
mouse

According to this derivation, children should, and do, first produce catch-mouse then catchermouse before producing the adult-like mouse-catcher. Because catch can only move after –er has
merged into the structure, children should not, and are not attested to, produce mouse-catch.
The claim that children’s non-adult forms are a reflection of their building of the
adult structure makes a particular prediction for how these children should interpret
compounds. The first step in the derivation merges the verb with the internal argument,
which means that as soon as children have acquired this step in the derivation, they should
only allow for verb-internal argument interpretations of verbal compounds. This prediction
was tested in the next study.

126

6.2 Child Study 5: Preferred Interpretations of Verbal Compounds
The final study examined how children interpret verbal compounds. The ultimate
goal of this study, in conjunction with the previous study, was to determine if children’s
interpretation of verbal compounds is dependent on their knowledge of verbal compound
syntax. An immediate goal of this study was to provide evidence that children do not start
off with a bias towards the adult interpretation of verbal compounds. That is, the goal was to
show that, at some point in a child’s language development, a towel-cleaner is equally likely to
be a person who cleans towels (internal argument reading of towel) or a person who cleans
things using a towel (adjunct reading of towel).
This study was designed to examine these claims by pairing nouns and verbs together
where the noun could easily serve as either the internal argument of the verb or an adjunct
to the verb. For example, clean and towel were paired since a towel is both something that can
be cleaned and can be used to clean. In the task the child had to choose which of two people
was a towel-cleaner, a person who cleans towels (internal argument) or a person who cleans
things using a towel (adjunct). Additionally, as in Adult Study 2, telicity was manipulated in
the items to determine if children were sensitive to the differences in argument structure of
atelic and telic verbs.
The hypotheses put forth in this dissertation made particular predictions about
children’s behavior on this task. Children who had not yet learned that the modifier noun in
a verbal compound is the internal argument of the verb should have had no preference
between an internal argument interpretation and an adjunct reading for verbal compounds.
In other words, these children should have been equally likely to choose an internal
argument reading or an adjunct reading of towel-cleaner. Children who had already learned that
the modifier noun in a verbal compound is the internal argument of the verb should have
127

only preferred internal argument readings of verbal compounds. With respect to telicity,
children were predicted, like adults (Borer 1994), to allow more adjunct interpretations for
atelic verbs and less adjunct interpretations for telic and particle verbs.
6.2.1 Method
6.2.1.1 Participants
31 children (18F, 13M) and 10 adults (8F, 2M) participated in the study. 15 children
made up the Younger Children group (3;0 – 5;0, Mean 4;2) and 16 children the Older
Children group (5;7 – 6;7, Mean 6;2). Children were recruited from local schools in Western
Massachusetts and adults from an undergraduate linguistics course at Michigan State
University. The adult participants received extra credit in their course for their participation.
6.2.1.2 Materials and Procedure
Four verbs were chosen for each of three verb types (atelic, telic, and particle). Each
verb was then paired with two different nouns, for a total of 24 items. Nouns were chosen
for each verb that could reasonably serve as either the internal argument of the verb or an
adjunct, usually an instrument, to the verb. See Table 26 for the full list of items.

128

Table 26. VC Interpretation: Items
Verb
Type

Verb

Compound

carry

helicopter-carrier
airplane-carrier
bird-tickler
penguin-tickler
sword-protector
shield-protector
truck-puller
tractor-puller
pencil-drawer
crayon-drawer
mouth-painter
foot-painter
block-maker
box-maker
sponge-cleaner
towel-cleaner
stick-picker-upper
fork-picker-upper
spoon-passer-outter
cup-passer-outter
bowl-filler-upper
bucket-filler-upper
basket-putter-awayer
bag-putter-awayer

tickle
Atelic

protect
pull
draw
paint

Telic

make
clean
pick up
pass out

Particle

fill up
put away

Argument
Description
carries helicopters
carries airplanes
tickles birds
tickles penguins
protects swords
protects shields
pulls trucks
pulls tractors
draws pencils
draws crayons
paints mouths
paints feet
makes blocks
makes boxes
cleans sponges
cleans towels
picks up sticks
picks up forks
passes out spoons
passes out cups
fills up bowls
fills up buckets
puts away baskets
puts away bags

129

Adjunct Description
carries things using a helicopter
carries things using an airplane
tickles things using a bird
tickles things using a penguin
protects things using a sword
protects things using a shield
pulls things using a truck
pulls things using a tractor
draws things using a pencil
draws things using a crayon
paints things using their mouth
paints things using their foot
makes things using blocks
makes things using boxes
cleans things using a sponge
cleans things using a towel
picks up things using a stick
picks up things using a fork
passes out things using a spoon
passes out things using a cup
fills up things using a bowl
fills up things using a bucket
puts away things using a basket
puts away things using a bag

To begin the task, the child was introduced to Monkey, a curious animal who liked to
learn things only from children. The following instructions were read to the child:
(17) Verbal Compound Interpretation Instructions
“This is my friend Monkey. There’s something that you need to know about this
monkey. She’s a funny monkey who only likes listening to children. She has a
problem, though, because she knows some names for people, but she’s not quite
sure who the names belong to. Do you think you can help her? She is going to tell
us about two people and then give us a name and we have to figure out which
person it goes with. Does that sound okay?”
On each trial two people were introduced, one at a time, with an image and an audio
recording describing the person in the pictures. On the third slide, both images appeared
side by side and the child was asked to determine which person should have a particular
name. See Figure 31 for a sample item. Children responded by pointing to the person on the
screen that they believed should have the name under question.
Figure 31. VC Interpretation: Sample Item

Slide 1

Slide 2

Slide 3

“This is a person who

“And this is a person who

“Which one is the sword-

protects swords.”

protects things using a

protector?”

sword.
The general procedure for this task was the same as for the verbal compound elicited
production task. If children participated in both the verbal compound elicited production
task and the verbal compound preference task, the preference task was administered on a

130

second day of testing. All adults participated in both the elicited production task and the
preference task in the same testing session.
6.2.1.3 Analysis
The data were analyzed in two ways. First, a 3 Verb Type (Atelic, Telic, Particle) X 3
Age Group (Younger Children, Older Children, Adults) ANOVAs was carried out with
proportion of argument responses as the dependent measure to determine if there are ageor verb type-related differences in response patterns.
Next, the results of the elicited production study were compared to the results of the
preference task for those children were participated in both. Children were divided into
groups based on the types of forms that they were producing in the elicited production task.
A 5 Production Group ANOVA was carried out on the proportion of consistent head
nouns with post-hoc comparisons to examine the differences in the productions groups. A
Bonferroni correction was used on the post-hoc comparisons. Finally, one-sample t-tests
were used to examine each group’s performance on the preference task against chance.
6.2.2 Results
6.2.2.1 Preference Task Analysis
Figure 32 shows the proportion of argument responses for each participant in the
study by age group, sorted within each group by the age of the participants. The horizontal
grey bar shows the range from .4 to .6 argument responses, which is the chance range for
this forced-choice task. There are a few patterns to take note of. First, no adult participants
fell in this region; all adult participants were well above chance for selecting argument
interpretations of verbal compounds. Second, both of the child groups had participants in
this range, eight in the Younger Children group and three in the Older Children group.
Finally, no participants fell below the chance region.
131

Figure 32. VC Interpretation: Proportion of Argument Responses by Participant

Proportion of Argument
Responses

1.0
.8
.6
.4
.2
Younger Children

Adults

Older Children

.0

Figure 33 shows the proportion of argument responses by each age group in the
verbal compound interpretation task.
Figure 33. VC Interpretation: Proportion of Argument Responses

Proportion of Argument
Responses

1.0

Younger Children

.5

Older Children
Adults

.0
Atelic

Telic

Particle

The results of the 3 Verb Type X 3 Age Group ANOVA revealed a significant main
effect of Verb Type (F(2, 76) = 3.496, p = .035). Post-hoc comparisons revealed that Particle
Verbs received marginally more Argument responses than Telic Verbs (p = .086). There was
132

no difference between Atelic Verbs and Telic Verbs (p = .158) or between Atelic Verbs and
Particle Verbs (p = 1.000).
The main effect of Age Group was also significant (F(2, 38) = 7.634, p = .002). Posthoc comparisons revealed that Adults used significantly more argument responses than
Younger Children (p = .001). There were no differences between Younger Children and
Older Children (p = .140) or Older Children and Adults (p = .124).
The interaction between Verb Type and Age Group was not significant (p = .827).
6.2.2.2 Summary of Results
The results provide evidence that children do not start out with a preference for
verb-internal argument readings of verbal compounds, which suggests that children do not
have a bias towards argument readings in advance of their syntactic knowledge constraining
their interpretation. By age five or six, however, this preference for verb-internal argument
readings has clearly developed. Replicating the findings from Adult Study 2, adults in the
current study had a strong preference for verb-internal argument readings only. The effect of
telicity, however, failed to replicate in this study. The lower number of nouns per verb type
group (four, as opposed to eight in Adult Study 2) could have contributed to the lack of this
finding, in addition to changes to the methodology to make the study child-appropriate
(picturable verb-internal argument/adjunct combinations). See Appendix D for results by
individual verb.
6.2.3 The Interaction of Syntax and Interpretation
To examine the predicted interaction between children’s acquisition of verbal
compound syntax and its effect on children’s interpretation of verbal compounds, the
children who participated in both the verbal compound elicited production task and the

