.m..,.. H. ".NFSiS MICHIGAN STATE UNIVERS i/ I I II I!!! (Ill/1W! U n 3 1293 cost/)ILMEOG nix/1mm I'. This is to certify that the dissertation entitled The Specification of Grammatical Number in Engiish presented by Kathieen M. Eberhard has been accepted towards fulfillment of the requirements for W degree in JZhflosophL £3»an Major professor Date W0 MSU is an Affirmative Action/Equal Opponum'ty Institun'on 042771 LIIRARY Michigan State University PLACE iN RETURN BOX to remove this checkout from your record. TO AVOID FINES return on or before date due. DATE DUE DATE DUE DATE DUE i : l J' v.) ,w L fiv '9 ___i —_IF_—i MSU Is An Affirmative Action/Equal Opportunity Institution cmmhd THE SPECIFICATION OF GRAMMATICAL NUMBER IN ENGLISH By Kathleen M. Eberhard A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Psychology 1993 ABSTRACT THE SPECIFICATION OF GRAMMATICAL NUMBER IN ENGLISH By Kathleen M. Eberhard Subject-verb agreement provides a means for examining the nature of the information that must be transmitted among sentence constituents during language production. The four experiments reported here exploited this phenomenon in an effort to gain insight into the nature of the grammatical information that underlies the overt marking of number in English. Results from previous investigations of verb agreement errors (Bock and Eberhard, 1993) suggest that the overt morphological properties of singular and plural count nouns are indicative of an underlying difference in the specification of a grammatical feature of number. The concepts of markedness theory further informed this hypothesis and four agreement-error elicitation experiments empirically examined it. The results of all four experiments suggest that singular count nouns are grammatically unmarked or unspecified for number whereas plural count nouns are marked or specified and that the syntactic operations undeflying agreement are sensitive to whether number is specified. ACKNOWLEDGMENTS I am deeply indebted to my mentor, Kay Bock, whose constant expert guidance traversed many electronic mail miles to ensure that this dissertation would be completed in an ambitious time frame. Her patience and support are matched only by that of my family and friends. I thank Rose Zacks, Tom Carr, Fernanda Ferreira, Barbara Abbott, and Carolyn Harford for their enlightening comments and challenging questions. I especially thank Rose Zacks for her willingness to take on the "chairperson" duties during Kay's absence. My parents deserve special recognition for their steadfast belief that I could do this. Their constant encouragement was a source of motivation and hope even in my most dismal moods (of which there were many). I owe much of my success to them. I have come to realize that I can face just about any challenge with the love and loyal support of my fiance Roger. I acknowledge that neariy every decision over the past five years was made after consulting the wise counsel of my brother Michael and his wife Suzanne. There are no limits to their generosity and love. My sister Patty was also a source of expert advice, and more importantly, she made me laugh when I needed to most. I am grateful for Teddy Rawlin's friendship; her letters of support always arrived at the most appropriate times. Barb Dock taught me more about myself and the world than I could ever possibly learn from a text book. This past year would have been considerably more difficult to live through if I didn't have Jim Zacks just down the hall. Words cannot express the appreciation that I have for his overwhelming kindness and generosity. Finally, I wish to acknowledge the unconditional love and patience that I received from Fluff and Phonzey. TABLE OF CONTENTS Page LIST OF TABLES ............................................................................................................ vi LIST OF FIGURES ......................................................................................................... vii CHAPTER I. INTRODUCTION ..................................................................................................... 1 Previous investigations of errors of attraction .......................................... 4 The experimental paradigm for investigating errors of attraction ............ 5 An alternative account of agreement errors ................................................ 12 The insecurity of singular agreement .......................................................... 14 Markedness in Linguistics ............................................................................ 17 Psycholinguistic inquiries into the markedness relationships ................... 21 Markedness and the specification of the grammatical feature of number ..................................................................................................... 26 The experiments ........................................................................................... 2 8 ll. EXPERIMENT 1 ..................................................................................................... 32 Participants .................................................................................................. 33 Materials ...................................................................................................... 34 Procedure ..................................................................................................... 36 Scoring .......................................................................................................... 37 Analysis ........................................................................................................ 37 Results .......................................................................................................... 37 Effects of Quantifier Type ............................................................................. 43 Discussion ..................................................................................................... 46 III. EXPERIMENT 2 ................................................................................................... 49 Participants .................................................................................................. 49 Materials ...................................................................................................... 49 Analysis ........................................................................................................ 51 Results .......................................................................................................... 5 1 Quantifier Type Examination ...................................................................... 58 Discussion ..................................................................................................... 59 IV. EXPERIMENT 3 .................................................................................................... 61 Participants .................................................................................................. 6 1 Materials ...................................................................................................... 61 Analysis ........................................................................................................ 62 Results .......................................................................................................... 63 Quantifier Examination ................................................................................ 67 Discussion ..................................................................................................... 68 V. EXPERIMENT 4 ..................................................................................................... 69 Participants .................................................................................................. 69 Materials ...................................................................................................... 69 Analysis ........................................................................................................ 72 Results .......................................................................................................... 72 Discussion ..................................................................................................... 79 VI. General Discussion .............................................................................................. 80 Conclusion ..................................................................................................... 88 FOOTNOTES ................................................................................................................ 89 APPENDIX A .............................................................................................................. 90 APPENDIX B .............................................................................................................. 94 REFERENCES ............................................................................................................. 96 LIST OF TABLES Table Page 1 Markedness values for 12 of the English consonants .......................................... 18 2 Examples of sentence preambles in the four conditions of Experiment 1 .......... 34 3 Numbers of responses in each scoring category in Experiment 1 ...................... 38 4 Numbers of miscellaneous responses that had correct, error, or ambiguous verb agreement ................................................................................. 42 5 Numbers of responses in each scoring category for the three quantifier types in Experiment 1 ....................................................................... 44 6 Examples of sentence preambles in Experiment 2 ............................................. 50 7 Numbers of responses in each scoring category in Experiment 2 ...................... 52 8 Number of miscellaneous responses in each agreement score category ............. 5 6 9 Number of responses in each score category for the three quantifier types in Experiment 2 ......................................................................................... 58 10 Examples of sentence preambles in Experiment 3 ............................................. 62 11 Numbers of responses in each scoring category in Experiment 3 ...................... 63 12 The number of miscellaneous responses in each agreement score category ....... 66 13 Number of responses in each scoring category for the three quantifiers in Experiment 3 ................................................................................................. 67 14 Examples of sentence preambles in Experiment 4 ............................................. 7O 15 Numbers of responses in each scoring category in Experiment 4 ...................... 73 16 Numbers of miscellaneous responses containing complex subject phrases in each verb-agreement score category .................................................. 77 vi LIST OF FIGURES Figure Page 1 Net proportions of agreement errors in the determiner "the" and quantifier conditions of Experiment 1 ............................................................ 39 2 Proportions of agreement errors for the match (plural local noun) and mismatch (singular local noun) conditions .................................................... 53 3 Proportions of agreement errors for the match (singular local noun) and mismatch (plural local nouns) conditions ..................................................... 64 4 A graph of the net errors proportions for the two mismatch conditions ............ 74 vii INTRODUCTION Language exhibits oppositional relationships at all levels of linguistic representation. This is readily apparent at the conceptual level where there are opposing concepts such as 9mg and Lad, male and ma, big and 51mm. Perhaps less obvious are the oppositions at the level of phonology where two opposing sounds differentiate words. For example, the phonemes /p/ and /b/ differentiate the English words pi_t_ and bit. Just as opposing sounds at the level of phonology differentiate words, the oppositions at the level of morphology differentiate basic conceptual categories. For example, the conceptual opposition between male and female gender is sometimes reflected in contrasting morphological forms such as, M]; and MESS. waiter and wages; he and she. A more consistent morphological opposition in English is number. The contrasting forms of count nouns (e.g., cat and §_a_ts_) differentiate the meanings 'one' and 'more than one'. Markedness theory, pioneered by Jakobson and Trubetzkoy, views the poles of every linguistic opposition as expressing a difference in complexity. One pole, the unmarked, is simple; the other, the marked, is more complex. At the phonological level, complexity is in terms of the acoustic or articulatory elaboration of the sound. For example, the marked phoneme /b/ may be considered as more complex than the unmarked phoneme /p/ because it is produced with the vibration of the vocal cords whereas /p/ is produced without the added vibration. At the morphological level, complexity is in terms of the augmentation of the morphological form. For example, in English the base form of singular count nouns, e.g., cat, is the unmarked form whereas the inflected form of plural count nouns, e.g., gs, is the marked form. As will be discussed in greater detail below, Jakobson defined the difference in complexity as one between the possession versus absence of a feature or property. For example, the opposition between the phonemes /b/ and /p/ can be defined as the presence versus absence of the articulatory feature of voicing (vibration of the vocal cords), respectively. The concepts of markedness theory were used to inform a hypothesis about the nature of the grammatical information that may underiie the overt morphological markings of number in English. In particular, it is proposed that the opposition between singular and plural count nouns may be characterized as an unspecification (absence) versus specification (presence) of the grammatical feature of number, respectively. Since grammatical number information is important for the implementation of subject- verb agreement (Le, a finite verb agrees in number with the subject noun phrase of the clause), the research examined whether this difference in the specification of number differentially affects the syntactic operations underlying agreement in language production. The impetus for the research came from the consistent findings of an asymmetrical distribution of agreement errors both in sentence production tasks (Bock and Cutting, 1992; Bock and Eberhard, 1993; Bock and Miller, 1991) and in natural observations (e.g., Strang, 1966; Zandevoort, 1961; Bock and Miller, 1991). The agreement errors are referred to as errors of attraction (Strang,1966), a term which characterizes the erroneous verb-agreement with an immediately preceding noun (or local noun), rather than with the subject noun of the clause as in sentence (1). 1.)*T he key to the green filing cabinets were beautiful. ln sentence (1), the verb agrees in number with the local noun Lam rather than with the subject noun, key. The subject phrase in sentence (1) also exemplifies the circumstances that create the asymmetrical distribution of attraction errors, Le, a singular subject noun followed by a phrase ending in a plural local noun. In other words, there is a much greater likelihood of an attraction error following a subject phrase that has a singular subject-plural local noun number pattern than following a subject phrase that has the reverse number pattern (i.e., plural subject-singular local noun). The question that arises from the asymmetry of the singular-plural attraction errors is, what is the attractive property that a plural local noun possesses, and a singular local noun lacks, that causes the verb-agreement process to occasionally derail? The findings from a series of error-elicitation studies by Beck and Eberhard (1993) led them to suggest that the property is a specification of lexical or grammatical number for plural nouns which is lacking or unspecified for singular nouns. According to Bock and Eberhard, lexical or grammatical number is separate from, though often consistent with, a noun's notional number, i.e., the number of the noun's denotation. The determination of a noun's grammatical number comes from the grammatical role of the noun, not its denotation. For example, the noun peg may be used to denote one object, but since its grammatical role is obligatoriiy plural (e.g., th_e_ pants were on the menneguin/ *the panes was en the menneggin) its grammatical number specification is plural. Consistent with the representations in several formal linguistic theories (e.g., Lexical Functional Grammar, Kaplan and Bresnan, 1982), it is assumed that, in addition to grammatical category information (e.g., noun, verb, etc.), the lexical entries of nouns contain grammatical number information. The grammatical number information may be represented as a feature that is either unspecified with a value as in the case of singular unmarked nouns or is specified as in the case of plural marked nouns. The syntactic mechanisms that implement verb-agreement in production are assumed to operate on the basis of this number information. Bock and Eberhard propose that because singular nouns are unspecified for grammatical number, the retrieval of an agreeing singular verb form is not based on explicit number information; rather, singular verbs are retrieved by default. In contrast, since plural nouns are grammatically specified for number, plural-agreeing verbs may be retrieved directly on the basis of this explicit information. The singular- plural pattern in a subject phrase creates the circumstances for an attraction error because the explicit plural specification of the local noun may occasionally preempt the default retrieval of a singular verb form for the unspecified (singular) subject noun. In other words, even though the explicit number specification is not from the subject noun, its existence within the subject phrase (i.e., on a local noun in a postmodifying phrase) may overshadow the subject noun and attract the mechanism that is responsible for retrieving agreeing verb forms. The plural-singular pattern, however, does not create the circumstances for an attraction error because the explicit number on the plural subject noun is not likely to be overshadowed by the number specification on a singular local noun if it has none. Previous investigations of errors of attraction. As stated above, this hypothesis was motivated by the asymmetrical distribution of agreement errors that points to the grammatical plurality of the local noun as the overriding cause of the errors. The investigations of experimentally-elicited agreement errors have systematically eliminated other potential properties that may be influential in the verb-agreement processes. The following section provides an overview of these previous investigations. The experimental paradigm for investigating errors of attraction. Beginning with Bock and Miller (1991 ), the experimental paradigm for investigating errors of attraction involves the auditory presentation of sentence preambles (e.g., The picture on the postcards). The sentence preambles serve as subject phrases for sentences that are generated by the experimental participants (e.g., The picture on the postcards was beautiful). The generated sentences are subsequently inspected for correct and incorrect verb-agreement. To mimic the circumstances that result in naturally occurring attraction errors, the experimental preambles contain a subject noun followed by a phrase containing a local noun that differs in number (i.e., a singular-plural pattern or a plural-singular pattern). Control preambles contain subject and local nouns that are matched in number (singular-singular pattern or plural-plural pattern) and are designed to indicate the effects of extraneous errors and dialects that do not mark agreement. Thus, the incidence of errors for the experimental conditions are assessed according to the incidence of errors that occur for the match control conditions. The properties of the preambles are manipulated under the assumption that if particular properties are found to reliably induce verb-agreement errors, then these same properties are also likely to be involved in the processes behind normal error- free agreement (Bock and Eberhard, 1993). A similar assumption has guided most research on speech errors since Fromkin (1971). Bock and Miller (1991) examined several potential properties that might influence the verb-agreement process. In one experiment (Experiment 2), they examined whether attraction errors are a result of the misidentification of the local noun as the subject of the verb. They proposed that if misidentification is a culpable factor, then the incidence of errors should be greater following preambles that contain a "subject-like” local noun (one that is animate and concrete) than for preambles that contain a local noun that does not possess subject-like properties. It was predicted that if concrete, animate local nouns attract verb-agreement then there would be a greater incidence of errors in the conditions represented in (3) than in (4). 