133

verbal compound preference task (N = 28) were divided into groups based on their elicited
production task performance.
Categorizing children was difficult for two reasons. First, children did not necessarily
use the same verbal compound form across all of the items when they used verbal
compound forms. For example, a child may have produced both VO and VerO forms. Only
one child produced all three forms (VO, VerO, and OVer), while the others who produced
more than one form produced either VO and VerO or VerO and OVer, suggesting that
these children were still learning the features in the next step of the derivation. The second
issue was that children, like adults, did not always use verbal compound forms. For example,
a child may have used a mix of verbal compound forms and noun-noun compound forms.
For these reasons, children had to reach a threshold of 50% of the form under consideration
and any more advanced form in the derivation. For example, a child who produced 25%
VerO and 25% OVer would be considered to be a VerO-producer, since at least 50% of their
trials showed them to have knowledge of at least the VerO step. Similarly, a child who
produced 25% VO and 25% VerO was considered to be a VO-producer.
The resulting grouping was the following. Children who produced more than 50%
OVer forms were in the OVer-producer group. Children who produced more than 50%
combined OVer and VerO forms were in the VerO-producer group. Children who produced
more than 50% combined OVer, VerO, and VO were in the VO-producer group. Children
who failed to produce these forms on a total of 50% of the trials were in the non-verbal
compound-producer group. Table 27 shows the distribution of children by group.

134

Table 27. Groups based on Elicited Production Task Performance
Group
Non-VC-producer
VO-producer
VerO-produce
OVer-produce
Adult

N
4
3
11
10
10

Mean Age
4;5
4;8
5;4
5;11
---

Production Performance
< 50% combined VO, VerO, OVer forms
≥ 50% combined VO, VerO, OVer forms
≥ 50% combined VerO and OVer forms
≥ 50% combined OVer forms
---

To assess how children’s knowledge of verbal compound syntax affected their
interpretation of verbal compounds, two analyses were performed. First, a one-way
Production Group (Non-VC-producer, VO-producer, VerO-producer, OVer-producer,
Adult) ANOVA was performed on the proportion of argument responses for each
production group to determine at which stage in development children’s preference
increased towards only argument interpretations. Second, one-sample t-tests were performed
on the proportion of argument responses for each production group against chance (.5) to
determine at which stage children reliably chose above chance argument interpretations.
See Figure 34 for the average proportion of argument responses by elicited
production group.

135

Figure 34. VC Interpretation: Proportion of Argument Responses by Production Group

Proportion of Argument
Responses

1.0

.5

.0
Non-VC- VOVerOOVerProducer Producer Producer Producer

Adult

The one-way Age Group ANOVA on the proportion of argument responses
revealed a main effect of Production Group (F(4, 33) = 10.472, p < .001). Post-hoc
comparisons revealed that Adults responded with significantly more argument responses
than VO-producers (p = .005) and Non-VC-producers (p < .001). OVer-producers
responded with significantly more argument responses than Non-VC-producers (p = .001)
and marginally more argument responses than VO-producers (p = .074). VerO-producers
responded with significantly more argument responses than Non-VC-producers (p = .002).
There were no other significant differences between groups.
The results of the second analysis, comparing the groups to chance, reflected the
findings of the first analysis. While VerO-Producers (t(10) = 6.540, p < .001), OVerProducers (t(9) = 7.461, p < .001), and Adults (t(10) = 18.822, p < .001) were all significantly
above chance, Non-VC-Producers (p = .718) and VO-Producers (p = .423) were at chance.
In summary, children’s performance on the preference task, matched their predicted
performance based on the production task. Children who reliably produced 50% or more
OVer and VerO forms performed adult-like on the preference task, while children who
136

produced 50% or more non-verbal compound forms did not. The VO-producers, while not
significantly above chance, do trend in the predicted direction and would be expected to
differ from chance given a large sample size within this group.
6.3 General Discussion
The goals of the studies in this chapter were (i) to replicate previous elicited verbal
compound production studies and (ii) to determine if children’s syntactic knowledge predicts
their interpretation of verbal compounds. In the first half of this dissertation we showed that
the syntax of verbal compounds alone restricts their interpretations for adults. The burden
of the studies in this chapter was to show there was a correlation between the way that
children produce and interpret verbal compounds. Namely the correlation that we were
looking for was between children’s knowledge of steps in the derivation and the reflection of
those steps in how children constrain the interpretation of verbal compounds.
The results of the two studies supported both guiding hypotheses of this
dissertation. First, the verbal compound forms that children produced were the same as
those that have been attested in previous production studies (Clark et al. 1986, Nagpal and
Nicoladis 2009, Gamache and Schmitt to appear). Children produced VO catch-mouse, VerO
catcher-mouse, and OVer mouse-catcher, but, again, failed to produce OV mouse-catch. As in
previous work, I argue that these forms reflect steps in the adult derivation (Gamache and
Schmitt to appear).
The fact that children are building the adult structure led to the prediction that
children who have sufficient verbal compound syntax knowledge should have adult-like
interpretations of verbal compounds. Sufficient verbal compound syntax knowledge is
defined as having learned the step in the derivation where the modifier noun merges as the
internal argument of the verb, which is the first step in the derivation. The results of the
137

comparison between children’s syntactic knowledge and their interpretation of verbal
compound showed that children with little or no knowledge of verbal compound syntax
allowed adjunct interpretations that are not allowed by adult verbal compound syntax, while
children with adult-like or near-adult-like knowledge of verbal compound syntax only
allowed internal argument readings of verbal compounds. The overall finding lends support
to the hypothesis that children are building steps in the adult derivation, since their syntactic
knowledge bore directly on their interpretation knowledge, and also to the hypothesis that
children use the same compositional system as adults to interpret compounds.

138

CHAPTER 7: SUMMARY AND CONCLUSIONS
7.1 Goal of Dissertation
This dissertation had two main goals: (i) to define the system that adults use to
interpret compounds and (ii) to understand how child come to use this system.
The first half of this dissertation presented the results of two studies examining
adults’ interpretation of noun-noun compounds and verbal compounds. Adults’
interpretation of compounds was predicted to be guided by (i) the syntactic structures of
compounds and (ii) the lexical structure of the words in the compounds and tendencies of
different natural classes to behave in particular ways. With respect to syntax, verbal
compound syntax was predicted to constrain adults’ interpretation of noun-verb-er strings to
a reading where the noun is interpreted as the internal argument of the verb (Roeper and
Siegel 1978, Selkirk 1982, Lieber 1983, Di Sciullo and Williams 1987, Bobaljik 2003). For
example, a cucumber-holder was predicted to be interpreted as only a person who holds
cucumbers and not a person who holds things using a cucumber. Noun-noun compound
syntax was predicted to not constrain adults’ interpretations in any particular way, except by
determining which of the two elements is the head. Therefore, cucumber-basket was predicted
to lead to many interpretations. With respect to lexical structure, adults were predicted to
interpret compounds in line with the tendencies of the natural classes of their constituents
(Downing 1977, Leonard 1985, Štekauer 2005). For example, a cucumber-basket would most
likely to lead to a reading of “a basket that cucumbers are put in” because basket is an artifact
and artifacts are most often defined in terms of their function (to hold/carry).
The first study explored how syntactic structure (verbal compound, noun-noun
compound) and natural classes (animate natural kinds, inanimate natural kinds,
locative/temporal nouns, humans, artifacts) interacted in adults interpretation of
139

compounds. The findings supported a theory of compound interpretation that requires
syntax and structured lexical entries. First, there were striking differences between nounnoun strings, which were presumably parsed with noun-noun compound syntax, and nounverb-er strings, which were presumably parsed with verbal compound syntax. Noun-noun
strings were interpreted using the full range of available relationships (agentive, constitutive,
formal, telic), while noun-verb-er strings were mainly interpreted with telic, verb-internal
argument relationships. This pattern held for known and nonce head nouns in noun-noun
strings and for known and nonce verbs in noun-verb-er strings, suggesting that this pattern
was arising from the syntax, not lexical-item specific knowledge. Second, there were
systematic differences in the way that noun-noun strings were interpreted based on their
natural classes. Noun-noun strings with an artifact as the head noun were often interpreted
as telic, while noun-noun strings with animate natural kind head nouns were almost never
interpreted as telic. For example, paper-scissors was likely to be interpreted as scissors used to
cut paper, while paper-salamander was unlikely to be interpreted with a functional connection
between paper and salamander. These findings reflect previous findings by Downing (1977),
Leonard (1985), and Štekauer (2005).
The second adult study examined adults’ ability to assign either licit (internal
argument) or illicit (adjunct) interpretations to verbal compounds. The results showed that
adults overwhelmingly prefer internal argument readings of verbal compounds. Participants
only allowed non-verb-internal argument readings when the verb did not require an object.
Atelic verbs, which may lack lexically-realized internal arguments (Borer 1994), were more
likely to lead to adjunct interpretations, such as a person who cleans using towels for towelcleaner, than telic verbs and particle verbs, which both require lexically-realized internal
arguments. For example, a basket-carrier (atelic) was more likely than a towel-cleaner (telic) to
140

allow an adjunct reading where basket was the instrument that did the carrying, rather than
was the object that was carried.
Taken together, the results of the adult studies provide evidence that the
compositional system that adults use to interpret compounds includes: (i) the syntax of
compounds, (ii) the relationships made available by the lexical structure of the head noun,
and (iii) the most salient lexical properties of particular nouns associated to particular natural
classes.
The second half of this dissertation presented the results of five studies that
examined children’s production and interpretation of noun-noun compounds and verbal
compounds. Children were predicted to behave exactly like adults, except in cases where
systematic differences in syntactic or lexical semantic representations caused children to have
different interpretations. The first series of studies, exploring children’s production and
interpretation of noun-noun compounds, examined how children use lexical semantic
information in compounds. The second series of studies, exploring children’s production
and comprehension of verbal compounds, examined how children use syntactic information
in compounds.
The first noun-noun compound study, a corpus study, compared children’s
production of novel noun-noun compounds to these compounds in their input based on the
natural classes of the modifiers and heads and the relationship between them. The goal of
this study was to provide evidence that children use the tendencies of natural classes, rather
than specific items in the input, to produce novel compounds. The results showed just that.
Children were matched to the input in terms of the distribution of relationships used at the
natural class level, but not at the individual items level. Children also produced compounds
composed of nouns that were not in their compound input, suggesting that templates are
141

not necessary for productive compound production. The findings of this study run counter
to a claim that children use head-relationship templates for novel compounds (Krott et al.
2009).
The second noun-noun compound study, an elicited production task, examined
children’s production of novel compounds based on the relationships between the head and
modifier. The goal of this study was to provide evidence that children produce noun-noun
compounds for the same range of meanings as adults. The results showed that there were
some relationships that children were more likely to produce noun-noun compounds for
than others. For example, children were very likely to produce noun-noun compounds for
shape, formal relationships, such as a hand-chair for a chair that is shaped like a hand, but
much less likely to produce noun-noun compounds for creation, agentive compounds, such
as a rain-puddle for a puddle created by rain. However, this pattern also held for adults. While
adults produced more noun-noun compounds overall than children, the relationships for
which they were less likely to produce noun-noun compounds patterned with the children.
The third, and final, noun-noun compound study, a forced-choice interpretation
task, examined children’s interpretation of noun-noun compounds given (i) the natural
classes of the head and modifier and (ii) the permanence of the relationship between the
head and modifier. The goal of this study was to provide evidence that children interpret
compounds according to the tendencies of the natural classes of their constituents in the
same way that adults do. The results showed that children had the same the interpretation as
adults for each of the item types. Children were able to taking into account both natural class
information and permanence information to arrive at different types of relationships for
compounds that shared the same head noun. The findings ran counter to Krott et al.’s
(2009) findings that children base their interpretations on previous head-relationship pairings
142

from the input; given that children were able to give adult-like interpretations for
compounds with the same head noun, but different relationships, they must be using a more
complex system than templates for interpretation.
To summarize the findings of the noun-noun compound studies, children were
adult-like by all measures. Unlike proposals from previous authors (Krott et al. 2009),
children did not rely on item-specific information to either produce or interpret novel nounnoun compounds. Moreover, children produced novel noun-noun compounds for the same
range of relationships as adults, contra Hiramatsu et al. (2000).
Turning to the verbal compound studies, the first study, an elicited production task,
examined children’s knowledge of the verbal compound syntactic structure. The goal of this
study was to create a baseline of syntactic knowledge for each child. The results replicated
previous findings showing that children only produce a particular set of non-adult forms,
clean-towel and cleaner-towel, but not towel-clean for towel-cleaner. As in previous research
(Gamache and Schmitt to appear), I argued that these forms are reflective of steps in the
adult derivation, and therefore carry with them all of the same syntactic constraints as the
adult derivation.
The second verbal compound study, a forced-choice interpretation task, examined
children interpretation of verbal compounds. The goal of the study was to provide evidence
that children’s interpretation of verbal compounds is determined by their knowledge of
verbal compound syntax, as measured by their productions in the previous study. The results
showed a coupling between syntactic knowledge and interpretation; children who were not
able to produce any of the verbal compound structure freely allowed either internal
argument or adjunct readings of the modifier in a verbal compound. On the other hand,
children who showed knowledge of the verb and internal argument merging in their verbal
143

compound productions only allowed adult, internal argument interpretations of verbal
compounds.
To summarize the findings of the verbal compound studies, children interpreted
verbal compounds in an adult-like way insofar as their syntactic representations were adultlike. If children’s production data showed that they had learned a particular step in the
syntactic derivation, their interpretation reflected the constraints that had been built up to
the step.
7.2 Conclusions
The findings of this dissertation have implications for both theoretical linguistics and
language acquisition research. First, to speak to the theoretical implications, one finding of
this dissertation is that speakers can arrive on common interpretations of compounds, even
those that lack constraint from the syntax, in the absence of context. This runs counter to
the claim that compound meanings are constructed completely based on context (e.g.,
Marchand 1966). This dissertation has demonstrated that there is sufficient information in
the linguistic signal to find the most likely meaning of a compound, and sufficient flexibility
in the structure of lexical items to account for the fact that a single compound can have
multiple meanings.
Another finding relevant to both theoretical and child language research is the
evidence from the child production data for the structure of verbal compounds. Contra
previous proposals (Harley 2009, Borer 2012), the children’s productions suggest that the
verbal compound derivation proceeds as (i) merge verb and internal argument, (ii) merge –er
in the structure, (iii) incorporate the verb into –er and (iv) move the internal argument to a
specifier position above verb-er. This finding has implications for the field that go beyond
the structure of verbal compounds. Children were shown to adhere to the principles of the
144

adult compound structures both in production and in interpretation, even before they have
acquired the full derivation. This finding suggests that the strong continuity hypothesis in
correct, at least in the domain of compounds, and data of this type can be treated more
seriously as evidence for features of adult structures.
Finally, the finding that children use the same source of information for both their
production and interpretation of compounds throughout development without recourse to
item-specific information suggests that we need to reconsider theories built on the notion
that children are completely dependent on the input. The finding that children are not
dependent on the input, even in the interpretation of noun-noun compounds, which are
extremely impoverished syntactically, suggests that in cases of rich syntactic structure,
children do not rely on item-specific information.

145

APPENDICES

146

APPENDIX A. Full Compound List for Adult Study 1

Artifact Modifier

Animate Modifer

Table 28. Full Compound List for Adult Study 1 by Head and Modifier Category
Animate
Head
bunny-dolphin
dolphin-bunny
dolphin-lobster
dolphin-mouse
horse-shark
lobster-dolphin
mouse-dolphin
pig-shark
shark-horse
shark-pig
shark-sheep
sheep-shark
bowl-bunny
bowl-deer
bowl-mouse
box-panda
chair-elephant
chair-horse
chair-pig
coat-panda
oven-panda
spoon-bird
telephone-bird
vacuum-bird

Artifact
Head
bird-spoon
bird-telephone
bird-vacuum
bunny-bowl
deer-bowl
elephant-chair
horse-chair
mouse-bowl
panda-box
panda-coat
panda-oven
pig-chair
box-key
bus-cup
bus-spoon
bus-vacuum
cup-bus
desk-key
key-box
key-desk
key-oven
oven-key
spoon-bus
vacuum-bus

Human
Head
bee-criminal
bee-man
bee-woman
bunny-lady
cat-author
deer-lady
elephant-author
hippo-lady
pig-author
tiger-dentist
tiger-musician
tiger-toddler
couch-criminal
couch-mother
couch-woman
crib-boy
cup-pilot
desk-boy
oven-boy
radio-dentist
radio-lawyer
radio-musician
soap-pilot
spoon-pilot