3 a.) The speech of the authors b.) The speeches of the author 4 a.) The author of the speeches b.) The authors of the speech The results, however, showed that, relative to controls, which had a match in number on the subject and local nouns, neither the animacy nor the concreteness of the local noun affected the occurrence of agreement errors. The only factor that reliably influenced the occurrence of errors was the plurality of the local noun (both the conditions represented in 3a and 4a above). Bock and Miller suggest that although animate entities in the to- be-communicated message may be more likely to be assigned to the grammatical role of subject than inanimate ones, once they are assigned to this role their animate or concrete features do not affect the verb-agreement process. Bock and Miller (Experiment 1, 1991) also examined whether plurality in the message (or meaning) of the preambles affects the occurrence of errors. The manipulation involved a comparison of the ability of two different types of preambles to elicit an error of attraction. One type of preamble was independently judged to refer to multiple entities. The other type of preamble was independently judged to refer to only a single entity. The contrasting preamble types are exemplified in (5). 5 a.)The picture on the postcards b.) The key to the cabinets In (Sa) the referent of the preamble is multiple tokens of the same picture, whereas in (5b) the referent is a single token of a key. The correct agreement for both preambles is singular which is consistent with the single referent meaning in (5b), but inconsistent with the multiple referent meaning in (Sa). Therefore, if plurality in the referent of the preambles causes attraction errors, it was predicted that there would be a greater incidence of errors following the multiple referent condition (5a) than following the single referent condition (5b). However, the number of errors following both types of preambles was equal, and there was no correlation between the judged multiplicity of the preamble's referent and the incidence of errors. This finding, together with the findings of the animacy manipulation, suggests that the disruptive source of information in attraction errors is the plurality of the local noun. Beck and Cutting (1992) provided evidence that the disruption by a plural local noun is greatest in cases where it is a clausernate of the subject noun and the agreeing verb. They found agreement errors to be more frequent following subject phrases that contained a plural local noun in a prepositional phrase (e.g., The claim about the babies), than following subject phrases that contained a plural local noun in a relative or complement clause (e.g., The claim that wolves had raised the babies). With the likely cause of attraction errors narrowed to the plural properties of a local noun within the subject phrase, Bock and Eberhard explored three possible sources of plural number that may be found on a local noun: semantic, grammatical, and morphophonemic. In one experiment they examined whether plurality in the denotation of a local noun alone would influence the occurrence of errors. This was accomplished by employing preambles that contained collective local nouns. Despite their singular form and consistent singular verb-agreement for American English speakers, collective nouns (e.g., a_r_my, eommittee, etc.) possess a plural meaning which results from their denotation of multiple entities. The notional plurality of collective nouns is reflected in several ways. Their singular form can often be the antecedent for a plural pronoun. This occurs even in the presence of singular verb agreement as in (6) (Nixon, 1972; Gemsbacher, 1990) . 6.) The W is going to meet tomorrow. They will decide on the issue then. Additionally, although American English speakers predominantly use singular verb agreement with the singular form of collectives, British English speakers are inclined to use plural verb agreement (Quirk, Greenbaum, Leech, and Svartvik, 1972; Nixon, 1972; Bauer, 1988). Lastly, in a judgment task Bock and Eberhard found that the singular forms of collectives (e.g., my) are judged to refer to 'more than one thing' reliably more often than the singular forms of non-collective count nouns (e.g., Mist). Given these notional characteristics of collective nouns, Bock and Eberhard investigated their effects on agreement errors by presenting preambles such as (7) below. 7 a.) The strength of the army b.) The strength of the armies c.) The strength of the soldier. d.) The strength of the soldiers They reasoned that if the plural information on a local noun that influences errors of attraction is semantic, then there should be a greater incidence of errors following singular collective local nouns (73) than following singular non-collective control local nouns (7c). Their results showed that there were no more errors following the singular collective local nouns than following singular non-collective local nouns; only when the local nouns were marked as grammatically plural (both collective (7b) and non- collective (7d)) was there a significant occurrence of errors. This provides additional evidence that the source of the disruptive plural information in attraction errors is not semantic. In three other experiments, Bock and Eberhard examined the effects of the morphophonological markings of number on attraction errors. Morphophonological markings are highly correlated with the grammatical number of lexical items. Both nouns and verbs indicate number with the allomorphs /z/, /s/, or lbzl, and there is a systematic covariation in the morphophonemic indications of number between a subject noun and the agreeing verb: When one contains an allomorph the other does not (e.g., girl talks versus girls talk). The result of this covariation is the elimination of a repetition of the same phonological ending for subject nouns and agreeing verbs. Bock and Eberhard suggested that this covariation may influence the verb- agreement process. The nature of this hypothesized influence is suggested in work by Stemberger and MacWhinney (1986). In a verb production task, they found that third person present tense verbs that end in either the phonological segment /2/ or /s/ (e.g., ehgeses) were more likely to result in a no-marking error (an erroneous deletion of the inflection, e.g., em->em'5_e) than verbs that do not end in a third person-sounding segment. They suggested that no-marking errors occur as a result of the connections between a phonological output buffer and the morphernes in the lexicon. For example, since both the morphernes egggse and the present tense inflection g require the segment /2/ as their final output segment, the insertion of /2/ into the ouput buffer by m occasionally fools the :s morpheme into believing that it has already been instantiated in the output buffer. According to Bock and Eberhard, a similar explanation could account for the singular-plural asymmetry in attraction errors. For example, an error such as T_he king of the islands mle... may be a side-effect of the encoding of the plural affix /z/ of the local noun islegds This encoding may have fooled the production system into believing that the singular affix on the verb (i.e., the /z/ for LINES) was instantiated. Based on this proposal, Bock and Eberhard's first experiment was designed to examine whether the verb-agreement processes could be fooled or influenced by a plural sounding segment on a singular local noun. This required the use of singular nouns that contained a final segment that was matched to the /z/ or /s/ of a true plural form. The incidence of errors in this "pseudoplural" local noun condition was compared to a 10 condition that contained a local noun that was actually marked as plural. The control condition contained a singular local noun that did not end in a plural sounding segment. The three conditions are represented in (8). 8 a.) The player on the course b.) The player on the courts c.) The player on the court The findings demonstrated no errors following either the pseudoplural local noun condition or the true singular local noun condition (8a and c, respectively). Only when the local noun was a true plural did errors occur (8b). Although the final segments of the pseudoplural local nouns were matched to the final segment of actual plural nouns, they were illegal plural allomorphs for the sequence of sounds that would be taken as the stems of the pseudoplurals. For example, the stem of the pseudoplural word ewe would be gee, and the legal plural allomorph for a stem such as m is /z/, not /s/. 50, in the second experiment , Bock and Eberhard employed local nouns that were homophones. One version of the homophone was a true plural noun (e.g., c_rews) and the other was a singular unmarked noun Me). As homophones, the local nouns were completely matched in phonology; only their grammatical number differed. Bock and Eberhard proposed that if the encoding of a local noun's plural phonology influences the agreement process, then the incidence of attraction errors following both versions of the homophones should be equal. Although this manipulation required visual rather than the usual auditory presentation, it was successful in eliciting errors. Consistent with the findings of the previous experiment, the agreement errors occurred only when the local noun was a true plural. This evidence suggests that the local noun's plural morphophonology is not the major source of disruptive information in attraction errors, and it points to grammatical number as being the prominent factor. 11 A third experiment offered additional confirmation of the role of grammatical number in errors of attraction. In this experiment, Bock and Eberhard examined the incidence of attraction errors following regularly and irregularly marked plural local nouns. This involved the comparison of preambles that contained local nouns such as Lag, beg, kids, etc. to those that contained local nouns such as mice. men. children. etc., respectively. The results showed that both types of plural marked local nouns were equally likely to elicit agreement errors, suggesting that it is the abstract grammatical information behind the overt marking of plurality that influences the verb-agreement process. Bock and Eberhard suggest that this abstract information is lexical information in the form of a grammatical specification of number for plural nouns. Since the asymmetrical distribution of attraction errors demonstrates a weakness of singular local nouns in attracting erroneous agreement, singular number information appears to be different from that of plural number. If the overt markings of number indicate the nature of the lexical/grammatical information that is behind them, then the difference may be in the underlying explicit markings or specification of number information. In particular, it is proposed that since plural nouns often contain an overt inflection, this inflection is indicative of the explicitly marked specification of the noun's grammatical role as plural. In contrast, since singular nouns are the base form from which the plural is derived, the unmarked basic form is indicative of unspecification of number. This markedness relationship in grammatical number specification is proposed to influence the syntactic operations that underlie verb-agreement. In particular, there may be a mechanism which operates during verb-agreement that is sensitive to whether there is grammatical number specification on the subject noun. If the mechanism does not detect any explicit specification of number, it may retrieve a singular verb by default or, in other words, as a last resort. 12 Thus, although the unmarked singular may be viewed as simpler in both meaning and morphological form than the marked plural, its simpler unspecification of grammatical number may result in a less direct retrieval of an agreeing verb. In other words, if the lexical entries for singular nouns do not supply a value or specification for the grammatical feature of number, then the mechanism that is responsible for retrieving an agreeing verb may have to determine the value by a default process. In contrast, if the lexical entries of plural nouns contain a specification for the grammatical feature of number, the agreement mechanism may be able to directly retrieve an agreeing verb on the basis of this explicit information. Based on this account, an error of attraction results when the explicit specification of number by a plural local noun in the singular subject phrase is mistakenly detected by the agreement mechanism. An alternative account of agreement errors. This marked/unmarked grammatical number account can be contrasted with an alternative account, proposed by Stemberger (1985), for what superficially appear to be agreement errors. An analysis of bound morpheme loss errors (e.g., The bird sings --> The birg sing) led Stemberger to suggest that they may originate in two different processing failures. First, a failure of the agreement process and second, a general ' failure to access bound rnorphemes. According to Stemberger, the first type of failure is an agreement error that arises from the blending of two sentence structure rules. The erroneous blending generates either a structure that contains a singular subject phrase followed by a plural verb phrase (singular NP-plural VP), or a structure that contains a plural subject phrase followed by a singular verb phrase (plural NP-singular VP). The phrases of these stmctures are assumed to contain diacritic features for number to ensure that the 13 lexical items inserted into them are appropriately singular or plural. The first erroneous structure (singular NP—plural VP) increases the rate of bound morpheme loss errors for present :s. The second erroneous structure (plural NP-singular VP) is reflected in the existence of addition errors for the present _-_s of regular verbs in the error corpus (e.g., The birds sing -> The birds sings). This structural account claims that the agreement errors are independent of the bxical items that are ultimately inserted into the phrase structures; thus, the number information of the inserted lexical items does not influence the occurrence of an agreement error. When applied to errors of attraction, this account claims that a singular subject phrase that contains a plural local noun (i.e., singular subject-plural local noun pattern) should result in a plural verb error equally as often as a singular subject phrase that contains a singular local noun (i.e., singular subject-singular local noun pattern). This is clearly not found in the attraction error experiments that were discussed above. In all investigations, the singular-plural pattern of the subject phrase resulted in a plural verb-agreement error reliably more often than the control subject phrase which had a matched singular-singular pattern. The second type of failure, which superficially results in an agreement error for regular verbs, is attributed to a difficulty in accessing morphological structures that contain bound rnorphernes. Stemberger argues that some characteristic of bound morphernes, in general, makes them difficult to access. The difficulty may arise from the greater superficial complexity of the bound morpheme form (i.e., they are marked by an additional segment), a greater semantic complexity (i.e., the additional morpheme often adds meaning), or, as he argues, a higher frequency related morphological structure (i.e., the base or unbound form) is accessed instead. The control conditions employed in the attraction error studies are designed to assess the incidence of this second type of error. The subject phrases in the control 14 condition always have a match in number between the subject and local nouns. If an experimental condition has a mismatch singular-plural number pattern, the corresponding control condition will have a match singular-singular number pattern. Thus, the match control condition requires the same agreeing verb form as its corresponding mismatch experimental condition. This means that, for whatever reasons bound-morpheme loss (or addition) errors arise from, there are as many opportunities for them to occur in the control conditions as in the experimental conditions. Since the incidence of errors is always greater in the mismatch experimental conditions than in the match control conditions, the errors in the experimental conditions cannot be explained as simple loss errors. The insecurity of singular agreement. As discussed above, because singular nouns are assumed to be unmarked or unspecified for grammatical number the implementation of singular agreement may be less direct than the implementation of plural agreement. This implies that singular verb agreement is less secure. If this is true then it may be reflected in a comparison of the verb agreement following the singular subject-singular local noun subject phrases with the agreement following the plural subject-plural local noun subject phrases. This comparison in Bock and Eberhard's third experiment supports the implication. This is the only experiment in that paper that contained both patterns of number in their experimental subject phrases (i.e., singular-plural as well as plural singular). Thus, both match control conditions were also employed (i.e., singular singular and plural -plural ). The results showed that singular subject nouns were associated with fewer correct responses than plural subject nouns in both the experimental mismatch and the control match conditions. 15 Bock and Eberhard point out that although these findings suggest that agreement with singular subject nouns is less ”secure", there are no similar findings in the studies of Bock and Miller. The experiments conducted by Bock and Miller also included both match control conditions (singular-singular subject phrases and plural-plural subject phrases). However, the comparison between their control conditions shows that correct agreement was equally likely following singular subject nouns and plural subject nouns. A very different sort of argument for a difference between singular and plural agreement comes from a study by Cowart and Calms (1987). They provide evidence from verb naming-latencies that suggests plural nominals delay the retrieval of singular verbs. In their study, subjects heard a sentence preamble, and at the offset of the preamble they named either is or ere which appeared on a computer screen. Naming latencies were collected and analyzed according to various features of the preambles. The preambles contained either the pronoun ygg or the plural pronoun tbey. Both pronouns were followed by a subordinate clause that ended in an amblguously plural or singular lexical noun phrase. The ambiguous noun phrase had a word ending in the suffix :im (e.g., byibg) followed by a plural noun form (e.g., eggs). The suffixed word could be interpreted either as an adjective (i.e., a present participle), in which case the phrase would be plural, or as a gerund, in which case the phrase would be singular. The conditions were designed to explore whether the pronoun M would bias the interpretation of the subsequent ambiguous noun phrase to be plural. This biasing effect was compared to a control condition that contained the pronoun ygg. Examples of the conditions are below. 