Inanimate
Head
camel-mud
cat-lettuce
chicken-banana
chicken-cherry
chicken-milk
cow-lettuce
deer-mud
eagle-lava
eagle-sand
eagle-snow
elephant-lettuce
hippo-mud
airplane-rain
bed-lava
bed-lemon
bed-snow
bicycle-peach
crib-peach
cup-rain
desk-peach
shoe-milk
shoe-water
shoe-banana
soap-rain

Temporal
Nonce
Head
Head
camel-month camel-dort
cat-mountain cow-quait
cow-mountain frog-bretch
giraffe-beach
frog-hade
giraffe-bedroom frog-zimit
giraffe-zoo
goat-quait
goat-mountain lion-dresh
hippo-month lion-garve
puppy-breakfast lion-gick
puppy-dinner owl-quait
puppy-year
snake-dort
snake-month
turtle-dort
airplane-evening airplane-dunt
bicycle-castle
bicycle-glesh
blanket-castle blanket-glesh
bottle-beach
broom-glesh
bottle-lunch
clock-dunt
bottle-zoo
comb-garve
clock-evening comb-gick
crib-castle
comb-mook
soap-evening diaper-bretch
truck-breakfast diaper-hade
truck-planet
diaper-masp
truck-year
door-dunt

147

Nonce-er
Verb-er
Head
Head
dog-bemainer fish-builder
dog-jender
fish-reader
dog-lorcer
fish-tickler
goat-fisser
horse-toucher
lobster-lalker
lobster-eater
monkey-cheamer mouse-eater
monkey-fleacher owl-toucher
monkey-ploser sheep-toucher
owl-fisser
snail-cleaner
sheep-fisser
snail-opener
snake-lalker
snail-thrower
turtle-lalker
turtle-eater
blanket-mielder box-driver
broom-mielder broom-driver
clock-cherver car-baker
coat-mielder
car-builder
door-cherver
car-reader
shirt-jender
coat-driver
shirt-lorcer
door-lover
shirt-malker
refrigerator-cleaner
table-cheamer refrigerator-feeder
table-ploser
refrigeratortable-praster
thrower
telephonetelephone-lover
cherver
vacuum-lover

Inanimate Modifier

Human Modifier

Table 28 (con’t)
Animate
Head
author-cat
author-elephant
author-pig
criminal-bee
dentist-tiger
lady-bunny
lady-deer
lady-hippo
man-bee
musician-tiger
toddler-tiger
woman-bee

Artifact
Head
boy-crib
boy-desk
boy-oven
criminal-couch
dentist-radio
lawyer-radio
mother-couch
musician-radio
pilot-cup
pilot-soap
pilot-spoon
woman-couch

Human
Head
athlete-butcher
athlete-criminal
athlete-mother
butcher-athlete
criminal-athlete
judge-poet
lawyer-poet
mother-athlete
musician-poet
poet-judge
poet-lawyer
poet-musician

Inanimate
Head
butcher-apple
coach-apple
doctor-banana
doctor-sun
doctor-water
judge-wind
lawyer-wind
mother-apple
teacher-broccoli
teacher-lemon
teacher-snow
uncle-wind

banana-chicken
cherry-chicken
lava-eagle
lettuce-cat
lettuce-cow
lettuce-elephant
milk-chicken
mud-camel
mud-deer
mud-hippo
sand-eagle
snow-eagle

banana-shoe
lava-bed
lemon-bed
milk-shoe
peach-bicycle
peach-crib
peach-desk
rain-airplane
rain-cup
rain-soap
snow-bed
water-shoe

apple-butcher
apple-coach
apple-mother
banana-doctor
broccoli-teacher
lemon-teacher
snow-teacher
sun-doctor
water-doctor
wind-judge
wind-lawyer
wind-uncle

broccoli-dirt
carrot-rock
carrot-sun
carrot-water
dirt-broccoli
dirt-lemon
dirt-raisin
lemon-dirt
raisin-dirt
rock-carrot
sun-carrot
water-carrot

Temporal
Head
aunt-day
butcher-forest
child-beach
child-circus
child-lunch
coach-forest
judge-day
nurse-breakfast
nurse-planet
nurse-school
sister-forest
uncle-day

Nonce
Nonce-er
Verb-er
Head
Head
Head
accountant-bley accountant- accountant-biter
aunt-bley
pliver
dentist-biter
coach-blump aunt-pliver
man-finder
girl-cload
baby-jender soldier-baker
girl-garve
baby-malker soldier-reader
girl-mook
baby-youcher soldier-rinser
politicianchef-cheamer son-cleaner
bretch
chef-ninker
son-feeder
politician-masp chef-praster son-puller
politician-zill man-heigner teenager-finder
sister-blump sister-heigner toddler-biter
teenager-blump teenagerwoman-finder
uncle-bley
heigner
toddler-pliver
broccoli-party cloud-cload
cherry-zainer cherry-juggler
cucumber-house cloud-mook
cucumberlava-kicker
egg-planet
cloud-zesh
zainer
milk-juggler
egg-school
cucumberflower-mainter mist-kicker
egg-store
blosh
flower-malker potato-baker
fire-circus
grape-hink
flower-youcher potato-drinker
fire-lunch
mist-hink
grape-shrapper potato-rinser
fire-morning
pea-brist
mist-shrapper sand-kicker
grape-party
pea-masp
sand-shrapper stone-juggler
raisin-party
pea-zill
stone-zainer tree-feeder
rock-house
raisin-hink
tomato-ninker tree-puller
sun-house
rock-blosh
tomato-praster tree-pusher
stone-blosh
tomato-yolder

148

Nonce Modifier

Locative/Temporal Modifer

Table 28 (con’t)
Animate
Artifact
Human
Head
Head
Head
beach-giraffe
beach-bottle
beach-child
bedroom-giraffe breakfast-truck breakfast-nurse
breakfast-puppy castle-bicycle
circus-child
dinner-puppy castle-blanket day-aunt
month-camel castle-crib
day-judge
month-hippo evening-airplane day-uncle
month-snake
evening-clock forest-butcher
mountain-cat evening-soap
forest-coach
mountain-cow lunch-bottle
forest-sister
mountain-goat planet-truck
lunch-child
year-puppy
year-truck
planet-nurse
zoo-giraffe
zoo-bottle
school-nurse

bretch-frog
dort-camel
dort-snake
dort-turtle
dresh-lion
garve-lion
gick-lion
hade-frog
quait-cow
quait-goat
quait-owl
zimit-frog

bretch-diaper
dunt-airplane
dunt-clock
dunt-door
garve-comb
gick-comb
glesh-bicycle
glesh-blanket
glesh-broom
hade-diaper
masp-diaper
mook-comb

Inanimate
Temporal
Head
Head
circus-fire
bathroomhouse-cucumber winter
house-rock
circus-winter
house-sun
cottage-sky
lunch-fire
morning-winter
morning-fire
school-sky
party-broccoli sky-cottage
party-grape
sky-school
party-raisin
sky-store
planet-egg
store-sky
school-egg
winterstore-egg
bathroom
winter-circus
winter-morning

bley-accountant blosh-cucumber
bley-aunt
blosh-rock
bley-uncle
blosh-stone
blump-coach
brist-pea
blump-sister
cload-cloud
blump-teenager hink-grape
bretch-politician hink-mist
cload-girl
hink-raisin
garve-girl
masp-pea
masp-politician mook-cloud
mook-girl
zesh-cloud
zill-politician
zill-pea

brind-afternoon
brind-cottage
brind-store
brist-city
cload-weekend
drign-autumn
drign-bathroom
drign-morning
drunt-city
hote-weekend
zesh-weekend
zill-city

149

Nonce
Head
afternoon-brind
autumn-drign
bathroom-drign
city-brist
city-drunt
city-zill
cottage-brind
morning-drign
store-brind
weekend-cload
weekend-hote
weekend-zesh

brist-rond
drunt-rond
hote-pame
pame-hote
pame-quelt
pame-zesh
quelt-pame
rond-brist
rond-drunt
rond-weach
weach-rond
zesh-pame

Nonce-er
Verb-er
Head
Head
afternoonautumn-hugger
fanker
afternoon-helper
autumn-mouter bedroom-hugger
bathroomdinner-helper
mouter
night-hitter
bedroomnight-puller
mouter
night-pusher
cottage-fanker ocean-drinker
dinner-fanker ocean-kisser
kitchen-ninker ocean-rinser
kitchen-troker year-helper
kitchen-yolder zoo-hugger
summer-larter
summer-mainter
summer-youcher
dresh-trider
dresh-catcher
drunt-flader
faise-hitter
hote-trider
faise-pusher
klend-jeeder
faise-watcher
klend-larter
gick-catcher
klend-mainter hade-rider
luse-drosher
quelt-catcher
luse-troker
weach-rider
luse-yolder
zimit-rider
quelt-trider
zove-climber
weach-flader
zove-drinker
zimit-flader
zove-kisser