1.) Even though they use very little oil, frying eggs... 2.) Even though you use very little oil, frying eggs... 3.) Even though they eat very little oil, frying eggs... 4.) Even though you eat very little oil, frying eggs... 16 In (1) the ambiguous noun phrase frying eggs may be coreferenced with the plural pronoun mey, thereby biasing a plural interpretation. However, a coreference of t_he_y with fgying eggs in (3) results in an anomalous meaning due to the selectional restrictions of the verb egg. The naming latencies showed that is was named 28 milliseconds more slowly in the preambles containing eh_ey than in the preambles containing m This slowing effect for naming is occurred regardless of whether a coreferencing of mey with the ambiguous noun phrase was anomalous (e.g., sentence 3 above). The naming latencies for ere did not differ when the preceding context contained yeg or tbey. Furthermore, the mean naming latencies for ere and is were about equal in the context that contained the pronoun m These findings suggest that the presence of a plural nominal (i.e., t_h_ey) interferes with the naming of a singular verb form, whereas a singular pronoun does not seem to influence the time that it takes to name either a singular or plural verb form. This is similar to the asymmetrical distribution of attraction errors that suggests a plural local noun interferes with correct singular verb agreement more so than a singular local noun interferes with correct plural agreement. In the case Of attraction errors, the interference by a plural local noun is proposed to be a result of a marked specification for grammatical number that, by virtue of being explicit, may become mistakenly detected by the verb agreement operations. The apparent undisruptiveness of a singular local noun is proposed to result from an unmarked specification of number that, by virtue of being less specific, does not attract the verb agreement operations. Both linguistic and psycholinguistic inquiries into the markedness relationship view the unmarked term as less specific than the marked. The following two sections address this view. 17 Markedness in Linguistics. Jakobson and colleagues' early work on markedness theory focused on analyzing the phonemic systems of various languages (e.g., Jakobson and Halle, 1956). These analyses resulted in several core principles of markedness theory. One principle is that bxical representations can be decomposed into sets of features. Each feature is binary; i.e., it can be specified as either "present" (+) or "absent" (-). A feature that is specified as "present" is referred to as a marked feature. Jakobson's application of these markedness principles to the level of phonology resulted in phonemic systems that exhibited a definite structure, pattern, and symmetry. A simplified illustration of this is in Table 1 which contains the phonemic feature classification of 12 of the English consonants.‘ Each of these 12 consonants is opposed to every other creating a total of 66 oppositions. This number, however, is reduced to six when the consonants are decomposed into sets of articulatory or acoustic features. This is because the decomposition allows phonemes to be classified according to shared feature specifications. The oppositions can then be defined for classes of phonemes. For example, the consonants /m/, /n/, and /'I/ are classified as nasals because they share a marked specification for the feature [nasal]. The opposition between these three nasals and the consonants /p/, /t/, and /k/, respectively (as illustrated in the word pairs seem/seep, sea/sea, sieg/gelg) is due to the difference in the specification of this one feature. 18 Table 1. Markedness values for 12 of the English consonants. 12 of the Erplish Consonants Features2 p t k b d g f s "’ m n . nasal - - - - - - - - - + + + compact - - + - - + - - + - - + continuant - - - - - - + + + - - - strident - - - - - - - + + - - - voiced - - - + + + - - - - - - Lgrave + - + + - + + - - + - + Markedness theory not only provides an economical description of the phonemic oppositions but it also defines the nature of the oppositions. In particular, because the unmarked phoneme of an opposing pair lacks a feature that the marked phoneme possesses, its articulation is considered to require less deformation of the vocal tract from the neutral rest position; i.e., it is articulated more simply or naturally than the marked phoneme (Battistella, 1990). Greenberg (1966) argues that the simplicity of the unmarked phoneme is reflected in its distributional properties. One such property is its ability to occur in environments of neutralization, i.e., environments where the opposition between two phonemes is suspended. For example, although the opposition between the consonants /p/ and /b/ is evident in the environment immediately preceding ill where they differentiate the two words bill and bifl, the opposition between them is neutralized in the environment following 5; where only the simpler unmarked phoneme /p/ can occur, e.g., spjfl but not *smfl. The ability to occur in environments of neutralization leads to the wider distribution of unmarked phonemes, and Greenberg (1966) has found that 19 unmarked phonemes consistently show a greater frequency of occurrence than marked phonemes. The ability to structure the phonemic system through markedness theory inspired Jakobson to extend the theory beyond the phonological level of language. In his analysis of the Russian verbal system (Jakobson, 1957), he structured the verbal categories according to the presence or absence of semantic features which were determined by the meanings of verbs. For example, he viewed the opposition between finite and infinitive verbs as a difference in the specification of the feature of tense. Since infinitives do not specify tense, whereas finite verbs do, he considered infinitives to be the unmarked or more basic verb. The application of the principles of markedness to semantic categories led Jakobson to the following statement The general meaning of a marked category states the presence of a certain property A; the general meaning of the corresponding unmarked category states nothing about the presence of A and is used chiefly but not exclusively to indicate the absence of A. (Jakobson, 1984, p.47). Thus, just as the absence of an articulatory feature specification permits the unmarked phoneme to occur both in environments of contrast and in environments of neutralization, the absence of a semantic feature specification should allow the unmarked semantic meaning to occur both in contexts where it represents the specific opposite of the marked meaning and in contexts where it represents a neutral meaning. This duality in usage is illustrated with the opposing lexical items hes:= and hestess. In traditional English, male gender is semantically unmarked and female gender is marked. As the unmarked term, bo_st_ may be used in a context that reflects the specific opposite of the marked female meaning as in (7), (7) Your host and hostess this evening are Mr. and Mrs. Smith. 20 or best may be used in a context that reflects a neutral meaning (i.e., one that represents both male and female simultaneously) as in (8). (8) Your hosts/*hostesses this evening are Mr. and Mrs. Smith. Waugh (1982).illustrates the duality in the use of the unmarked meaning with opposing singular and plural lexical items. The specific opposition is evident in the contrasting meanings of the singular phrases gne man and a_hg_rs_e and the plural phrases many men and sever_a| horses. However, the unmarked singular can also be used to express a neutral meaning of number as in the statements man is an animal and the horse is_en unbredicteble gas; where the meaning of number is not known or pertinent. The markedness assignments of semantic oppositions are not only reflected in contextual distributions but they are also consistently reflected in the morphological forms. The reflection is in what Jakobson referred to as the "zero-marking" of the unmarked item (Jakobson, 1939; Greenberg, 1966). Zero-marking refers to the tendency of the semantically unmarked item to be the base form of a lexical paradigm and the more specific marked form to be a derivation of the base by affixation or inflection, e.g., m and hgstess, horse and horses. Thus, the inflection on a marked lexical form is viewed as indicative of the presence of a semantic feature that is absent for the opposing unmarked form. Finally, just as the markedness relation between two opposing phonemes was reflected in the greater frequency of occurrence of the unmarked phonemes than the marked phonemes, with respect to the markedness relation of the semantic category of number, Greenberg found that the unmarked singular form of nouns has an overall greater text frequency than the marked plural form (Greenberg, 1966). 21 Psycholinguistic inquiries into the markedness relationships. Several psycholinguistic inquiries into the markedness relationships at the semantic level have empirically examined the view that the unmarked meaning is less specific than the marked meaning. Clark and Chase (1974) found that when speakers were presented with the task of simply describing two objects, one above the other, they prefer to say "A is above 8" rather than "B is below A". This led Clark and Chase to claim that "above" is the default or unmarked description; i.e., the one that is reliably elicited when there is no constraint on the focus of the description. The contrast between the unfocused and focused meanings of opposing adjectives is illustrated in the questions "How short/small are you?" versus "How tall/big are you?" There is a presupposition of size in the first question that is not present in the second. The unmarked adjectives Ed and big are used in a general or unspecified way when inquiring about size. Recently, Schriefers (1990) examined these dimensional adjectives in a production task. He presented subjects with a display of two geometric figures that differed in either the size dimension (bigger versus smaller) or the length dimension (shorter versus taller). One of the figures was indicated with a cross that cued the subject to name the relation from the perspective of that figure. For example, subjects responded with "smaller" when presented with two circles that had the smaller one indicated by a cross. The findings showed that the unmarked responses (e.g., bugger, rebel) were significantly faster than the marked responses (e.g., s_rr_1elle_r, sbgfier). When the same task required a key press response, rather than a naming response, the markedness effects disappeared. Furthermore, when subjects merely pronounced the adjectives instead of naming them on the basis of making a dimensional judgment, the markedness effects also disappeared. These findings led Schriefers to suggest that the markedness effects were located at a stage of lexical access rather than at an earlier conceptual encoding stage or at a later articulatory stage. 22 Schriefers‘ findings are consistent with the view that the more specific meaning of marked lexical items reflects an underlying specification of semantic features or properties that the less focused unmarked meaning lacks. This view has been applied to research into the memory for both adjectives (Clark and Card, 1969) and verbs (Clark and Stafford, 1969). The assumption underlying this research is that if the more specific meaning of marked lexical items is due to the possession or specification of semantic features these additional features may become lost during the retrieval or recall process. As a result there should be a systematic observation in recall errors that results in the unmarked items being substituted for the marked items. Clark and Stafford (1969) provided evidence for this assumption in their examination of recall for verb forms that varied in the markedness oppositions of tense and aSpect. They defined the markedness relations according to which pole possessed a greater restriction on the verb's meaning. For example, since the non-progressive verb's meaning ( e.g., walked) is less limited or restricted in duration than the progressive verb's meaning (e.g., was walking), the non-progressive was considered to be unmarked and the progressive was considered to be marked. Participants were presented with sentences that had one of eight different verb fomis that varied according to either tense or aspect oppositions. Recall was cued by the subject noun phrase of each sentence. The results showed that recall errors for verb forms were systematically simplified in the direction of the unmarked form being recalled instead of the marked form (i.e., non-progressive forms were recalled in place of progressive forms). Lapointe and Dell (1989) conducted a similar investigation of the effects of markedness relations among verbs, but in a short- rather than long-term recall task. Subjects were presented with two simple sentences each of which had one of four possible verb fomis. Ordered from least complex (most unmarked) to most complex (most marked), the four verb forms were: is V+ing, was V+ing, has V +ed, or must V. 23 The ordering was based on the complexity of semantic notions that the particular verb forms expressed (Lapointe, 1985; 1988). The participants viewed the sentences for eight seconds and then were cued with the subject noun phrase of one of the sentences to produce the remainder of the sentence. The results showed that 9% of the responses resulted in a substitution error. The substitution errors were generally favorable to the predictions of the markedness rankings. The less marked progressive forms ("was V+-ing", and "is V+-ing") replaced the more complex perfect forms. However, the past form of the progressive was produced in place of the present form, and "must V" was substituted more often than "has V+ed". Lapointe and Dell suggest that the tense reversal may not be reflective of the markedness rankings, but rather due to a task bias: Since the sentences were presented before a production cue, producing them in the past tense may have seemed more accurate. However, the reversal of the ranking for the two most complex forms was viewed as more challenging to the predictions, and Lapointe and Dell suggest that the reversal may be attributable to factors other than their semantic markedness. Although the overall evidence from memory tasks suggests that the additional features of the marked meaning become lost during the retrieval or recall process, there is an alternative explanation for these results. The additional features may have been lost during the initial encoding or comprehension of the stimuli in the study phase. Harris and Brewer (1973; Brewer and Harris, 1974) argued that a relevant context is needed to fully understand the more complex meaning of verb tense; i.e., a shift in recall to simpler fomis is also predicted if full understanding is not achieved in the first place. Whether these findings of less secure processing of marked items at the semantic level are viewed as having occurred during either comprehension or recall, they nonetheless contrast with recent findings of less secure processing of unmarked items at the phonological level in a production task. In particular, Stemberger (1991) found 24 that in both naturally occurring and experimentally elicited contextual errors (i.e., errors in which a phonological segment was mispronounced as a segment that appeared in a nearby word) marked phonological segments tended to replace unmarked segments more often than the reverse. The asymmetries were systematically explored in several experiments employing an error induction technique. The technique involved the presentation of word pairs for participants to silently prepare to say aloud. On some trials, the presentation of two words was followed by a cue for their overt production. On critical trials, the cued (target) pairs were preceded by prime pairs that prepared the participants to say the initial consonants of the target pair in a particular'order, e.g., the prime pair, msbig prepared the initial consonant order /s/-/sh/. The target pair, however, had the initial consonants reversed, e.g., the target pair 5M; sifi. Thus, a primed error resulted in the replacement of /sh/ with /s/ in the production of the first word of the target pair (e.g., sbgcjg~> seek). Because there was an equal number of target pairs that had preceding primes that prepared the opposite replacement error (e.g., the pair, sMQ-stbn primed the target pair, slit-M) Stemberger was able to examine whether there were asymmetries in the direction of replacement errors between the two opposing consonants. His findings showed that consonants unmarked for voicing were replaced by consonants marked for voicing more often than reverse; consonants unmarked for the feature continuant (i.e., stop consonants) were replaced by consonants marked as continuants more often than reverse, and consonants unmarked for nasal were replaced by nasals more often than the reverse. These findings were explained in terms of the underspecification of features. According to Stemberger, early phonological theories assume that all features in the lexical entries of segments must be specified as either [+F] or [-F]. However, more recent theories allow for some feature values to be left blank or underspecified. In 25 particular, it has been proposed that the underspecified value should correspond to the universally unmarked value of the feature, i.e., the most frequent value for the segment class across languages. For example, Stemberger claims that most phonologists view unvoiced non-continuants (stops) as more frequent cross-linguistically than voiced non—continuants; therefore, unvoiced is usually considered to be an unmarked feature for these consonants. Given this, it may not be necessary to specify this feature value in the lexical entries of unvoiced non-continuants. The blank value for the feature may be specified at a later point in the processing. Because the values of features are assumed to be binary (+, present; -, absent), the specification of a blank value can be determined by virtue of knowing that it is not the other (specified) value. Thus, the process that is responsible for specifying blank values would use the most frequent (i.e., unmarked) value of the segment's class as a default specification unless the contrasting binary value is specified. Stemberger argues that by eliminating specified values for features in the lexical entries, the amount of information that is stored in the lexicon is reduced. However, he points out that the need for a default specification process does increase the complexity of the processing system. This increase in the complexity of processing segments with underspecified values causes them to be processed less securely than segments that have their feature values specified. Stemberger claims that the asymmetry in the contextual errors involving underspecified and specified segments reflects a processing failure that occurs at a point before the underspecified value becomes specified. He suggests that the failure occurs when the specified value of an upcoming marked segment is highly activated and is enoneously selected as the value for a current underspecified segment. The asymmetry arises because an activated specification of a current marked segment will not have any competition from the specification of an upcoming unmarked segment if it has none. 26 This account of the asymmetry in errors at the phonological level is similar to Bock and Eberhard's account of the asymmetry of agreement errors at the grammatical level. They propose that the singular forms of count nouns lack a specification for the grammatical feature of number whereas the plural forms contain an explicit specification. The syntactic mechanisms underlying verb agreement are assumed to be sensitive to whether number is explicitly specified. When it isn't, singular verbs are retrieved by default. The greater number of agreement errors following subject phrases with a singular subject-plural local noun number pattern is due to the explicit specification of number on a local noun overshadowing the unspecification of number on the subject noun. The type of number information that the agreement mechanism is claimed to be sensitive to, however, is grammatical- not semantic or morphophonemic. The next section addresses this issue. Markedness and the specification of the grammatical feature of number. The category of number is not only a semantic property that is highly correlated with morphological properties, but it is a grammatical or syntactic property as well. Although there is a relatively consistent reflection of notional number in the grammatical roles and morphological properties of lexical items, there are many cases where the reflection is not explicit or is misaligned. Number is not explicitly reflected in the morphology of "zero-plural" nouns such as sabnbb, sbe_eb, deer, etc., where the same form is used to denote both singular and plural entities. This is also the case for a few invariable plural nouns in English such as ea_ttl_e, mice, and beegle. Number is misaligned in both the morphology and grammatical roles of pluralia tantum nouns (i.e., the summation plurals, sc_iss_ors, pa_nts, etc.) when their use refers to single objects. There are also nouns whose morphological and semantic properties would suggest that 27 their grammatical role could or should be plural (e.g., linggisties, be_ws_, mm, yet they are grammatically singular. These cases suggest that there is a grammatical category of number (Bock and Eberhard, 1993), and that the determination of whether a lexical item (e.g., a noun) is grammatically singular or plural depends on the role the item plays in the grammar. Furthermore, as discussed above, the evidence from studies of errors of attraction suggests that it is grammatical number, rather than number meaning or morphophonology, that effectively influences the verb agreement process. Thus, although the markedness relation is typically applied to the conceptual opposition of number meaning (one versus more than one), which is correlated with the morphological properties of count nouns (uninflected versus inflected), it is proposed that there is also a markedness relation in the opposition of the grammatical number of nouns (unspecified singular versus specified plural). In terms of the operations that underlie subject-verb agreement, the greater specificity of grammatically plural nouns may exert a dominance over the unspecification of grammatically singular nouns. The asymmetric distribution of attraction errors appears to reflect this dominance. The experiments reported here examine this issue by investigating whether singular number may become grammatically marked or explicitly specified as singular. If so, then the marked plural specification of a local noun should be less dominant; i.e., it should be less likely to attract erroneous agreement. A grammatical marking of singular number may arise from the combination of an unmarked noun with a singular quantifier such as obs. For example, the phrase erg berse not only semantically specifies the opposite of the plural meaning (i.e., 'one'_ versus ' more than one'), but the quantifier gbe grammatically marks the noun phrase to be singular. This singular marking is reflected in the quantifier's inability to combine 28 with a noun that is grammatically marked as plural (e.g., * gne boys, * gne scissors). In contrast, the determiner file may combine with either a grammatically unmarked singular noun ( e.g., me_bgy) or a grammatically marked plural noun (e.g., the boys, the scissgrs). All previous investigations of subject—verb agreement employed noun phrases that contained the determiner tbe and therefore did not examine whether singular number can be specified by an overt morpheme. The quantifiers ea_ch and e_very, like the cardinal quantifier gbe, must also combine with an unmarked count noun. Although the quantifiers are not bound to the noun that they combine with, their requirement of singular verb agreement resembles the requirement of plural verb agreement imposed by a bound plural morpheme. Thus, it may be assumed that all three of these quantifiers grammatically mark or specify number information to be singular (e.g., Pollard and Sag, 1991). The experiments. Based on the above analysis, when an unmarked subject noun is preceded by the quantifiers eyegr, mar me, the operations of verb-agreement should be able to access the singular verb form on the basis of specified number information, rather than by an indirect default computation. If there is a mechanism that is sensitive to the specification of number, a detection of singular specification may result in more secure retrieval of a singular agreeing verb form. Thus, a plural specification by a local noun should be less likely to cause an error of attraction. This was examined in the first experiment. It was argued that a plural-singular number pattern in subject phrases (e.g., The pictures en the gsteerd) does not result in as many attraction errors as the reverse pattern because singular unmarked local nouns are less likely to interfere with the plural marking on the subject nouns in the agreement process. However, a marked 29 specification of singular number by a quantifier on a local noun phrase may interfere with the retrieval of a plural verb form for a plural marked subject noun (e.g., 1mg pietgres en eveg bfiteerd). If so, there should be an increase in attraction errors for the plural-singular pattern of subject phrases. This was examined in the second experiment. The proposed effect of a quantifier is that it marks or specifies as singular an otherwise unmarked noun phrase. However, there is an alternative interpretation of the effects of a quantifier. It may be that unmarked singular nouns do contain a singular specification but, relative to the plural specification of marked nouns, it is weak. Thus, a quantifier may not specify singular number, but rather enhance or strengthen a specification. If so, plural quantifiers (e.g., many, mesa, stem should enhance the plural specification of a following plural marked noun. This issue was explored by examining the incidence of attraction errors following local nouns that were only marked as plural, or marked as plural plus plurally quantified. If there is a greater incidence of errors following plurally quantified local nouns than simple plural local nouns, quantifiers may be regarded as simply enhancing the number specification on their following nouns. However, if the incidence of attraction errors is the same then this would be consistent with the hypothesis that the singular is grammatically unmarked. This was explored in the third experiment. The combination of a quantifier with an unmarked noun should create a noun phrase that is grammatically marked or specified for number. However, this does not necessarily mean that an unmarked noun, itself, becomes grammatically specified as singular. To examine whether a noun may on its own carry explicit specification of singular number, the fourth experiment exploited the grammatical distinction between mass and count nouns. 30 Although count nouns tend to denote objects (e.g., _c_a_l_t_, flower. book) and mass nouns tend to denote substances (e.g., geld, ester, m_ud_) the determination of whether a noun is mass or count depends not on the properties of its denotation, but rather on whether the grammar allows it to be individuated (Quirk, Greenbaum, Leech, and Svartvik, 1985; Gordon, 1975). Only count nouns have the ability to be individuated by combining with distributive quantifiers such as eyery and eecb (e.g., evety cat, 22M flower versus *evegr mud, *each water). Additionally, count nouns, as the term implies, have the ability to be counted by combining with cardinal quantifiers (e.g., o_ne deg, tyvg boys versus *one gold, *two mud). Consistent with their ability to be individuated and counted, only count nouns may be grammatically marked as plural (e.g., bass, dggs versus *muds, *wateg). Although some mass nouns may become reclassified as count nouns (Quirk, Greenbaum, Leech, and Svartvik, 1985) when there is a semantic shift that results in a "kind of" denotation (e.g., the waters of the great lakes), the majority of mass nouns cannot become plurally inflected. Furthermore, only mass nouns may combine with non-individuating quantifiers such as _r_n_u_eb and Ii_ttl_e (e.g., much water little mud versus *much a , *little book where little is used as a determiner and not an adjective). The claim that the distinction between mass and count nouns is a grammatical distinction is supported by the existence of mass nouns whose denotation should allow them to become individuated, but they do not. For example, the mass nouns fumitgre, silverware, and lafle denote groups of objects, yet they cannot grammatically combine with individuating quantifiers (*each furniture *everv traffic), nor can they become grammatically marked as plural (*furnitures. *silverweres). In addition, the assignment of abstract nouns as mass or count (e.g., aflice *edviees, 31 examgle/examples, etc.) appears to be based purely on their grammatical roles since their denotation is irrelevant (Gordon, 1985). Gordon (1985) provided developmental evidence that also supports this grammatical distinction. The results of his study suggest that although children from the ages two to five show that they know there is a correlation between the semantic distinction of mass and count nouns (i.e. substance versus object, respectively), they nonetheless rely on the grammatical distribution in their assignment of nouns to these categories. As a grammatical category that obligatoriiy requires singular agreement, mass nouns may contain grammatical specification that is equivalent to an underlying marked specification of grammatical number as singular. This marked singular specification may be manifested in a reduced incidence of attraction errors when a mass noun is the subject noun of a subject phrase that contains a plural local noun. The fourth experiment was designed according to this prediction. CHAPTER II Experiment 1 The first experiment was designed to examine whether the grammatical specification of singular number on the subject noun reduces the occurrence of errors of attraction. To do this, the quantifiers every, each, and erg were compared with the determiner tbe. Since the quantifiers combine only with singular nouns, it may be assumed that they specify singular number (e.g., Pollard and Sag, 1991) so that the resulting noun phrase becomes specified as singular. In contrast, because the determiner tbe can grammatically combine with either a singular or plural noun, it cannot specify number information. Thus, the number of a singular noun phrase containing tbe may remain unspecified. A finding of fewer errors of attraction following preambles that contain subject noun phrases quantified by event, each, or gm , would support the hypothesis that number is grammatically unspecified for the singular form of count nouns. It would also argue that this unspecification is a contributing factor to errors of attraction. Although the quantifiers eaeb and gym obligatoriiy require grammatically singular agreement, there appears to be a conflict in their notional number. When egb and eLecy combine with a noun, a notion of plurality results from the presupposition that there is more than one entity in the resulting noun phrase's referent set. However, despite the plurality in the referent set, both quantifiers focus the interpretation on the individuals or units that constitute the referent set (e.g., Croft, 1991 ). In particular, they force an interpretation in which each member of the referent set acted 32 33 independently of the other members. This focus on the individuals in the set by the quantifiers may be contrasted with the group focus that arises from a plural noun phrase as illustrated in the sentences below (cited by Roberts, 1987). 1.) The women brought a salad to the potluck. 2.) Each/Every woman brought a salad to the potluck. The plural noun phrase "the women" focuses the interpretation of (1) on the group of women who, together, brought a single salad. This group interpretation is not possible with either version of the quantified noun phrases in the second sentence. The quantified noun phrases focus the interpretation of the sentence on the individual members who independently brought separate salads. The individualistic interpretation may be reflected in the grammatically singular specification of these quantifiers. The quantifier 9_ne, unlike eac_h and eyeg, does not have conflicting notional number. It is both grammatically and nationally singular. The previous investigations of Bock and Miller (1991) and Bock and Eberhard (1993) did not find an influence of notional number in errors of attraction. However, a comparison of the agreement for the two quantifiers ea_m and eyety with the agreement for the quantifier gee will illuminate the possible effects the referent meanings may have on the agreement process. Participants. Thirty-two native English speaking participants were recruited from Michigan State University introductory psychology classes for this experiment. They were given extra credit or credit that counted toward a course requirement in return for their participation. 34 Materials. Thirty-six sets of experimental preambles were constructed. Each of the 36 preamble sets contained four preambles, one in each of four conditions represented in Table 2. The complete list of preambles is in Appendix A. Table 2. Examples of Sentence Preambles in the four conditions of Experiment 1. Local Noun Phrase Number Sentence Preamble "'the" Subiect Noun Phra_se Singular (Match Condition) The slogan on the poster Plural (Mismatch Condition) The slogan on the posters Qrgntified "eech. everyt or one" Subiect Noun Ph_res_e_ Singular (Match condition) Every slogan on the poster Plural (Mismatch Condition) Every slogan on the posters Each preamble consisted of a complex subject phrase in which the subject noun phrase was followed by a prepositional phrase. The noun phrase that terminated the prepositional phrase is called the local noun phrase. Each member of a set of preambles was identical except for the determiner that preceded the subject noun (quantifier or tbe) and the number of the local noun phrase (singular or plural). The three quantifiers were e_a_c_h, eyegr and see. There were twelve sets of preambles for each of the three quantifiers. Each of them appeared in one third of the preambles. Within a preamble set, two of the four preambles represented a mismatch condition. Both mismatch preambles contained a singular subject noun phrase that was followed by a plural local noun phrase. However, the mismatch preambles differed in 35 whether the singular subject noun phrase had a quantifier or the determiner _t_h_e. The other two preambles represented the control conditions and contained singular subject and local noun phrases. Because the control preambles required the same correct verb form as the mismatch preambles but cannot create attraction errors, they reveal errors attributable to other problems in creating apprOpriate verb forms, such as bound morpheme loss errors and verb form variations attributable to dialect differences. In addition to the experimental preambles, 64 filler items that consisted of simple noun phrases were created. Half of the filler items consisted of a determiner- adjective-noun sequence; the other half consisted of a determiner-noun sequence. Since all 36 experimental items contained a singular subject noun phrase, 14 of the filler items were singular and 50 were plural. This created an equal number of items on each list that were singular and plural. In order to make the filler items comparable to the experimental items, half of the filler items contained quantifiers and half contained the determiner t_be. Seven of the singular filler items contained me and seven contained a quantifier (three geb, two gem and two gee). Twenty-five of the plural filler items contained tbe and twenty-five contained a quantifier (nine my, eight some, and eight seyerel). Four lists were constructed with each list containing all the filler items and one preamble from each of the 36 experimental condition sets. Each list contained an equal number of preambles that represented each of the four conditions and an equal number of items that represented the three quantifier types (gbe, eyeg, egg) within the two quantifier conditions. The order of the fillers and experimental items was random except for a restriction that two experimental items could not occur consecutively. The order of the filler items and the items representing a particular experimental set were identical across the four lists. 36 Eight filler items began each list. Four of these items were plural and four were singular. Two of the singular items contained tbe and two contained a quantifier (cm and every). Two of the plural filler items contained tbe and two contained a quantifier (m_a_ny and gage). All of the items were digitally recorded by the author at a sampling rate of ZOkHz. The items were edited with sound editing software (MacSpeech Lab II, GW Instruments). The items were edited using the visual display of the waveform accompanied by the audio playback. The editing removed long silent intervals in order to create a moderately fast rate of speech that is natural and perceptible. The items were then reconverted to analog form and recorded onto cassette tape in the order that they appeared on each list. Procedure. Participants were run individually. They were informed that they would hear phrases which were to be used as the beginnings of sentences. They were asked to repeat each phrase and to quickly continue aloud with a sentence conclusion. No other instructions about the forms of their completions were given. The experimenter played each phrase from a cassette recorder. She paused the tape following each phrase to cue the participant for a completion. If the participant did not apprehend the phrase, the experimenter repeated it aloud. The experimenter also encouraged the participants to speak more quickly if their speech rate slowed during the experimental session. The sessions were recorded on audio tape. They lasted about 15 minutes. 37 Scoring. The responses for the experimental preambles were transcribed from audio tape. Each response was assigned to one of four categories. A correct score was given to responses that contained an exact repetition of the preamble that was immediately followed by a number-inflected verb that was correct with respect to the number of the subject noun phrase. An error score was given to responses that met the criteria for the correct score, except that the number of the inflected verb was incorrect with respect to the number of the subject noun phrase. Responses that contained an exact repetition of the preamble that was immediately followed by a verb uninflected for number were assigned to an "ambiguous" category because of the undiagnosability of the number of the verb form. These responses included the past tense of verbs and modals such as should, could, would, etc.. All responses that did not meet the criteria for the above three categories were scored as miscellaneous. The majority of the responses in this category were preamble repetition failures. Analysis. An analysis of variance for each of the score categories was performed first with participants as a random factor and then with items as a random factor. Each analysis contained the within-participant variables of subject noun phrase type (quantified or non-quantified) and local noun number (singular or plural). An effect was considered significant if it was reliable at or beyond the .05 level. The test statistics for participants and items are designated F1 and F2, respectively. Results. Application of the scoring criteria yielded a total of 696 correct scores (60% of all the responses), 123 agreement errors (11% ), 232 ambiguous scores (20% ), and 38 101 miscellaneous response scores (9% ). Table 3 gives the number of responses in each scoring category for each of the four conditions. Table 3. Numbers of responses in each scoring category in Experiment 1. Response Type Local Noun Phrase Number Correct Error Ambigrous Miscellaneous Wse with "the" Singular (Match condition) 216 2 57 13 Plural(Mismatch condition) 121 65 54 48 Subject noun phrase with qgentifiers Leach. every; or one: Singular (Match condition) 209 9 63 7 Plural (Mismatch condition) 150 47 58 33 Agreement error proportions for each of the four conditions were calculated. A condition's error proportion is the number of errors divided by the total number of agreement-marked responses (i.e., number of agreement errors divided by number of agreement errors and correct agreement scores). Figure 1 shows a graph of the error proportions. To discount extraneous agreement errors, as assessed in the control (number match) conditions, the control proportions were subtracted from the mismatch proportions. The resulting net error proportion in the determiner the and the quantifier conditions are shown in Figure 1. 