Verb-er Modifier

Nonce-er Modifier

Table 28 (con’t)
Animate
Head
bemainer-dog
cheamermonkey
fisser-goat
fisser-owl
fisser-sheep
fleachermonkey
jender-dog
lalker-lobster
lalker-snake
lalker-turtle
lorcer-dog
ploser-monkey
builder-fish
cleaner-snail
eater-lobster
eater-mouse
eater-turtle
opener-snail
reader-fish
thrower-snail
tickler-fish
toucher-horse
toucher-owl
toucher-sheep

Artifact
Head
cheamer-table
cherver-clock
cherver-door
cherver-telephone
jender-shirt
lorcer-shirt
malker-shirt
mielder-blanket
mielder-broom
mielder-coat
ploser-table
praster-table

Human
Head
cheamer-chef
heigner-man
heigner-sister
heigner-teenager
jender-baby
malker-baby
ninker-chef
pliveraccountant
pliver-aunt
pliver-toddler
praster-chef
youcher-baby

baker-car
baker-soldier
builder-car
biter-accountant
cleaner-refrigerator biter-dentist
driver-box
biter-toddler
driver-broom
cleaner-son
driver-coat
feeder-son
feeder-refrigerator finder-man
lover-door
finder-teenager
lover-telephone
finder-woman
lover-vacuum
puller-son
reader-car
reader-soldier
thrower-refrigerator rinser-soldier

Inanimate
Temporal
Nonce
Nonce-er
Verb-er
Head
Head
Head
Head
Head
mainter-flower fanker-afternoon drosher-luse bemainer-shraller bemainermalker-flower fanker-cottage flader-drunt drosher-farker
painter
ninker-tomato fanker-dinner flader-weach farker-drosher brenser-climber
praster-tomato larter-summer flader-zimit farker-fleacher brenser-kisser
shrapper-grape mainter-summer jeeder-klend farker-troker
brenser-tickler
shrapper-mist mouter-autumn larter-klend fleacher-farker drosher-player
shrapper-sand moutermainter-klend jeeder-shraller fesker-hitter
yolder-tomato bathroom
trider-dresh larter-shraller
fesker-opener
youcher-flower mouter-bedroom trider-hote shraller-bemainer fesker-watcher
zainer-cherry ninker-kitchen trider-quelt shraller-jeeder fleacher-player
zainertroker-kitchen troker-luse shraller-larter
jeeder-painter
cucumber yolder-kitchen yolder-luse troker-farker
lorcer-painter
zainer-stone youcher-summer
ploser-player
baker-potato
drinker-potato
feeder-tree
juggler-cherry
juggler-milk
juggler-stone
kicker-lava
kicker-mist
kicker-sand
puller-tree
pusher-tree
rinser-potato

150

drinker-ocean catcher-dresh
helper-afternoon catcher-gick
helper-dinner
catcher-quelt
helper-year
climber-zove
hitter-night
drinker-zove
hugger-autumn hitter-faise
hugger-bedroom kisser-zove
hugger-zoo
pusher-faise
kisser-ocean
rider-hade
puller-night
rider-weach
pusher-night
rider-zimit
rinser-ocean
watcher-faise

climber-brenser
hitter-fesker
kisser-brenser
opener-fesker
painter-bemainer
painter-jeeder
painter-lorcer
player-drosher
player-fleacher
player-ploser
tickler-brenser
watcher-fesker

builder-washer
climber-washer
opener-wiper
thrower-wiper
tickler-washer
washer-builder
washer-climber
washer-tickler
watcher-wiper
wiper-opener
wiper-thrower
wiper-watcher

APPENDIX B. Full Compound List for Corpus Study
Table 29. Compound List for Corpus Study by Speaker
Adam
air-man
air-plane-man
air-plane-part
air-plane-port
alphabet-box
animal-book
animal-car
animal-duck
apartment-house
apple-juice
apple-snake
army-gun
automobile-car
baby-chicken
baby-dog
baby-dragon-fly
baby-horse
baby-jesus
baby-mail-bag
baby-soup
baby-tulip
baby-whale-fish
barn-tree
barrel-man
base-ball-ball
base-ball-hat
base-ball-statue
base-man
beach-ball
bee-count
bird-dinner
birdie-house
bird-nest
bird-pie
bird-seed
boot-snow
bowelmovement
box-thing
boy-bath
boy-bath-tub
boy-bed
boy-bird

boy-fish
bunny-day
bus-man
butter-birdie
butter-fly-bug
butter-flyprogram
button-hole
cage-house
cake-fork
camera-movie
car-mosquito
car-swing
car-truck
cat-cage
cat-car
cat-name
cereal-baby-soup
cheek-kiss
chick-egg
chicken-pox
children-shout
chocolate-syrup
choo-choo-train
choo-choo-traincow-boy-hat
christmas-tree
church-party
circus-animal
circus-dan
circus-elephant
circus-ring
clown-thing
cocker-shell
color-egg
comb-hair
cotton-ball
country-box
cow-boy-coffee
cow-boy-fireengine
cow-boy-holster
cow-boy-hoot

cow-boykangaroo
cow-boy-shirt
cow-boy-sock
cow-boy-train
crown-number
cry-baby
daddy-chicken
daddy-dragon
daddy-tiger
dance-time
Dan-circus
day-time-light
diesel-motor
dinosaur-baby
dirt-shovel
doctor-board
doctor-nothing
doctor-thing
dodge-thing
dog-baby
doggie-baby
doggie-house
dog-horse
dollar-book
door-apartmenthouse
dough-nut-pail
drain-hole
egg-noodle
egg-robin
elephant-foot
elephant-monkey
engineer-man
engine-truck
farm-house
fashion-car
fire-car
fire-chief
fire-chief-hat
fire-engine-boy
fire-hat
fish-flies
151

fish-fork
fluffer-nutterday
freight-fire
gas-thing
gentle-man
giraffe-mosquito
glove-hand
graham-cracker
grocery-storeman
guitar-drum
gun-pen
halloweenchurch-party
head-ache
high-school
horse-dress
house-circus
ice-cone
ice-cream-horn
ice-cube
ink-finger
jeep-man
jello-pie
jet-air-plane
juice-cranberry
key-lock
king-hat
kitty-box
kitty-cat
koala-bear
lady-bug
lady-dinosaur
lady-elephant
library-school
locomotivecaboose
log-car
macaroni-cheese
magic-boy
mail-paper
man-texaco-star

mantle-shelf
marble-gun
marble-race
medicine-box
middle-shoe
milk-box
milk-glass
milk-pond
milk-tire
Mommydinosaur
mommy-tiger
morning-time
mountain-sea
mouse-house
neck-lace
news-paper
numbers-bet
nurse-hat
office-train
pan-cake
paper-ball
paper-thing
parking-place
peanut_butterpickle
peanut-bug
peanut-butterman
peanut-butterpeanut
peanut-butterpickle-man
pencil-pen
pickle-pie
pie-boy
pie-turkey
pilot-thing
pin-cushion
pine-juice-apple
pirate-costume
play-check
play-ticket

Table 29 (con’t)
police-gun
police-song
polish-shoe
punch-man
puppy-dog
razor-toy
record-box
record-folder
rifle-gun
rubber-bandband-aid
rubber-bandman
sailor-man
salad-meat

sand-shovel
santa-claus
santa-claus-boy
school-church
sea-gull
ship-hat
snack-pop-corn
snow-boot
space-man
spices-cake
spider-book
spider-web
star-fish-animal
station-wagonwheel

steam-trailer
stick-boy
stick-whale
stone-bug
stone-fence
stream-line
sugar-snack
sweep-broom
tail-bird
tail-part
tattoo-man
taxi-cab
telephonenumber
telephone-pole

tennis-racket
tennis-thing
time-lunch
tire-clock
toe-finger
tomato-juice
tommy-spider
tooth-brush
totem-pole
toy-rhinoceros
toy-train
tractor-puzzle
train-car
train-man
trapeze-elephant

trapeze-lion
truck-game
truck-man
truck-track
truck-wheel
tuna-fish
tv-camera
vanilla-pie
water-fountainboy
water-queen
whale-fish
wind-mill-house
wind-shield-man
wood-friend

Adam Input
air-plane
air-plane-movie
air-port
alphabet-basket
animal-doctor
animal-man
animal-skin
apple-cider
army-man
art-gallery
arts-festival
baby-barn
baby-bear
baby-boy
baby-carriage
baby-chick
baby-cow
baby-doggie
baby-doll
baby-dragon
baby-duck
baby-elephant
baby-food
baby-giraffe
baby-lamb
baby-sitter
baby-whale

back-yard
bag-car
baggage-car
ball-player
banana-song
band-aid
band-box
bar-bell
bar-bell-set
barber-shop
barn-yard
base-ball
base-ball-bat
base-ball-player
basket-ball
bath-room
bath-tub
bean-bag
bean-bean
bean-stalk
beauty-salon
bee-hive
bird-carriage
bird-house
birth-day
birth-day-cake
birth-day-party

birth-day-time
boat-car
boat-pie
boat-ship
Bob-fish
book-shelf
bostonuniversity
box-car
boy-baby
boy-chair
boy-house
boy-scout
bread-box
breakfast-time
brief-case
brother-dinosaur
bubble-gum
buddy-cake
bull-dozer-song
bullet-gun
bunk-house
bunny-rabbit
butter-fly
cake-knife
candle-stick
candy-cane

car-train
cave-man
center-pole
chair-trick
chimney-house
chocolate-hair
chocolate-spray
Christmaspresent
Christmasvacation
church-hat
church-music
cigarette-smoke
circus-man
circus-parade
circus-people
circus-train
clothes-line
clothes-pin
coal-bin
cocoa-bean
cocoa-marsh
coffee-can
coffee-cup
coffee-pot
control-panel

cookie-dough
cottage-cheese
cough-drop
country-doctor
country-doctorbox
cow-boy
cow-boy-bill
cow-boy-book
cow-boy-boot
cow-boy-bunny
cow-boy-hat
cow-boy-lady
cow-boy-rope
cow-boy-shale
cow-girl
cow-horse
cracker-crumb
cranberry-juice
crayon-box
cream-pitcher
cub-scout
cup-cake
curtain-rod
dairy-farm
dairy-products
day-time

152

Table 29 (con’t)
diaper-man
dick-tracy-car
diesellocomotive
dinner-bell
dinner-time
doctor*dan
doctor-box
doctor-dan
doctor-kit
dodge-car
dog-cake
doggie-horse
dog-house
door-bell
dough-nut
dragon-fly
drum-stick
ear-ache
ear-phone
ear-phone-plug
easter-egg
egg-apple
egg-plant
egg-shape
extension-cord
eye-brow
face-cloth
farm-animal
farm-puzzle
farm-yard
farm-yard-truck
feather-brain
ferris-wheel
film-viewer
finger-fish
finger-nail
finger-paint
finger-print
fire-alarm
fire-department
fire-engine
fire-father
fire-hoop
fire-hydrant
fire-man

fire-place
fire-station
fire-truck
fish-cart
fish-fly
fish-house
fish-net
fish-school
flash-light
flower-pot
fluffer-nutter
fly-paper
foot-ball
foot-ball-game
footie-ballie
foot-print
fruit-basket
gas-station
gas-tank
ginger-ale
god-mother
grape-fruit
green-light
grocery-store
hair-pin
Halloweenparade
Halloween-party
Halloween-time
hand-kerchief
hard-ware
hard-ware-store
Harvard-shirt
Harvard-Squarebus
head-shake
heart-beat
house-bird
ice-berg
ice-cream
ice-cream-cake
ice-cream-cone
ice-creamdessert
ice-vanilla
ice-vanilla-food

ink-pad
iron-fish
jelly-bean
jesus-knot
jingle-pole
kangaroo-house
key-case
key-chain
key-hole
kitchen-door
knight-book
license-number
light-time
lion-fish
lion-heart
lion-tamer
lip-stick
liquid-medicine
log-truck
lumber-company
lunch-time
magic-word
mail-bag
mail-box
mail-car
mail-man
marble-bag
marble-basket
marble-box
matching-game
meadow-pond
microphonecord
milk-man
milk-truck
mommy-chicken
money-gravy
monkey-train
morning-supper
motherkangaroo
motor-boat
motor-scooter
movie-camera
mud-pie
muffin-man
153

nail-file
neck-tie
news-cast
night-stick
night-time
note-book
numbers-bit
nursery-school
oat-meal
oat-meal-box
orange-juice
orange-seed
palm-tree
papa-bear
papa-dragon
paper-clip
paper-cup
paper-plunge
paper-punch
paper-sponge
paper-towel
peach-jam
peanut-butter
pet-shop
pickle-man
picnic-cake
pill-box
pillow-station
pine-apple
pine-apple-juice
pin-game
pin-wheel
plastic-bag
plastic-flower
pocket-book
police-car
police-man
police-thing
police-whistle
popeye-film
popeye-movie
popsicle-stick
post-man
post-office
power-shovel

presidentkennedy
pumpkin-bird
pumpkin-candle
pumpkin-seed
puppet-show
question-mark
rail-road
rail-road-train
rail-road-truck
rain-coat
rain-hat
raisin-bran
rattle-snake
razor-blade
red-light
rein-deer
ring-master
roast-beefchicken-flower
rocket-ship
rope-trick
rubber-band
rubber-tip
safety-book
safety-rule
sail-boat
sales-man
school-bus
school-children
school-room
science-museum
scissors-finger
scotch-tape
sea-amoeba
sea-creature
sea-green
sea-greenmonster
sea-shell
sea-shore
seat-belt
see-saw
service-station
ship-air-plane
ship-boat

Table 29 (con’t)
shoe-ball
shoe-lace
shoe-polish
shoe-store
side-walk
silver-ware
sky-dart
smoke-stack
snow-ball
snow-man
snow-plow
snow-suit
snow-time
soda-water
space-boat
space-chart
space-ship
spice-cake
stamp-basket
stamp-pad
star-fish

station-wagon
steam-shovel
stocking-cap
stone-fly
story-book
straw-berry
street-broom
street-car
street-light
street-lights
sugar-dish
sugar-ice-cream
sugar-spoon
suit-case
summer-time
sun-glasses
sun-shine
sun-suit
supper-time
surprise-game
swan-boat

sword-fish
tagalong-man
tail-light
tank-car
tank-truck
taxi-car
tea-cake
tea-cup
telegraph-key
telephonecompany
televisionprogram
television-set
texaco-star
tinkey-toy
toilet-seat
tomato-soup
tooth-paste
top-floor
toy-box

toy-car
toy-clock
toy-company
toy-house
toy-snake
track-train
train-track
train-whistle
trash-can
trash-truck
treasure-chest
treasure-horse
treasure-house
tree-barn
tree-top
tricycle-stick
t-square
tummy-ache
turkey-pie
turtle-pie
tv-set

valentine-cookie
vanilla-ice
vegetable-part
water-bug
water-color
water-fountain
water-hydrant
water-juice
water-scorpion
water-show
water-tea
week-end
wheaties-box
wheel-barrow
wheel-barrowhorse
white-wash
wind-mill
wind-shield
winter-time
x-ray

Sarah
animal-book
apple-ear
apple-sun
baby-bath
baby-feet
baby-kangaroo
baby-rattle
baby-swing
ball-game
beauty-parlor
bubble-game
bunny-rabbit
cart-wheel
cat-monster
chocolate-kind
clothes-line
cuckoo-clock
dance-school

diamond-ring
doll-blanket
donkey-face
duck-house
duck-sponge
easter-bow
eye-glasses
eye-lash
fairy-god-mother
finger-polish
fire-mandepartment
fishie-water
flower-house
god-mother
hair-pin
halloween-candy
halloween-hat

halloweensomething
hand-cream
hand-kerchief
hand-kerchiefclothes
home-work
house-tree-top
indian-thing
iron-board
jelly-bean
lady-leg
lollipop-door
magazine-beach
mary-poppinsrecord
meat-ball
mike-shake

mommy-girl
monkey-ear
motor-cycle
mountain-rock
night-nightpillow
note-book
peanut-lollipopdoor
phone-book
piano-chord
poodle-swing
powder-thing
puppet-show
root-beer
rope-thing
school-pencil
soap-powder

sugar-plum
sugar-snap
sugar-wheat
tea-cup
teddy-bear-bear
telephone-book
thumb-nail
tiger-hair
tooth-ache
toy-bike
toy-boat
tree-top
valentine-doll
valentine-watch
wall-paper
water-sled
water-thing