39 Net Error Proportion Quantified Determiner "fl'ell Subject Noun Phrase Type Figure 1. Net proportions of agreement errors in the determiner "the" and quantifier conditions in Experiment 1. The conditions each contained either a singular subject noun phrase that was marked with a singular quantifier (each, every, or one) or a singular subject noun phrase that was unmarked with the determiner “the". As evident in Table 3, there were more errors and fewer correct scores in the mismatch conditions than in the match conditions, showing the usual attraction error effect. However, as shown in Figure 1, this effect was not the same for the quantified preambles as for the th_e preambles. Preambles with singular quantifiers in the subject noun phrase resulted in fewer errors than preambles with generic me in the subject noun phrase. These effects on verb agreement of match versus mismatch (i.e., local noun number) and type of determiner in the subject noun phrase were confirmed statistically. An analysis of variance performed on the correct agreement responses 4o yielded a main effect of local noun number in both the participant and item analyses (F1(1,31) = 49.64 and F2(1,35) = 89.64). This reflected the greater number of correct responses following singular local nouns (match conditions) than following plural local nouns (mismatch conditions) (425 versus 271, respectively). The participant analysis yielded a marginally significant main effect of subject noun phrase type (F1(1,31) = 3.53, p = .07) which reflected the greater number of correct scores in the conditions that contained a quantified subject noun phrase than in the conditions that contained a subject noun phrase with the determiner tbe (359 versus 337, respectively). This main effect was not significant in the item analysis (F2(1,35) = 2.19). Most importantly, there was a significant interaction between the factors of subject noun phrase type and local noun phrase number in both the participant and item analyses of the correct scores (F1(1,31) = 5.94, F2(1,35) = 3.99). The interaction resulted from the disparity in the number of correct scores in the two mismatch conditions. Planned comparisons showed that the greater number of correct agreement responses in the mismatch condition with quantified subject noun phrases (150) than in the mismatch condition with the determiner tbe (121) was reliable for both participants and items. The 95% confidence intervals for these comparisons were 21.34 and 26.06 for participants and items, respectively. Relative to the correct agreement scores, the agreement errors were distributed in a similar but complementary fashion across the four conditions, and this was also confirmed statistically. There was a main effect of local noun phrase number (F1(1,31) = 38.87 and Fz(1,35) = 74.21) which reflected the greater number of agreement errors in the plural local noun conditions than in the singular local noun conditions. The main effect of subject noun phrase type was not significant in either the participant or item analyses (F1(1,31) = 1.31 and F2(1,35) = .97). However, the 41 interaction between subject noun phrase type and local noun phrase number was significant (F1(1,31) = 6.77 and F2(1,35) = 5.65). The significant interaction was again a result of the disparity in the number of responses in the two mismatch conditions. The planned comparison between these two conditions showed that the smaller number of attraction errors following quantified mismatch preambles was reliable. The 95% confidence intervals for the comparisons were 13.88 for participants and 15.21 for items. The distribution of ambiguous scores was fairly even across all four conditions and produced no significant main effects or interactions (all Fs < 1). The analyses of the miscellaneous scores resulted in a significant main effect of local noun number in both the participant and item analyses (F 1(1,31) = 6.00 and F2(1,35) = 4.34). There were more miscellaneous responses in the plural local noun (mismatch) preamble conditions than in the singular local noun (match) preamble conditions (81 versus 20, respectively). There was also a significant main effect of subject noun phrase type in both the participant and item analyses (F1(1,31) = 21.88 and F2(1,35) = 28.23). This effect reflected the greater number of miscellaneous responses for the me subject noun phrase preambles than for the quantified subject noun phrase preambles (61 versus 40, respectively). The interaction was not significant (both Fs < 1.09). Appendix B contains examples of the miscellaneous responses as well as the various reasons for the assignment of the score. The most common reason (64 responses) was a repetition failure that changed the number of either the subject or local noun of the presented preamble. The miscellaneous responses were also examined for verb agreement. Of the 101 responses, 95 contained complex subject phrases in which there was a subject noun phrase followed by a local noun phrase. These 95 responses were categorized as either 42 match subject phrases (the subject and local noun number was the same) or mismatch subject phrases (the subject and local noun number differed). All mismatch subject phrases had a singular subject noun that was followed by a plural local noun. Six responses were not assigned to either subject phrase category because of an addition of a plural morpheme on a singular quantified subject noun (e.g., Each orders for the cooks) which made the number of the subject noun phrase indeterrninable. The remaining 89 responses were given a verb agreement score that was with respect to the number of the actually produced subject noun phrase. The results of this scoring are given in Table 4. Table 4. Numbers of miscellaneous responses that had correct, error, or ambiguous verb agreement. The percentages are based on a total of 89 responses. The actual numbers of responses are given in parentheses. AfiementType Correct Error Am ' uous _llvl_atch Subject th 51% (45) 5% (4) 20% (18) Mismatch Su ' Phrases 7% (6) 11% (10) 6% (6) Table 4 shows that correct agreement was more likely following match than mismatch subject phrases, and incorrect agreement was more likely following mismatch than match subject phrases. This pattern is consistent with the pattern in the main experimental analysis. However, though not evident in Table 4, the verb agreement in the miscellaneous responses did differ in one important respect from that of the main experiment: 0f the ten miscellaneous mismatch responses that resulted in an agreement error, six had subject noun phrases containing a singular quantifier (e.g., Each mistake 43 in the programs) and four had singular subject noun phrases containing the generic tb__ (e.g., The advertisement in the magazines). Although this pattern of agreement errors is opposite to that of the main experiment, it is not responsible for the main experiment's pattern. When these ten miscellaneous errors are added to the errors of the respective preamble conditions in the main experiment, the resulting net error proportions for the mismatch quantified condition and the mismatch she condition are .22 and .35, respectively. A comparison of these net proportions with those given in Figure 1 shows that the error pattern does not change. Effects of Quantifier Type. For the quantified subject noun preambles, the numbers of scores for each of the quantifiers (gne, each, and every) in each of the score categories were calculated and are listed in Table 5. 44 Table 5. Numbers of responses in each scoring category for the three quantifier types in Experiment 1. Response Type Subject Noun Phrase Correct Error Ambiguous Miscellaneous Quantifier Type Singular Local Noun Phrase (Match Condition) One 66 1 26 3 Each 73 4 17 2 Every 70 4 20 2 Plural Local N un Phrase Mi m ch ondi i n One 61 8 19 8 Each 45 16 23 12 Every 44 23 16 1 3 The numbers in the table show that when the subject noun phrases of mismatch preambles contained the quantifier gee, there were more correct responses and fewer agreement errors than when the subject noun phrases contained either the quantifier eacb or em. The differential effects of quantifier type (em, e_vegr, or gn_e) and local noun phrase number (match or mismatch) were examined in a 3X2 analysis of variance on the number of responses in each score category. A repeated measures analysis of variance was conducted with participants as a random factor, and a mixed analysis of variance was conducted with items as a random factor (quantifier type was a between- item factor). The analysis of the correct responses yielded a significant main effect of local noun phrase number (F1(1,31) = 20.05, Fz(1,33) = 22.59) which, consistent with 45 the overall experimental analysis, reflected the greater number of correct responses for match quantified preambles than for mismatch quantified preambles. The main effect of quantifier type was not significant (both Fs < 1.00); however, the interaction between quantifier type and local noun phrase number was significant in the participant analysis (F1(2,62) = 5.20) and marginally significant in the item analysis (F2(2,33) = 3.16, p=.06). Post hoc comparisons in the participant analysis showed that the number of correct responses following mismatch preambles with ane (61) was reliably greater than the number following preambles with eac_h (45) and eyeer (44) (the 95% confidence interval was 14.04); however these differences were not significant in the item analysis (the 95% confidence interval was 18.46). The analysis of the errors resulted in a significant main effect of local noun phrase number (F1(1,31) = 20.31, F2(1,33) = 41.58) as well as a significant main effect of quantifier type (F1(2,62) = 5.53, F2(2,33) = 3.96). As in the overall experimental analyses, the main effect of local noun phrase number reflected the greater number of errors following plural local nouns than following singular local nouns. The main effect of quantifier type reflected the difference in the total number of errors following the three quantifier types: There were only nine attraction errors following preambles containing _on_e whereas there were 20 following preambles containing _e_aeb and 27 following preambles containing em. Post hoc comparisons showed that only the difference between the number of errors following preambles with gee and preambles with eye_ry was significant (the 95% confidence interval was 13.73 for participants and 16.64 for items). There was also a significant interaction in the participant analysis (F1(2,62) = 3.57)which was marginally significant in the item analysis (F2(2,33) = 3.14, p=.06). The 95% confidence intervals for the post hoc comparisons between the three mismatch preambles were 8.01 and 9.81 for participants and items, respectively. Thus, the difference in the number of errors 46 following mismatch preambles with gee (8) and mismatch preambles with eyety (23) was reliable. However, the difference between preambles with o_ne and those with eaeb (16) just missed reliability, and the difference between mismatch preambles with eam and those with em was not reliable. The numbers of scores in the ambiguous category were fairly evenly distributed across the three quantifier types, and the analysis confirmed this as there were no significant effects (all Fs < 1.62). The analysis of the miscellaneous response category resulted in only a significant effect of local noun number (F1(1,31) = 22.68, F2(1,33) = 14.93). Consistent with the overall experimental analysis, there were more miscellaneous responses for quantified preambles with plural local nouns (33) than for quantified preambles with singular local nouns (7). Neither the main effect of quantifier type nor the interaction was significant (all Fs < 1.00). Discussion. The results of Experiment 1 revealed a large number of attraction errors for singular subject phrases that had plural local nouns. The most important finding was that when the subject noun phrase was marked as singular by a quantifier such as every, gel; or gag a following plural-marked local noun was significantly less likely to disrupt the implementation of correct singular verb agreement. Thus the results are consistent with the hypothesis that number is grammatically unspecified for the singular form of count nouns and that this is a contributing factor in errors of attraction. A post hoc examination of the differences among the quantifiers eac_h, m and gee suggested that the quantifier gee may be more effective at marking singular number than the other two. Specifically, there were fewer agreement errors following 47 preambles with the quantifier me than following preambles with either e_vgy or eac_h. Furthermore mismatch preambles that contained bee in the subject noun phrase had not only fewer attraction errors than those that contained eyegy or ea_cb, but they also had a greater number of correct agreement responses. However, because different preambles were associated with the different quantifiers in this experiment the differences in agreement patterns cannot be confidently attributed to just the properties of the quantifiers. Nevertheless, the indication of a decrease in the effectiveness of singular marking by eym and gee relative to that of o_ne may be due to the resulting notion of plurality; i.e., every/each postca_rd presupposes more than one postcard in the referent set. Despite the grammatical specification of singular number, this notional plurality may occasionally cause the use of plural agreement for these quantifiers. Alternatively the decrease in effectiveness of ev_ery and ea_ch to mark singular number may be due to the greater semantic complexity that results when the they combine with a noun which in turn may cause verb agreement to be less reliable for these quantifiers. So, to examine whether singular agreement is the accepted and consistent pattern for all three quantifiers a written validation test was administered. The test contained all the subject nouns that were employed in Experiment 1. They were presented as simple noun phrases (i.e., ones without postmodifying phrases) which were either preceded by their assigned quantifier or by the generic th_e. Two test lists were constructed. Each list contained all the nouns with half combined with a quantifier and the other half combined with the determiner me. A noun that was combined with a quantifier on one test list was combined with the determiner _t_h_e_ on the other. The construction of the lists was the same as in the main experiment. 48 Sixteen participants were recruited from the same pool as the main experiments. None of them participated in any of the other experiments that are reported. They were instructed to write a sentence completion for each noun phrase. The written responses were then scored according to whether an agreeing verb that was inflected for number was singular or plural. A total of 72 singular quantified noun phrases was presented in the validation test (36 noun phrases were the subject noun phrases of Experiment 1, and 36 were the local noun phrases of Experiment 2, to be discussed further below). All 366 number- inflected verbs that were produced to complete the singular quantified noun phrases were singular (64% of all the completions for the singular quantified phrases). All of the 404 number-inflected verbs that were produced to complete the corresponding singular generic tbe noun phrases were also singular (70% of all the completions for the singular generic noun phrases). Thus, although simple quantified noun phrases such as mgslggae or eagh x ture presuppose more than one sbgfl and p_i_ctgre in the referent set, they nevertheless elicited complete singular verb agreement in the validation tests. These results provide evidence that the accepted pattern of verb agreement for each, evety, and gee is singular. CHAPTER III Experiment 2 The results of Experiment 1 provided evidence that noun phrases containing one of the quantifiers gee, e_v_e_ry, or gn_e_ may be specified as singular versus those that contain the determiner the Experiment 2 examined the same question from a different perspective: If a quantified singular noun phrase is in a local position following a plural subject noun phrase, then the opposing number specification may interfere with the use of a correct plural verb form. This would yield an increase in singular attraction errors. The results of Experiment 1 also suggested that the quantifier gee may be more effective at marking or specifying singular number than ease or elegy. If this is the case, then the incidence of singular attraction errors should be the greatest following preambles that contain _o_ee. Participants. Eighty-seven native English speaking participants were recruited from the introductory'psychology classes at Michigan State University. They were given extra credit or credit that counted toward a course requirement in return for their participation. None had participated in the first experiment. Materials. The experimental preambles were modified from the preambles that were employed in Experiment 1. Each member of a set of preambles was identical except for 49 50 the determiner preceding the local noun (quantifier or th_e) and the number of the local noun (singular or plural). The subject noun was always plural. Table 6 provides examples for each of the three conditions. The complete list of preambles is in Appendix A Table 6. Examples of sentence preambles in Experiment 2. Local Noun Phrase Condition Sentence Preamble Control Plural (Match ) The slogans on the posters "the" Singular (Mismatch) The slogans on the poster Quantified Singular (Mismatch) The slogans on every poster The three items in each of the 36 sets represented the two experimental conditions and the control condition. The two experimental conditions were mismatch conditions that contained a plural subject noun that was followed by a singular local noun. The mismatch conditions differed in whether the singular local noun was combined with a quantifier or the determiner th_e . The control condition was designed to assess the incidence of non-attraction errors on plural verbs, and so contained both a plural subject and a plural local noun. Since each set of the preambles for this experiment was derived from a set employed in Experiment 1, if a set in Experiment 1 contained the quantifier gee in the subject noun phrase, then the corresponding set for this experiment contained me in the local noun phrase. This maintained an equal number of sets (i.e., 12) that represented each of the three quantifiers (eyeg, em and en_e_) in the local noun phrase. 51 The 64 filler items from Experiment 1 were also employed in this experiment. However, because all of the experimental items in this experiment contained a plural subject noun phrase, the fillers were adjusted so that 14 were plural and 50 were singular, otherwise they were the same as Experiment 1. Three lists were constructed in the manner described in Experiment 1. The procedure and scoring were also the same as in Experiment 1. Analysis. A one-way analysis of variance was performed on the data for each of the four scoring categories. The single factor in the analysis was the type of number specification of the local noun phrase and had three levels (i.e., preceded by a quantifier, by the determiner th_e, or marked as plural by a plural inflection). Planned comparisons were used to evaluate the difference between the two mismatch conditions and between each mismatch condition and the control condition. Results. Application of the scoring criteria yielded 2,333 correct agreement responses (74% of all the responses), 40 agreement errors ( 1%), 481 ambiguous responses (15%), and 278 miscellaneous responses (9%). The number of scores in each of the four scoring categories for the three conditions are in Table 7. 52 Table 7. Numbers of responses in each scoring category in Experiment 2. Response Type Local Noun Phrase Condition Correct Error Ambiguous Miscellaneous Control Plural 760 6 151 127 (Match) Generic "the" 81 1 11 134 88 (Mismatch) Quantified Singular 762 23 196 63 (Mismatch) The table shows that were fewer correct agreement responses and more singular attraction errors when a singular local noun phrase contained a singular quantifier than when it contained the generic tbe . This difference is clearly evident in the error proportions of the two mismatch conditions which, along with the error proportion for the match condition, are shown in Figure 2 . 