Sarah Input
abc-book

after-noon

air-plane

alarm-clock

animal-cracker

154

Table 29 (con’t)
animal-farm
anirmal-farmhouse
apple-eye
apple-juice
apple-pie
apple-teeth
ash-tray
automobileaccident
baby-book
baby-brother
baby-butter-cup
baby-carriage
baby-chair
baby-clothes
baby-doll
baby-monkey
baby-pillow
baby-powder
baby-rabbit
baby-talk
baby-toy
back-yard
ball-pole
ballet-shoe
bamboo-plate
banana-finger
band-aid
barber-shop
base-ball
basket-ball
bath-robe
bath-room
bath-roomkitchen
beach-ball
beach-chair
beach-house
bed-room
bed-time
bed-time-story
belly-ache
belly-button
belt-buckle
bermuda-shorts

bible-song
bird-house
bird-party
birdie-house
birth-day-cake
birth-day-party
birth-day-present
birth-day-record
birth-day-shorts
Bobby-doll
bogey-man
bookdepartment
boy-chickie
bubble-bath
bubble-gum
bulk-head
bunny-bee
bunny-sticker
butter-cup
butter-finger
butter-fly
cake-pan
candy-bar
candy-cane
candy-desk
card-board
carriage-robe
carrot-candle
cash-register
child-thought
china-children
china-man
chocolate-candy
chocolate-chip
chocolate-cookie
chocolate-chipcookie
chrismas-tree
christmas-bell
christmas-card
christmas-dress
christmas-egg
christmas-eve
christmas-play

christmaspresent
christmas-song
christmas-time
christmas-tree
city-child
cocoa-crispies
cocoa-pop
cocoa-puff
coffee-cake
coffee-pot
collection-box
comic-book
comic-strip
conversationpiece
corn-flake
cotton-tail
cow-boy
cow-girl
cow-boy-Bill
cream-soda
cross-bone
crown-princess
cub-scout
cub-scout-party
cuckoo-nut
cup-board
cup-cake
dance-costume
dance-recital
day-light
day-time
dish-pan
dish-towel
dog-baby
dog-food
dog-house
dog-pretzel
doggie-bank
doll-carriage
doll-house
donald-ducksoup
door-bell
door-prize
155

door-prize-cake
dough-nut
dust-pan
ear-drum
ear-muff
ear-muffs
ear-plug
ear-ring
easter-basket
easter-bunny
easter-card
easter-egg
easter-time
eater-card
egg-shell
emergencybutton
extension-cord
eye-brow
face-cloth
fan-club
farm-animal
farm-house
ferris-wheel
field-trip
finger-nail
fink-test
fire-department
fire-engine
fire-house
fire-man
fire-truck
fish-bowl
fish-head
foot-stool
fruit-loop
garden-set
gas-station
gilligan's-islandtree
girl-chickie
girl-friend
globe-santa
grape-tonic
halloween-party
hand-ball

hand-grenade
hand-lotion
hand-lotionpowder
hat-box
head-band
heart-attack
heart-failure
heir-loom
honey-bee
honey-comb
honey-moon
house-key
house-wife
house-work
human-being
ice-box
ice-cream
ice-cube
ice-skate
ice-creamjimmies
indian-tent
jelly-jar
jewelry-box
jingle-bell
kitchen-table
kitty-cat
lady-bug
lamp-shade
light-house
lip-stick
liver-pill
lunch-bag
macaroni-salad
mail-box
mail-man
man-meal
mary-poppinsdoll
mass-noun
mayonnaisesandwich
medicine-cabinet
milk-man
mink-stole

Table 29 (con’t)
mirror-image
monster-apple
monster-picture
movie-picture
nail-polish
nick-name
noon-time
nose-powder
nurse-baby
nursery-rhyme
nursery-school
oat-meal
orange-cake
orange-juice
pan-cake
paper-boy
paper-doll
peanut-butter
peanut-door
peanut-lollipop
peanut-buttersandwich
peg-board
peg-game
penny-bank
pepper-mint

pet-department
pet-shop
phone-call
picture-book
pie-plate
pig-farm
pig-sty
pig-tail
pigeon-coop
piggie-back
piggie-bank
piggie-box
pine-apple
play-dough
play-groundapple
pocket-book
police-man
pony-express
pony-tail
poodle-dress
post-office
pot-luck
potato-chip
potato-salad
poultry-farm

powder-puff
preschool-kid
princesstelephone
pussy-cat
pussy-dog
rabbit-people
raisin-bran
rein-deer
roof-thing
sail-boat
salad-roll
sand-box
sand-man
sand-shark
santa-clauspuzzle
savings-bond
scooter-doll
Scooter-suit-case
septic-tank
shower-curtain
snow-man
space-ship
sport-jacket
sticker-bur

156

sticker-bur-bush
straw-berry
street-corner
sugar-baby
sugar-bear
sugar-bowl
sugar-daddy
sugar-pop
suit-case
summer-squash
summer-time
sun-apple
sun-burn
sun-glasses
sun-suit
sweet-heart
table-cloth
table-cover
tea-bag
tea-party
tea-pot
tea-spoon
teddy-bear
telephone-call
telephone-car

telephoneconversation
telephone-pole
thanksgivingdinner
toe-nail
toe-shoe
tooth-brush
tooth-paste
tootsie-roll
toy-box
toy-department
toy-party
toy-piano
train-track
tree-light
tummy-ache
tuna-fish
tuna-fish-icecream
tv-desk
vanity-table
variety-store
watch-band
week-end

APPENDIX C. Production Forms for VC Elicited Production Task
Table 30. VC Production: Counts for All Production Forms By Age Group
Form
OVer
Ver
NN
Noun
Phrasal
No Response
VO
MVer
VPO
OVerPer
VerO
VerOer
OVPer
VOM
VPOM
VerM
VerOM
VPer
VingM
MOVer
VPM
OV
OVerM
VOer
MVerO
VerPer
VerP
VM
VingOM
VingO

Example
table-cleaner
cleaner
table-girl
girl
a girl who cleans tables
--clean-table
girl-cleaner
clean-up-table
table-cleaner-upper
cleaner-table
cleaner-tabler
table-clean-upper
clean-table-girl
clean-up-table-girl
cleaner-girl
cleaner-table-girl
clean-upper
cleaning-girl
girl-table-cleaner
clean-up-girl
table-clean
table-cleaner-girl
clean-tabler
girl-cleaner-table
cleaner-upper
cleaner-up
clean-girl
cleaning-table-girl
cleaning-table

Younger
Older
Children Children
33
86
17
25
13
19
24
6
15
15
10
9
15
0
2
9
8
2
1
0
2
5
4
2
0
4
1
3
2
2
2
2
3
0
3
0
2
1
1
0
0
2
1
1
0
1
1
0
0
1
0
1
0
1
1
0
0
1
1
0

157

Adults
145
3
12
9
3
0
0
0
0
6
0
0
1
0
0
0
0
0
0
1
0
0
0
0
0
0
0
0
0
0

Total
264
45
44
39
33
19
15
11
10
7
7
6
5
4
4
4
3
3
3
2
2
2
1
1
1
1
1
1
1
1

APPENDIX D. Responses by Verb for Child Verbal Compound Interpretation Study

1.0

.5

Younger Children

Older Children
Adults

Atelic

Telic

Particle

158

put-away

pick-up

pass-out

fill-up

paint

make

draw

clean

tickle

pull

protect

.0
carry

Proportion of Argument Responses

Figure 35. VC Interpretation: Proportion of Argument Responses by Individual Verb

REFERENCES

159

REFERENCES

Altman, D. G. (1990). Practical statistics for medical research. CRC Press.
Bisetto, A., & Scalise, S. (2005). The classification of compounds. Lingue e Lingaggio, 4(2).
Bobaljik, J. D. (2003). Auspicious compounds. McGill Working Papers in Linguistics, 17(2), 6572.
Borer, H. (1994). The projection of arguments. In E. Benedicto and J. Runner (eds.)
Functional projections, University of Massachusetts Occasional Papers 17, 19-47. GLSA,
Amherst MA.
Borer, H. (2012). In the event of a nominal. The Theta System, 103-15.
Borer, H., & Wexler, K. (1987). The maturation of syntax (pp. 123-172). Springer
Netherlands.
Brisard, F., Laarman, E., & Nicoladis, E. (2008). Clausal order and the acquisition of Dutch
deverbal compounds. Morphology, 18(2), 143-166.
Brown, R. (1973). A first language: The early stages. Harvard University Press.
Chomsky, N. (1965). Aspects of the Theory of Syntax (No. 11). MIT press.
Chomsky, N. (1968). Language and mind. New York: Harcourt Brace Jovanovich.
Clark, E. V. (1981). Lexical innovations: How children learn to create new words. In W.
Deutsch (Ed.), The child’s construction of language. London: Academic Press, (299-328).
Clark, E. V., Gelman, S. A., & Lane, N. M. (1985). Compound nouns and category structure
in young children. Child Development, 84-94.
Clark, E. V., Hecht, F., & Mulford, R. C. (1986). Coining complex compounds in English:
Affixes and word order in acquisition. Linguistics, 24(1), 7-3.
Clark, E. V., & Berman, R. A. (1987). Types of linguistic knowledge: Interpreting and
producing compound nouns. Journal of Child Language, 14(03), 547-567.
Clark, E. & Barron, B. J. S. (1988). A thrower-button or a button-thrower: Children’s
judgements of grammatical and ungrammatical compound nouns. Linguistics, 26, 3–
19.
Cohen, J. (1960). A coefficient of agreement for nominal scales. Educational and Psychological
Measurement, 20, 37-46.
Davidson, N. S., & Gelman, S. A. (1990). Inductions from novel categories: The role of
language and conceptual structure. Cognitive Development, 5(2), 151-176.
160

Delfitto, D., Fábregas, A., & Melloni, C. (2008, November). Compounding at the interfaces.
In meeting of the North East Linguistic Society, Cornell University.
Di Sciullo, A. M. (2005). Decomposing compounds. Skase Journal of theoretical linguistics,
2(3), 14-33.
Di Sciullo, A. M., & Williams, E. (1987). On the definition of word (Vol. 14). Cambridge, MA:
MIT press.
Downing, P. (1977). On the creation and use of English compound nouns. Language, 810842.
Evers, A. and Kampen, J. van. (2001). E-language, I-language and the order of parameter
setting. UiL OTS Working Papers.
Fenson, L., Marchman, V., Thal, D., Dale, P., Reznick, S. & Bates, E. (2006). The MacArthur
Communicative Development Inventories: User’s Guide and Technical Manual. Second edition.
Brookes.
Ferreira, F. & Patson, N. (2007). The ‘good enough’ approach to language comprehension.
Language and Linguistics Compass, 1, 71-83.
Frege, G. (1884). Die Grundlagen der Arithmertik translated as Foundations of Arthmetic by J. L.
Austin, NY: Bantam Books.
Frazier, L & Rayner, K. (1982). Making and correcting errors during sentence
comprehension: Eye movements in the analysis of structurally ambiguous sentences.
Cognitive Psychology, 14, 178-210.
Fraizer, L. (1987). Sentence processing: A tutorial. In M. Colheart (Ed.), Attention and
Performance XXI. Hillsdale, NJ: Erlbaum.
Gamache, J. L, & Schmitt, C. (to appear). Children’s forms as derivational steps: External
evidence for a new synthetic compound structure. Proceedings of the Chicago Linguistics
Society Meeting. Chicago, IL.
Gamache, J. L., Latack, K., Hirzel, M., & Schmitt, C. (2013). Noun categories in children’s
production of noun-noun compounds. Presented at Generative Approaches to Language
Acquisition. Oldenburg, Germany.
Gelman, S. A., & Markman, E. M. (1987). Young children's inductions from natural kinds:
The role of categories and appearances. Child development, 1532-1541.
Gelman, S. A., Wilcox, S. A., & Clark, E. V. (1989). Conceptual and lexical hierarchies in
young children. Cognitive Development, 4(4), 309-326.
Gelman, S. A. (1988). The development of induction within natural kind and artifact
categories. Cognitive psychology, 20(1), 65-95.
161

Gelman, S. A., & Heyman, G. D. (1999). Carrot-eaters and creature-believers: The effects of
lexicalization on children's inferences about social categories.Psychological Science, 10(6),
489-493.
Gleitman, L, & Gleitman, H. Phrase and Paraphase: Some Innovative Uses of Language. New York,
NY: W. W. Norton & Company, Inc.
Hale, K., & Keyser, S. J. (1993). On argument structure and the lexical expression of
syntactic relations. The View from Building 20, 53-109.
Halle, M., & Marantz, A. (1993). Distributed morphology and the pieces of inflection. The
View from Building 20, 111-176.
Harley, H. (2009). Compounding in distributed morphology. The Oxford Handbook of
Compounding, 129-144.
Hiraga, M. (2010). Acquisition of recursive verbal compound nouns. In 32nd Annual
Conference of the Deutsche Gesellschaft für Sprachwissenschaft (DGfS), Humboldt University,
Berlin.
Hiramatsu, K., Snyder, W., and Roeper, T. (2000) Of musical hand chairs and linguistic
swing. In Proceedings of BUCLD 24. Somerville, MA: Cascadilla Press.
Johnston, M., & Busa, F. (1999). Qualia structure and the compositional interpretation of
compounds. In Breadth and depth of semantic lexicons (pp. 167-187). Springer
Netherlands.
Jurafsky, D. (1996). A probabilistic model of lexical and syntactic access and disambiguation.
Cognitive Science, 20, 137-194.
Krott, A., & Nicoladis, E. (2005). Large constituent families help children parse
compounds. Journal of Child Language, 32(1), 139-158.
Krott, A., Gagne, C. L., & Nicoladis, E. (2009). How the parts relate to the whole:
Frequency effects on children’s interpretations of novel compounds. Journal of child
language, 36(01), 85-112.
Krott, A., Gagne, C. L., & Nicoladis, E. (2010). Children's preference for HAS and
LOCATED relations: A word learning bias for noun–noun compounds*.Journal of
child language, 37(2), 373.
Kutas, M., DeLong, K., & Smith, J. (2011). A look around at what lies ahead: Prediction and
predictability in language processing. In M. Bar (Eds.), Predictions in the Brain: Using
Our Past to Generate a Future. Oxford, UK: Oxford University Press.
Lees, R. B. (1960). The grammar of English nominalizations. The Hague: Mouton.
Leonard, L. (1984). The Interpretation of English Noun Sequences on the Computer. Amsterdam:
North-Holland.
162

Levi, J. (1978). The Syntax and Semantics of Complex Nominals. New York, NY: Academic
Press.
Lewis, S. & Phillips, C. (in press). Aligning grammatical theories and language processing
models. Journal of Psycholinguistics Research.
Lieber, R. (1983). Argument linking and compounds in English. Linguistic inquiry, 251-285.
MacWhinney, B. (2000). The CHILDES Project: The database (Vol. 2). Psychology Press.
Marchand, H. (1966). The Categories and Types of Present-Day English Word-Formation. University,
Alabama: University of Alabama Press.
McRae, K., Spivey-Knowlton, M., & Tanenhaus, M. (1998). Modeling the influence of
thematic fit (and other constraints) in on-line sentence comprehension. Journal of
Memory and Language, 38, 283-312.
Mukai, M. (2008). Recursive compounds. Word Structure, 1(2).
Nagpal, J., & Nicoladis, E. (2009). Why are Noun-Verb-er compounds so difficult for
English-speaking children?. The Mental Lexicon, 4(2), 276-301.
Ó Séaghdha, D. Learning Compound Noun Semantics. Doctoral dissertation, University of
Cambridge.
Pinker, S. (1984). Language Learnability and Language Development. Boston, MA: Harvard
University Press.
Pustejovsky, J. (1995). The Generative Lexicon. Cambridge University Press: Cambridge.
Rips, L. J. (1989). Similarity, typicality, and categorization. Similarity and analogical reasoning, 2159.
Roeper, T., & Siegel, M. E. (1978). A lexical transformation for verbal compounds. Linguistic
inquiry, 199-26.
Roeper, T., & Snyder, W. (2005). Language learnability and the forms of recursion. UG and
external systems: Language, brain and computation, 155-169.
Selkirk, E. (1982) The Syntax of Words. Linguistic Inquiry Monograph Series,7.
Smith, L. B., Jones, S. S., & Landau, B. (1996). Naming in young children: a dumb
attentional mechanism? Cognition, 60(2), 143-171.
Snyder, W. (1995). Language Acquisition and Language Variation: The Role of Morphology. Doctoral
dissertation, MIT.
Štekauer, P. (2005). Meaning predictability in word formation: novel, context-free naming units (Vol.
54). John Benjamins Publishing.
163

Tomasello, M., & Brooks, P. J. (1998). Young children's earliest transitive and intransitive
constructions. Cognitive Linguistics, 9, 379-396.
Tomasello, M. (2000). Do young children have adult syntactic competence?.Cognition, 74(3),
209-253.
Tomasello, M. (2009). Constructing a language: A usage-based theory of language acquisition. Harvard
University Press.
Trotzke, A., Bader, M., & Frazier, L. (2013). Third factors and the performance interface in
language design. Biolinguistics, 7, 1-34.
Warren, B. (1978). Semantic patterns of noun-noun compounds. Acta Universitatis
Gothoburgensis. Gothenburg Studies in English Goteborg, 41, 1-266.
Williams, E. (1981). On the notions “lexically related” and “head of a word”. Linguistic
Inquiry, 245-274.
Yang, C.D. (2002). Knowledge and Learning in Natural Language. Oxford University Press.

164