53 0.04 0.03 ~ 0.02 J Error Proportion 0.01 J Match Mismatch Mismatch "the" Quantifier Preamble Conditions Figure 2. Proportions of agreement errors for the match (plural local noun) and mismatch (singular local noun) conditions. The match and mismatch conditions all contained a plural subject noun. The mismatch conditions either contained a singular quantified local noun or a singular unmarked noun preceded by the determiner 'the'. The one-way analysis of variance of the correct scores was significant for both participants and items (F1(2,172) = 3.86, F2(2,70) = 3.41). Planned pairwise comparisons were conducted on the three condition means in both participant and the item analyses in order to determine which of the differences were significant. The 95% confidence intervals for each comparison was 36.52 in the participant analysis and 39.33 in the item analysis. Thus, the comparisons showed that the large number of correct responses in the mismatch determiner tifi condition (811) differed significantly from both the number in the match condition (760) and the number in the mismatch quantifier condition (762). The difference between the number of correct 54 responses in the mismatch quantifier condition and the match control condition was not reliable. The analysis of the error scores also showed that there were significant differences among the three conditions (F1(2,172) = 4.24 and F2(2,70) = 5.10). Planned pairwise comparisons with 95% confidence intervals of 10.54 for participants and 9.73 for items confirmed that the number of singular attraction errors following quantified local noun phrases (23) was reliably greater than the number following the tbe local noun phrases (11). The difference in the number of errors following the quantified preambles and the match control preambles (6) was also reliable, but the difference in the number of errors following the control match and mismatch tee preambles was not. The one-way analysis of variance of the ambiguous scores was also significant (F1(2,172) = 8.88, and F2(2,70) = 6.08). Unplanned comparisons in the participant analysis showed that the number of ambiguous responses in the mismatch quantified condition (196) was significantly greater than the number in the match condition (151). The 95% confidence interval for each of the participant comparisons was 37.76. The difference between these two conditions was marginally reliable in the item analysis which had a greater confidence interval of 46.19. The large number of ambiguous responses in the quantified mismatch condition was also reliably greater than the number in the mismatch tbe condition (134) in both analyses, and the number of ambiguous responses in the mismatch m condition and the match condition did not differ significantly from each other. There were also significant differences in the number of miscellaneous responses for the three conditions (F1 (2,86) = 12.19, Fz(2,70) = 9.75). Post hoc comparisons with 95% confidence intervals of 32.46 for participants and 36.72 for items revealed that the number in the match control condition (127) was significantly greater than 55 both the number in the mismatch the condition (88) and the number in the mismatch quantifier condition (63). The difference between the two mismatch conditions was not reliable in either the participant or item analysis. Appendix 8 contains examples of the miscellaneous responses as well as the number of responses that occurred for the reasons for the assignment of this score. The majority of the miscellaneous scores (223) were due to preamble repetition failures. As in the first experiment, the most common failure was a change in the number of either the subject or local nouns of the presented preambles (151 miscellaneous responses). The most frequent direction of the change (118) was from plural to singular. This direction may be due to a greater difficulty in the perception of the plural morphemes in the presented preambles and/or to a greater difficulty in the production of them during the repetition (resulting in a bound morpheme loss error). Since the match preamble condition had two plural nouns that could experience these difficulties, it is not surprising that this condition had more repetition failures than either of the mismatch conditions. Seventy-one of the 127 miscellaneous match condition responses were due to a number change in the repetition compared to 24 of the 88 mismatch _t_he responses and 11 of the 63 quantified mismatch responses. Thus, this particular repetition failure not only caused the large number of miscellaneous responses in the match condition, but it also appears to be the reason for this condition's low number of correct agreement responses. As in Experiment 1, verb agreement in the miscellaneous responses was examined. 0f the 278 miscellaneous responses, 256 contained complex subject phrases. Ninety-six of these had a match in number between the subject and local nouns (either both singular or both plural), and 160 had a mismatch in number (either singular subject-plural local noun or plural subject-singular local noun). The verbs following these subject phrases were assigned an agreement score (correct, error, or ambiguous) 56 with respect to the number of the actually produced subject noun phrase. The results of this scoring are given in Table 8. Table 8. Number of miscellaneous responses in each agreement score category. The total percentages are given in parentheses and are based on all 256 responses. Agreement Type Correct Error Ambiguous Subiect 8: Local Match smfit pbrags Noun Number "the" Local Noun Plural Subject & Local 46 10 Singular Subject & Local 5 14 5 Quantified Local Noun Singular Subject & Local 6 9 1 TOTAL 57 (22%) 23 (9%) 16 (6%L Mismatch subject mrases "the" Local Noun Plural Subject/Singular Local 82 3 25 Singular Subject/Plural Local 1 10 4 Quantified Local Noun Plural Subject/ Singular Local 27 8 TOTAL 110 (43%) 13 (5%) 37 (15%L The numbers in Table 8 show the opposite pattern of agreement from that of the main experiment: There were fewer errors and more correct responses following mismatch subject phrases than following match subject phrases. Beginning with the opposite error pattern, all 23 of the match errors resulted in a plural-marked verb 57 that was preceded by a subject phrase in the repetition that contained a singular subject noun. Likewise, ten of the mismatch errors resulted in a plural-marked verb that was preceded by a singular subject. Since all of the presented experimental preambles contained a plural subject noun, these verbs correctly agreed with the presented preambles. It is impossible to determine whether the produced singular subject nouns were misperceptions of the presented plural subject nouns, and therefore were intended, or were production errors. If some of these repetitions contain production errors that resulted in the loss of the plural morpheme on the subject nouns, then the plural verbs would be in agreement with intended plural subject nouns. Thus, the opposite pattern of errors in the miscellaneous responses may reflect processing failures that differ from those underlying the error pattern in the main experiment. The three errors that followed the mismatch subject phrases containing a plural subject and singular local noun appear to be genuine agreement errors. The question is whether they are attraction errors. Since the actually presented preambles for these three repetition responses were from the match condition, the repetition responses changed the presented local noun from plural to singular. As in the previous discussion, it is impossible to determine whether the produced singular noun was a perception or production error. If the singular local noun in any of these three responses was intended (e.g., as a result of a perception error), then a singular attraction error occurred. If, however, a plural local noun was intended, but a singular local noun was erroneously produced, then an agreement error that is not an error of attraction occurred. Despite the indeterminate reason for the repetition failure, adding these three agreement errors to the number in either the match or mismatch tbe conditions of the main experiment does not change the overall pattern: The quantified mismatch condition still has the greatest number of agreement errors. 58 The greater incidence of correct agreement among the mismatch subject phrases than the match subject phrases appears to also be a consequence of the large number of repetition failures in the match condition that changed one of the nouns in the presented preamble from plural to singular. Fifty-seven of the 82 plural subject-singular local noun subject phrases that had correct agreement scores were repetition failures of presented match preambles. If some number of these failures are production errors then the intended preamble would have been a match preamble. Thus, as in the case of the error pattern, the pattern of correct agreement among the miscellaneous responses may also reflect processing failures that are not evident in the main experiment results. Quantifier type examination. The numbers of scores in each score category for the three quantifier types (gm, each, and eveer) employed in the mismatch quantifier condition are given in Table 9. Table 9. Number of responses in each scoring category for the three quantifier types in Experiment 2. Response Type Local Noun Phrase Correct Error Ambiguous Miscellaneous Quantifier Type 0'8 234 1 3 77 24 Each 261 3 67 1 7 Every 267 7 52 22 When the quantifier me was in the local noun phrase, there were fewer correct agreement responses and more agreement errors relative to when the quantifiers each 59 and eym were in the local noun phrase. A one-way analysis of variance was conducted on each of the four response score categories with quantifier type as a single factor. The factor of quantifier type was a within-participants factor and a between-items factor in the respective analyses. The analysis of the correct scores was significant for participants (F1(2,172) = 3.91) but not for items (F2(2,33) = 1.64, p > .05). Post hoc comparisons in the participant analysis showed that only the difference between the number of correct agreement responses for preambles with gee (234) and eyety (267) was reliable (the 95% confidence interval was 31.23). The one-way analysis of variance of the error scores was also significant in both the participant and item analyses (F1(2,172) = 3.73, F2(2,33) = 3.47). Post hoc comparisons in the participant analysis, with a 95% confidence interval of 9.14, revealed that only the difference between the 13 attraction errors following preambles with gee and the three attraction errors following preambles with eaeh was significant. None of the post hoc comparisons in the item analysis were reliable as the conservative 95% confidence interval was much greater at 1 7.08. The differences among the numbers of ambiguous scores for the three quantifiers was only marginally significant in the analysis with participants random (F1(2,172) = 2.59, p = .08), but not significant with items random (F2< 1). The analysis of the miscellaneous scores was not significant in either the participant or item analyses (both Fs<1). Discussion. The error results of Experiment 2 demonstrated that a singular local noun phrase that is grammatically marked as singular by a quantifier is significantly more 60 likely to elicit an error of attraction than a singular unmarked local noun phrase. This offers additional support for the claim that singular count nouns are grammatically unspecified for number. When they combine with a singular quantifier such as eagb, elegy, or me the resulting noun phrase becomes grammatically marked or specified as singular. When this grammatically specified noun phrase is a local noun phrase in a sentence preamble which contains a plural subject noun phrase, the singular specification of the local noun may interfere with the correct implementation of plural verb-agreement. An examination of the differences between the three quantifier types in their ability to attract an agreement error offered some indication that the quantifier gee may be more attractive than the quantifiers m and fiery. The preambles that contained a local noun quantified by gee had the fewest number of correct agreement responses and the greatest number of attraction errors. These findings are consistent with those of Experiment 1 which showed that the quantifier _(fl was somewhat more effective than the other two in reducing attraction errors when it was in the subject noun phrase. However, as in the first experiment, because different preambles were associated with the different quantifiers caution must be observed in drawing conclusions based on these results. The results will nonetheless be considered further in the General Discussion. CHAPTER IV Experiment 3 The results of the first two experiments support the proposal that a singular quantifier specifies an otherwise unmarked noun phrase as singular. However, there is another plausible interpretation of the results: The singular quantifier may merely enhance or strengthen a weak singular specification on an unmarked count noun. If the effect of a quantifier is to enhance the number specification of a following noun then a plural quantifier (e.g., many, several, a few) should enhance the plural specification of a following plural marked noun. This proposal predicts that attraction errors should be more frequent following mismatch preambles that contain both a plural local noun and a plural quantifier than following mismatch preambles that contain only a plural local noun and the determiner tea. Participants. Twenty-four participants were recruited for this experiment from the same source as in the previous two experiments. Materials. The 36 sets of preambles employed in the first two experiments were employed in this experiment. Each of the 36 sets contained three preambles to represent three conditions. The three conditions are shown in Table 10. 61 62 Table 10. Examples of sentence preambles in Experiment 3. Local Noun Phrase Condition Sentence Preamble Singular Control (Match) The slogan on the poster Generic "the" Plural (Mismatch) The slogan on the posters Qtrantified Plural (Mismatch) The slogan on several posters The three items in each of the 36 sets represented the two experimental conditions and the control condition. The two experimental conditions were mismatch conditions that contained preambles with singular subjects that were followed by plural local nouns. The mismatch conditions differed in whether the plural local noun had a quantifier or the determiner me. The control preambles contained singular subject and local nouns. The plural quantifiers that were employed in this experiment were m_aey, several and a fee, which may only occur with plural count nouns (Quirk, Greenbaum, Leech, & Svartvik, 1985). As in the previous experiments, each set of preambles contained only one of these quantifiers in the mismatch quantifier condition. The number of sets that employed each of these three types of quantifiers was equal. The fillers were designed as in the previous experiments, and the list constnrction also followed the methods described earlier. Analysis. A one-way analysis of variance was performed on the data for each of the four scoring categories. The single factor in this analysis was the type of number specification of the local noun phrase. It had three levels, unmarked singular, plural 63 marked, and plural marked plus plural quantified. Planned comparisons were used to evaluate the differences between the two mismatch conditions and between each mismatch condition and the control condition. Results. The number of scores for each condition in each scoring category are presented in Table 11. Figure 3 contains the net error proportions for the two mismatch conditions. Table 11. Numbers of responses in each scoring category in Experiment 3. Regronse Type Local Noun Phrase Correct Error Ambiguous Miscellaneous Condition Singular Control (Match) 214 1 67 6 "the" Plural (Mismatch) 123 67 67 31 Quantifier Plural 152 57 54 25 (Mismatch) There was a total of 489 correct agreement responses (57% of all responses), 125 agreement errors (14% ), 188 ambiguous responses (22% ) and 62 miscellaneous responses (7% ). Table 11 shows that there were more correct responses and fewer errors following match preambles than following mismatch preambles. Furthermore, the 64 error proportions in Figure 3 show that within the two mismatch conditions, local noun phrases containing quantifiers did not attract more agreement errors than local noun phrases containing th_e. 0.4 Error Proportion Match Mismatch Mismatch "the" Quantifier Preamble Conditions Figure 3. Proportions of agreement errors for the match (singular local noun) and mismatch (plural local noun) conditions. The match and mismatch conditions all contained unmarked singular subject nouns. The mismatch conditions contained a local noun that was plural and was either preceded by a plural quantifier or by the determiner 'the'. The one-way analysis of variance of the correct scores confirmed that there were significant differences among the conditions (F1(2,46) = 25.79, Fz(2,70) = 25.74). Planned pairwise comparisons were conducted with 95% confidence intervals of 23.29 and 23.03 in the participant and item analyses, respectively. The comparisons showed that the large number of correct responses in the match condition (214) reliably 65 differed from the number in both the mismatch tm condition (123) and the number in the mismatch quantifier condition (152). Furthermore, there were significantly more correct responses in the mismatch quantified condition than in the mismatch t3 condition. The one-way analysis of variance of the errors was also significant with both participants and items as random factors (F1(2,46) = 14.62), F2(2,70) = 24.84). The 95% confidence intervals for the pairwise comparisons were 23.67 for the participant analysis and 17.95 for the item analysis. Thus, the number of errors in the mismatch conditions differed reliably from the one error that in the match control condition; however, the difference in the number of attraction errors in the quantified and the unquantified mismatch conditions was not reliable (57 and 67, respectively). There were no significant differences in the numbers of ambiguous responses across the three conditions (both Fs< 1.17). However, the overall analysis of the miscellaneous responses revealed that there were significant differences (F1(2,46) = 10.05, F2(2,70) = 6.07). Post hoc comparisons with 95% confidence intervals of 14.81 and 18.82 for participants and items, respectively, revealed that the number of miscellaneous responses in the match control condition (6) differed reliably from the number in the mismatch quantified condition (25). The match control condition also differed reliably from the mismatch tee condition (31 ). However, the difference between the two mismatch conditions was not reliable. As in the previous two experiments, examples of miscellaneous responses are given in Appendix B. Again, the most frequent reason (31) for the assignment of a miscellaneous score was due to a change in the number of either the subject or local noun of the presented preamble. Verb-agreement in the miscellaneous responses of this experiment was also examined. Fifty-four of the 62 responses contained complex subject phrases that had 66 either a match or mismatch in number between the subject and local nouns. The verb- agreement in these 54 responses was scored with respect to the number of the actually produced subject noun phrase, and the results of this scoring are given in Table 12. Table 12. The number of miscellaneous responses in each agreement score category. The percentages given in parentheses are based on a total of 54 responses. Aggement Type Correct Error Ambiguous Subiect & Local Match sgbje' ct mrases Noun Number "the" Local Noun Plural Subject & Local 4 1 2 Singular Subject & Local 14 5 Quantified Local Noun Plural Subject & Local 3 1 2 TOTAL 21 (39%) 2 (4%) 9 (17%) Mismatch subject phrases "the" Local Noun Plural Subject/Singular Local 1 1 Singular Subject/Plural Local 3 3 Quantified Local Noun Singular Subject/Plural Local 8 2 4 TOTAL 12 (22%) 345%) 7 (13%) 67 Table 12 shows that there were more correct agreement responses and fewer agreement errors in the miscellaneous responses containing match subject phrases than in the responses containing mismatch subject phrases. Thus, the agreement pattern in the miscellaneous responses was consistent with the pattern in the main experiment. Quantifier examination. Table 13 gives the numbers of scores for each response type for each of the three quantifiers that were employed in the mismatch quantifier condition. Table 13. Number of responses in each scoring category for the three quantifiers in Experiment 3. Response Type Local Noun Phrase Correct Error Ambiguous Miscellaneous Quantifier Type A few 44 21 25 6 Many 55 21 12 8 Several 53 15 17 11 A one-way analysis of variance was conducted on the scores in each of the response categories with the single factor of quantifier type. The fairly even distribution of scores across the three quantifiers in each response category was confirmed statistically. Only the analysis of the ambiguous scores showed that there was a difference. The analysis of the ambiguous scores was significant with participants as a random factor (F1(2,26) = 3.12) but not with items as a random factor (F2(2,33) = 1.48). Post hoc comparisons in the participant analysis with a conservative 95% 68 confidence interval of 13.36 revealed that only the greater number of ambiguous responses in the condition with a few (25) than in the condition with many (12) was marginally reliable. The analyses of the other three score categories yielded Fs < 1.77. Discussion. The results of Experiment 3 offered additional support for the interpretation of the findings in the first two experiments. In particular, the results of Experiments 1 and 2 suggest that the singular form of count nouns is grammatically unmarked or unspecified for number. When an unmarked count noun combines with a singular quantifier the resulting noun phrase becomes grammatically specified as singular. The results of Experiment 3 suggest that the quantifier effect is not one of enhancing the specification of a following noun, but rather one of actual specification. If the effect were mere enhancement one would have expected an increase in the number of attraction errors following mismatch preambles that contained both a plural quantifier and a plural local noun relative to mismatch preambles that contained a plural local noun with the determiner tee. However, the error results showed that mismatch quantified preambles did not increase the likelihood of attraction errors. Although the evidence of the first three experiments suggests that the combination of a quantifier with an unmarked noun creates a noun phrase that is grammatically marked or specified for number, it does not mean that the noun itself becomes grammatically specified. It was proposed in the Introduction that the two noun classes that are referred to as mass and count may provide a test of whether a singular noun, itself, may carry explicit singular number specification. CHAPTER V Experiment 4 Experiment 4 examined whether mass nouns contain an underlying marked specification of singular number by examining their ability to prevent errors of attraction when they are the subject noun of a subject phrase that contains a plural local noun. A reduction in the incidence of errors for subject phrases that contain mass nouns relative to those that contain count nouns would support the claim that singular count nouns contain an unmarked number specification. Participants. Fifty-two participants were recruited in the same manner as described in the previous three experiments. Materials. Twenty sets of experimental preambles each containing four preambles were constructed to represent the four conditions. The conditions and examples of the preambles are shown in Table 14. 69 70 Table 14. Examples of sentence preambles in Experiment 4. Local Noun Phrase Number Sentence Preamble Singular Count Subiect Noun Singular (Match) The eraser for the blackboard Plural (Mismatch) The eraser for the blackboards Singular Mass Subject Noun Singular (Match) The chalk for the blackboard Plural (Mismatch) The chalk for the blackboards As in the previous experiments each preamble consisted of a complex subject noun phrase in which the subject was followed by a prepositional phrase. Each member of a set of preambles was identical except for the subject noun (mass or unmarked count) and the number of the local noun (singular or plural). The twenty mass nouns were selected from a dictionary of mass terms (Sparkes, 1985). Each mass noun was matched with a count noun that denotes an entity that is related in meaning. For example, the count noun bar was matched with the mass noun metre, Only mass nouns that do not readily undergo a reclassification as a count noun were employed. To enforce this, none of the selected mass nouns had a plural form listed in Francis and Kucera's (1982) word frequency norms. In addition, the mass nouns and their corresponding meaning-related count nouns were subjected to Allan's (1980) tests of countability. Allan constructed a set of four grammaticality judgment tests that are designed to determine the countability preferences of English nouns. A judgment of "unacceptable" in three of the tests suggests that the noun is not countable, whereas a judgment of 71 "acceptable" suggests that it is. These three tests include the following environments: ( 1) an environment in which a noun takes plural verb and/or pronoun agreement (e.g., *traffic are dangerous, *aren't they? versus cars are dangerous, aren't they?) (2) an environment in which a noun combines with a unit denumerator such as "a(n)" or "one" (e.g., *a/*one traffic versus a/one car ) and (3) an environment in which a noun combines with a plural quantifier such as "several" or "a few " (e.g., *several traffic versus several cars). The fourth countability test contains an environment in which a judgment of "acceptable" suggests that the noun is not countable, and a judgment of "unacceptable" suggests that it is. The environment is one in which the combination of the noun and the determiner "all" results in a genus-referring noun phrase that is singular (e.g., All traffic is dangerous versus *All car is dangerous). All twenty of the selected mass nouns were shown to be uncountable by all four of Allan's tests, and all twenty of the corresponding meaning-related count nouns were shown to be countable. Each of the twenty preamble sets consisted of two preambles that contained a mass subject noun and two that contained the corresponding meaning-related count noun in the singular form. One mass noun preamble and one count noun preamble each represented a mismatch condition. Both mismatch preambles contained a singular subject noun that was followed by a plural local noun. The other mass and count noun preambles represented match control conditions and contained singular subject nouns , followed by singular local nouns. In addition to the experimental preambles, filler items that consisted of simple noun phrases were created. Four lists were constructed with each list containing all the filler items and one preamble from each of the four experimental condition sets. The fillers were designed so that the number of singular and plural items within each list 72 was equal. Each list contained an equal number of preambles that represented each of the four conditions. The same Procedure and Scoring categories employed in the first three experiments were also employed in Experiment 4. Analysis. An analysis of variance for each of the score categories was performed first with participants as a random factor and then with items as a random factor. Each analysis contained the within participant variables of subject noun phrase type (mass or count) and local noun number (singular or plural). Results. Application of the scoring criteria yielded 695 correct agreement responses (67% of all responses), 88 agreement errors (8%), 181 ambiguous responses (17% ) and 76 miscellaneous responses (7% ). Table 15 contains the numbers of scores in each score category for each preamble condition. 73 Table 15. Numbers of responses in each scoring category in Experiment 4. Response Type Local Noun Phrase Number Correct Error Ambiguous Miscellaneous _Seigglar Count Seteect Noun Singular (Match) 200 2 46 12 Plural (Mismatch) 140 51 39 3O Singular Mass Subiect Noun Singular (Match) 192 6 48 14 Plural (Mismatch) 163 29 48 20 An examination of the numbers of correct responses in Table 15 shows that there were fewer correct agreement responses following mismatch preambles than following match preambles. The table also shows that the distribution of agreement errors was complementary to that of the correct agreement responses; i.e., there were more errors following mismatch preambles than following match preambles. Figure 4 contains the net error proportions for the two mismatch preamble conditions. The figure clearly shows that the incidence of plural attraction errors was reduced when the preambles contained a mass subject noun relative to when they contained a count subject noun. 74 0.4 Net Error Proportion M 355 Count Subject Noun Type Figure 4. A graph of the net error proportions for the two mismatch conditions. The net error proportions were obtained by subtracting the error proportions of the match conditions from the error proportions of the corresponding mismatch conditions. An analysis of variance of the correct agreement responses yielded a significant main effect of local noun phrase number in both the participant and item analyses (F1(1,51) = 25.67, F2(1,19) = 30.97). This main effect confirmed that there was a greater number of correct agreement responses in the match control conditions than in the mismatch conditions (392 versus 303, respectively). The main effect of subject noun type (mass or count) was not significant in either analysis (F1(1,51) = 1.11, F2(1,19) = .49), which indicated that the overall number of correct agreement responses following count noun preambles did not differ significantly from the overall number following mass noun preambles (340 versus 355, respectively). However, in the participant analysis, there was a significant interaction between the factors of 75 subject noun type and local noun phrase number (F 1(1 ,51) = 5.11). Planned pairwise comparisons in the participant analysis with a 95% confidence interval of 19.59 revealed that the greater number of correct agreement responses in the mass mismatch condition (163) than in the count mismatch condition (140) was reliable. The interaction between subject noun type and local noun phrase number was not significant in the item analysis (F2(1,19) = 2.04, p > .10). The complementary distribution of the errors relative to the correct responses was confirmed statistically. The main effect of local noun phrase number (F1 (1 ,51) = 34.75, F2(1,19) = 34.98) was significant in the error analysis. This reflected the greater number of agreement errors in the mismatch conditions than in the match control conditions (80 versus 8, respectively). The main effect of subject noun type was also significant in the participant analysis (F 1(1 ,51) = 6.14) but not in the item analysis (F2(1,19) = 2.20). This effect reflected the greater number of attraction errors following preambles that contained subject count nouns than following preambles that contained subject mass nouns (53 versus 35, respectively). The interaction between the factors of local noun phrase number and subject noun type was significant in both analyses (F1(1,51) = 13.53, F2(1,19) = 4.65). Planned pairwise comparisons showed that the interaction resulted from the large reduction in the number of attraction errors in the mass mismatch condition (29) relative to the count mismatch condition (51 ). The difference between these two conditions was reliable in both the participant and item analyses (the 95% confidence intervals were 10.10 and 17.84, respectively). As evident in Table 15, the numbers of ambiguous responses were fairly evenly distributed across the four conditions. This even distribution was confirmed in the analyses of the ambiguous responses as there were no significant effects or interactions (all Fs < 1). 76 The analysis of the miscellaneous responses resulted in a significant main effect of local noun phrase number (F1(1,51) = 7.54, F2(1,19) = 6.14). There were more miscellaneous responses in the mismatch conditions than in the match control conditions (50 versus 26, respectively). Neither the main effect of subject noun type nor the interaction between subject noun type and local noun phrase number was significant in either the participant or item analysis (all Fs < 2.33). Examples of the miscellaneous responses are given in Appendix B. As in the previous three experiments, the majority (55) were due to preamble repetition failures, with the most frequent failure (30) being a change in the number of the local noun in a presented mismatch preamble from plural to singular. Thus, the significantly greater number of miscellaneous responses in the mismatch conditions than in the match conditions was due to this particular failure. As discussed in the previous experiments, these "number-change" failures may reflect either a perception or a production error. Verb agreement in the 68 miscellaneous responses containing complex subject phrases was examined. Fifty-two had subject and local nouns that matched in number and 16 had subject and local nouns that had a mismatch in number. The greater representation of match subject phrases in the miscellaneous responses resulted from the frequent number-change repetition failures of the presented mismatch preambles. The verb agreement in these responses was scored with respect to the number of the actually produced subject noun. Table 16 presents the results of this scoring. 77 Table 16. Numbers of miscellaneous responses containing complex subject phrases in each verb-agreement score category. The percentages given in parentheses are based on a total of 68 responses. Aggement Type Correct Error Ambjmous Subiect & Local Match Subiect Phrases Noun Number Count Subject Nouns Plural Subject & Local 7 2 1 Singular Subject 8: Local 17 2 6 Mass Subject Nouns Singular Subject & Local 10 2 5 TOTAL 34 (50%) 6 (9%L 12 (17%) Mismatch Subject Phrases Count Subject Nouns Singular Subject/Plural Local 5 1 Mass Subject Nouns Singular Subject/Plural Local 7 1 2 TOTAL 12 (17%) 2 (3%) 2 (3%) Table 16 shows that there were more correct agreement responses following match subject phrases than following mismatch subject phrases, which is consistent with the findings in the main experiment. However, the distribution of errors was not consistent with the main experiment. There were more errors following match subject phrases than following mismatch subject phrases. 78 Two of the six errors in the match responses resulted in singular-marked verbs that disagreed with a produced plural subject noun. However, since all of the experimental preambles contained singular subject nouns, the singular-marked verbs agreed with the subject nouns of the presented preambles. Because it is impossible to determine whether the produced plural subject noun was a perception error of the presented preamble's subject noun or a production error, the apparent erroneous agreement cannot be confidently attributed to the same processing failures that are proposed to underlie the agreement errors in the main experiment. Because the remaining four of the six errors in the match subject phrase responses resulted in plural-marked verbs that disagreed with both the presented and produced singular subject nouns, these errors appear to be agreement errors. All four of the subject phrases of these responses resulted in a change in the number of a presented preamble's local noun from plural to singular. Again, it is impossible to determine whether this change is a result of a perception of production error. If the change is due to a production error, then the plural-marked verb may be incorrectly agreeing with an intended plural local noun. In other words, some or all of these four match phrase errors may be attributable to the same processing failures that are assumed to underiie errors of attraction for mismatch phrases. Assuming that the four plural-marked agreement errors are attraction errors, adding them to the errors in the respective singular subject-plural local noun conditions of the mismatch phrases results in a total of three errors following mismatch phrases with subject counts nouns and three following mismatch phrases with subject mass nouns. The equal number of errors for these two mismatch conditions is not consistent with the findings of the main experiment. However, adding these errors to the respective mismatch conditions in the main experiment does not change the error pattern in the main experiment. 79 Discussion. The error results of Experiment 4 provide additional evidence for the claim that the singular form of a count noun is grammatically unmarked or unspecified for number. Furthermore, the results suggest that, unlike singular count nouns, mass nouns may contain explicitly specified singular number information. These claims are based on the findings that when the subject noun of a sentence preamble was a singular unmarked count noun followed by a plural local noun, the incidence of attraction errors was significantly greater than when the subject noun of the preamble was a mass noun. Thus, one consequence of the unspecification of number for singular count nouns is the greater likelihood that the grammatical specification of plural number by a following local noun will disrupt the implementation of singular verb agreement. In contrast, the plural specification of a local noun appears to be less disruptive of the use of correct singular verb agreement with subject mass nouns. CHAPTER V General Discussion The results of all four experiments support the proposal that there is a markedness relation in the underlying specification of grammatical number for singular and plural count nouns. Singular nouns appear to be unmarked or unspecified for number whereas plural nouns appear to be marked or explicitly specified. This difference in the underlying markings of grammatical number is manifested in the asymmetric distribution of attraction errors. The asymmetry, which shows a greater incidence of agreement errors following subject phrases with a singular subject-plural local noun number pattern than following subject phrases with the reverse number pattern, appears to arise from the marked specification of a plural local noun dominating or overshadowing the unmarked specification of a singular subject noun. Experiment 1 showed that the dominance of plural number specification on a local noun can be significantly weakened by a subject noun phrase that is grammatically marked or specified as singular by an overt singular morpheme, i.e., a quantifier. In addition, the results of Experiment 2 demonstrated that when a local noun phrase is grammatically marked as singular by a quantifier, this explicit specification is able to overshadow the contrasting grammatical specification of plural number on a subject noun phrase, thereby increasing the likelihood of a singular attraction error relative to when there is no singular marking on a local noun phrase. Experiment 3 offered evidence that singular quantifiers do not merely enhance a weak specification or marking of singular number, but rather they mark an otherwise unspecification. 80 81 Finally, the results of Experiment 4 showed that mass nouns, unlike singular count nouns, appear to contain an underlying grammatical specification that is equivalent to a marked specification of singular number. Although the overall results demonstrate that singular number on a subject noun phrase can become grammatically marked or specified and therefore reduce the likelihood of an attraction error by a plural marked local noun, there is still an indication that plural number marking is more attractive or dominant than singular number marking. This indication comes from a comparison of the incidence of singular versus plural attraction errors following mismatch preambles that had contrasting number marking on the subject and local noun phrases. Experiment 2 was the only one to employ mismatch preambles in which the subject noun phrases were marked as plural and the local noun phrases were marked as singular (by a quantifier). The incidence of singular attraction errors, i.e., the net error proportion for this condition, was .02 (see Figure 2). Both Experiments 1 and 4 employed mismatch preambles in which the subject noun phrases were marked as singular (by either a quantifier or a mass noun) and the local noun phrases were marked as plural. The incidence of plural attraction errors, i.e., the net error proportions for these conditions, was .20 in Experiment 1 (see Figure 1) and .12 in Experiment 4 (see Figure 4). This comparison suggests that plural marking is more capable of overshadowing singular marking than the reverse. The difference in attractiveness may be attributable to the nature of the marking of singular versus plural number. In Experiments 1 and 2, singular noun phrases were grammatically marked by the unbound morphemes eagb, eyety, and gee. In contrast, plural noun phrases were grammatically marked by the bound morpheme s Thus, plural noun phrases received their number specification from their head nouns, whereas singular noun phrases received their number specification from a non-head 82 constituent, i.e., the quantifier. Therefore, in these cases, plural specification may be more attractive than singular specification (or less prone to being overshadowed) because the specification of plural number was at both the level of the noun phrase and the level of the head subject noun. In contrast, the specification of singular number was only at the level of the noun phrase. Unlike noun phrases specified as singular by quantifiers, noun phrases specified as singular by mass nouns should receive their specification from the head subject noun, itself. Thus given the above analysis, the specification on the subject noun should be more effective at reducing the incidence of attraction errors by plural local nouns. This appears to be the case in a post hoc comparison of the net error proportion for the mismatch condition with quantified subject noun phrases in Experiment 1 (.20) versus the net error proportion for the mismatch condition with mass subject noun phrases in Experiment 4 (.12). However because there are many differences between these two experiments this comparison should be treated cautiously. It is up to future research to examine this issue in a more controlled manner as well as to examine whether local noun phrases specified as singular by mass nouns are as effective at eliciting attraction errors as local noun phrases specified as plural by plural count nouns. Along related lines, the results of Experiments 1 and 2 indicated that there is a difference in the effectiveness of the three quantifiers me, exery, and gm to mark singular number. In Experiment 1, the results indicated that the singular quantifier gee was more effective than the other two at reducing the incidence of attraction errors by a plural local noun. In Experiment 2, the results indicated that the quantifier gee was more effective than the other two at attracting erroneous verb agreement. Because all three quantifiers obligatoriiy require singular verb agreement (the consistent use of this agreement was verified by the results of a validation test in Experiment 1) the difference does not appear to arise at the level of grammatical number specification. 83 However there is a level at which the quantifier m differs from the quantifiers ease and fiery, namely the semantic level. As previously discussed, the combination of a noun with either the quantifiers ease or eyefl results in a noun phrase whose referent set presupposes more than one entity. The combination of the quantifier gee with a noun, however, not only explicitly specifies number as singular at the grammatical level but it also explicitly specifies number as singular at the semantic level. The duality in the usage of the unmarked meaning of singular number was illustrated in the Introduction with the phrases geem and man is an animal where it was argued that the singular meaning of the first phrase reflects the specific opposite of the plural meaning of a phrase such as many men . The meaning in the second phrase was argued to reflect a neutral unmarked meaning of singular number. Thus, the apparent greater effectiveness of the quantifier gee to mark singular number may be due to its consistent marking of number at both the semantic and grammatical levels. This suggests that the notional markings of number may have some influence on the verb agreement process. This might appear to be inconsistent with the evidence from previous investigations of attraction errors which has shown that verb agreement is more sensitive to grammatical number than semantic or notional number (Bock and Miller, 1991; Bock and Eberhard 1993). However, the claims proposed here as well as previously do not assert that the verb agreement processes are completely blind to notional number. In fact recent evidence by Bock, Eberhard, and Cutting (1992) shows that the notional plurality of singular collective subject nouns (e.g., The choir at the mass) can affect the implementation of verb agreement. Thus, the claim is that when there is a conflict between notional number and grammatical number the verb agreement process appears to favor grammatical number (Bock and Eberhard, 1993). In the case of noun phrases with the quantifier _o_ne the consistent marking of number 84 information at both the semantic and grammatical level may result in more secure retrieval of an agreeing verb. The issue of "security" was also raised in the Introduction where it was argued that the unspecification of singular nouns should show evidence of less secure verb agreement even under circumstances where there is no plural specification to disrupt the agreement process. This is because the agreement mechanism is assumed to retrieve a singular agreeing verb form by default rather than on the basis of explicit number information. The match control conditions of Experiments 1 and 4 permit an examination of this issue. Both Experiments employed two match conditions. In all four conditions the subject and local noun phrases were singular; however, in two of the four conditions the subject noun phrases were marked as singular by either a quantifier (Experiment 1) or a mass subject noun (Experiment 4). The prediction, therefore, is that the match conditions that contained singular number specification on the subject noun phrases should show evidence of more secure verb agreement than those that contained unspecified subject noun phrases. Neither Experiment, however, offers support for this prediction. In Experiment 1, there were more correct agreement responses and fewer agreement enors following match preambles that had unmarked subject nouns than following match preambles that had subject nouns marked as singular by a quantifier. Similarly in Experiment 4, there were more correct agreement responses and fewer agreement errors following match preambles that had singular subject count nouns than following match preambles that had mass subject nouns. Although the pattern of agreement is opposite to the predictions based on the security of marking, it should be noted that the pattern arises from small differences in small numbers, none of which were significant. 85 In contrast to the small number of errors in the match condition, there were large numbers of agreement errors in the mismatch conditions in these four experiments. In fact, the overall error rates (total number of agreement errors divided by total number of number-inflected responses) in Experiments 1, 3, and 4 were somewhat larger than the overall error rates for previously reported experiments. The overall error rates in the experiments reported by Bock and Miller (1991) range from .04 to .08. Bock and Cutting's (1992) experiments had overall error rates ranging from .03 to .05, and Bock and Eberhard (1993) have slightly higher overall error rates ranging from .04 to .14. In comparison, the overall error rates for Experiments 1, 3, and 4 reported here are .15, .20 and .13, respectively. One possible reason for the larger error rates in these experiments may be the small number of experimental preambles that contained animate nouns. Bock and Miller (1991) found that the copula tgbe is used more often when subject nouns are inanimate than when they are animate. Because this irregular verb inflects for number both in the present and past tenses there are more opportunities for agreement errors to occur. None of the subject nouns in any of the four experiments reported here were animate and very few of the local nouns were. Thus the predominance of inanimate nouns in these experiments may have elicited the greater use of tgbe as a copula or auxiliary which led to an increase in the error rates. Although the vast majority of number-inflected responses, both correct and incorrect, were forms of the irregular verb tele there were responses that involved number-inflected regular verb forms. The error rate for the regular verb forms did not differ dramatically from the error rate for the irregular verb forms. This is consistent with previous investigations (Bock and Eberhard, 1993; Bock and Miller, 1991) and suggests that the verb agreement patterns are not particular to the retrieval of irregular verb forms. Across the four experiments there were 214 responses that 86 had number inflected regular verb forms. Of these responses 25 were errors (12%). In comparison, there were 4,374 responses that had number inflected irregular verb forms, 351 of which were errors (8%). Similar examinations by Bock and Eberhard and Bock and Miller found error rates for regular verb forms to be 19% and 5%, respectively, and error rates for irregular verb forms to be 9% and 8% ,respectively. Thus, the error rates for irregular verbs are quite consistent across these independent studies, and the error rate for regular verbs in this study is the average of the rates found in Bock and Eberhard's and Bock and Miller's studies. The less consistent error rate of the regular verb forms is likely due to the smaller numbers that determine the rate. Nevertheless, in both Bock and Eberhard's study and the one reported here the error rate for regular verbs appears to be slightly higher than the error rate for irregular verbs. The higher rate may be reflective of not only agreement errors but also of the regular verb form's proclivity to bound morpheme loss errors; i.e., the dr0pping of the third-person-present singular inflection. Stemberger (1984) has found that loss errors are much more likely than addition errors and he attributes their greater likelihood to a failure at the level of assembling the verb form. Specifically, he suggests that loss errors result when the higher frequency-less complex verb form (the plural form) is substituted for the lower frequency inflected form (the singular form). In three of the four experiments reported here and in three of the four experiments reported by Bock and Eberhard, all of the experimental preambles had singular subject noun phrases. Therefore, in both studies the correct form of the verb for the majority of the experimental preambles was singular and this means that there is a greater likelihood that some of the regular verb errors are actually bound morpheme loss errors. Although this type of error may have inflated the overall agreement error rate for regular verb forms, it should nonetheless 87 be equally distributed across the control and experimental conditions since both conditions contained singular subject nouns. Other difficulties in processing bound morphemes surfaced in the failures to correctly repeat the preambles in these experiments. In all four experiments the most common repetition failure resulted in a change in the number of one or both of the nouns of the presented experimental preambles. Across the four experiments 191 failures changed a plural noun in a presented preamble to a singular noun and 88 changed a singular noun to a plural noun. Although on the surface these failures appear to be production errors, i.e., bound morpheme loss and addition errors, it is very unlikely that all the failures are attributable to one or the same stage of processing. This is because the nature of the task minimally involved the perception, comprehension, elaboration, and production of the preambles. At any stage of processing the loss or addition of the plural morpheme may have occurred. Because it is impossible to determine when these errors occurred, the interpretation of the agreement patterns that follow these responses is speculative at best. It should be noted that a similar problem plagues the interpretation of the results of memory studies that have investigated the recall of marked versus unmarked lexical items (e.g., Clark and Stafford, 1969; Clark and Card, 1969; Lapointe and Dell, 1989). The consistent findings of a systematic simplification of the marked form to the unmarked form in recall errors may be reflective of the difficulty of perceiving, comprehending, recalling, or producing the marked forms. Despite the uncertainty as to what stage(s) the difficulty in processing the marked form occurs, the results of the agreement experiments here suggest that grammatical marking aids the processes underlying verb agreement. 88 Conclusion. The view of semantic markedness relations typically holds that the meaning of unmarked lexical expressions is simpler or less specific than the meaning of opposing marked lexical expressions which suggests that the processing of the unmarked is in some sense easier or more secure. However, the investigations of the markedness relations of number at the grammatical level reported here show that when singular noun phrases are marked for number, implementing correct verb agreement during production is less likely to be disrupted by conflicting number marking in the clause than when singular noun phrases are unmarked for number. These findings are consistent with the view of language processing that although there is a correlation between properties of lexical expressions at the various levels of linguistic representation, the representations within each level, as well as the processes that operate on them, are somewhat distinct The systematic manipulation of the grammatical property of number in the investigations of subject-verb agreement in production has offered a glimpse into both the nature of the processes at the syntactic level and the representations that they operate on. FOOTNOTES 1The table contains 12 of the 22 English consonants. The features that are listed in the table are from Jakobson and Halle (1956). The two features consonantal and vocalic have been ornrnitted since they do not define any of the oppositions among the phonemes listed in the simplified table. 2Jakobson and Halle (1956) defined each feature acoustically and "genetically" , i.e., in articulatory terms. The articulatory descriptions for each feature is as follows: Nasal refers to the stopping of the passage of air through the oral cavity by opening the velic to allow the air to pass through the nasal cavity. Compact refers to a place of articulation, namely the articulation in a dorsal location of the oral cavity. Continuant refers to the constriction of a location in the vocal tract that does not stop the passage of air, but is sufficient to cause a friction. Strident refers to the "high-intensity" noise from the passage of air that is caused by an obstruction at the point of articulation. Voiced refers to the periodic vibration of the vocal cords. Grave, like compact, refers to a place of articulation and includes both frontal and dorsal locations. 89 APPENDIX A Exgrimental preambles from Experiment 1: Match and Mismatch "the" The boat for the expedition(s) The check from the stockbroker(s) The donation to the charity(ies) The door to the office(s) The museum near the hotel(s) The order for the cook(s) The bridge to the island(s) The invitation to the party(ies) The memo from the accountant(s) The road from the mountain(s) The shipment from the warehouse(s) The warning from the expert(s) The appointment with the doctor(s) The entrance to the laboratory(ies) The key to the cabinet(s) The letter from the lawyer(s) The result of the survey(s) The song for the performer(s) The cover on the book(s) The stamp on the envelope(s) The name on the sign(s) The movie in the theater(s) The mistake in the program(s) The star on the flag(s) The slogan on the poster(s) The picture on the postcard(s) The crime in the city(ies) The prize in the cereal box(es) The advertisement in the magazine(s) The uniform for the employee(s) The chemical in the product(s) The label on the bottle(s) The date on the application(s) The defect in the car(s) The tag on the suitcase(s) The problem in the school(s) Match aee Mismateh Quantified Each boat for the expedition(s) Each check from the stockbroker(s) Each donation to the charity(ies) Each door to the office(s) Each museum near the hotel(s) Each order for the cook(s) Every bridge to the island(s) Every invitation to the party(ies) Every memo from the accountant(s) Every road from the mountain(s) Every shipment from the warehouse(s) Every warning from the expert(s) One appointment with the doctor(s) One entrance to the laboratory(ies) One key to the cabinet(s) One letter from the lawyer(s) One result of the survey(s) One song for the performer(s) Each cover on the book(s) Each stamp on the envelope(s) Each name on the sign(s) Each movie in the theater(s) Each mistake in the program(s) Each star on the flag(s) Every slogan on the poster(s) Every picture on the postcard(s) Every crime in the city(ies) Every prize in the cereal box(es) Every advertisement in the magazine(s) Every uniform for the employee(s) One chemical in the product(s) One label on the bottle(s) One date on the application(s) One defect in the car(s) One tag on the suitcase(s) One problem in the school(s) 9O 91 Exmrimental preambles from Experiment 2: _M_atch aflMismeLch ”the" The boats for the expedition(s) The checks from the stockbroker(s) The donations to the charity(ies) The doors to the office(s) The museums near the hotel(s) The orders for the cook(s) The bridges to the island(s) The invitations to the party(ies) The memos from the accountant(s) The roads from the mountain(s) The shipments from the warehouse(s) The warnings from the expert(s) The appointments with the doctor(s) The entrances to the laboratory(ies) The keys to the cabinet(s) The letters from the lawyer(s) The results of the survey(s) The songs for the performer(s) The covers on the book(s) The stamps on the envelope(s) The names on the sign(s) The movies in the theater(s) The mistakes in the program(s) The stars on the flag(s) The slogans on the poster(s) The pictures on the postcard(s) The crimes in the city(ies) The prizes in the cereal box(es) The advertisements in the magazine(s) The uniforms for the employee(s) The chemicals in the product(s) The labels on the bottle(s) The dates on the application(s) The defects in the car(s) The tags on the suitcase(s) The problems in the school(s) Mismatch Quantifed The boats for each expedition The checks from each stockbroker The donations to each charity The doors to each office The museums near each hotel The orders for each cook The bridges to every island The invitations to every party The memos from every accountant The roads from every mountain The shipments from every warehouse The warnings from every expert The appointments with one doctor The entrances to one laboratory The keys to one cabinet The letters from one lawyer The results of one survey The songs for one performer The covers on each book The stamps on each envelope The names on each sign The movies in each theater The mistakes in each program The stars on each flag The slogans on every poster The pictures on every postcard The crimes in every city The prizes in every cereal box The advertisements in every magazine The uniforms for every employee The chemicals in one product The labels on one bottle The dates on one application The defects in one car The tags on one suitcase The problems in one school 92 Exmrimental preambles from Experiment 3: Match and Mismatch "the" The boat for the expedition(s) The check from the stockbroker(s) The donation to the charit(y)ies The door to the office(s) The museum near the hotel(s) The order for the cook(s) The bridge to the island(s) The invitation to the party(ies) The memo from the accountant(s) The road from the mountain(s) The shipment from the warehouse(s) The warning from the expert(s) The appointment with the doctor(s) The entrance to the laboratory(ies) The key to the cabinet(s) The letter from the lawyer(s) The result of the survey(s) The song for the performer(s) The cover on the book(s) The stamp on the envelope(s) The name on the sign(s) The movie in the theater(s) The mistake in the program(s) The star on the flag(s) The slogan on the poster(s) The picture on the postcard(s) The crime in the city(ies) The prize in the cereal box(es) The advertisement in the magazine(s) The uniform for the employee(s) The chemical in the product(s) The label on the bottle(s) The date on the application(s) The defect in the car(s) The tag on the suitcase(s) The problem in the school(s) Mismatch Quantified The boat for many expeditions The check from many Stockbrokers The donation to many charities The door to many offices The museum near many hotels The order for many cooks The bridge to several islands The invitation to several parties The memo from several accountants The road from several mountains The shipment from several warehouses The warning from several experts The appointment with a few doctors The entrance to a few laboratories The key to a few cabinets The letter from a few lawyers The result of a few surveys The song for a few perfomrers The cover on many books The stamp on many envelopes The name on many signs The movie in many theaters The mistake in many programs The star on many flags The slogan on several posters The picture on several postcards The crime in several cities The prize in several cereal boxes The advertisement in several magazines The uniform for several employees The chemical in a few products The label on a few bottles The date on a few applications The defect in a few cars The tag on a few suitcases The problem in a few schools 93 Exgrimental preambles from Expement 4 : Match and Mismatch Coent Subiect Nome The garage behind the building(s) The crime at the soccer match(es) The shield for the soldier(s) The crop in the field(s) The fork on the napkin(s) The stain on the rug(s) The suggestion from the worker(s) The song for the performer(s) The tractor behind the fence(s) The ovation after the announcement(s) The cigarette in the tobacco advertisement(s) The glass for the bartender(s) The suitcase for the trip(s) The eraser for the blackboard(s) The shirt beside the dryer(s) The pass from the teacher(s) The car near the exit ramp(s) The assignment for the student(s) The couch near the window(s) The instruction in the computer program(s) Match fird Mismatch Mass Subiect Noues The parking behind the building(s) The violence at the soccer match(es) The armor for the soldier(s) The rice in the field(s) The silverware on the napkin(s) The mud on the rug(s) The advice from the worker(s) The music for the performer(s) The equipment behind the fence(s) The applause after the announcement(s) The smoking in the tobacco advertisement(s) The ice for the bartender(s) The luggage for the trip(s) The chalk for the blackboard(s) The laundry beside the dryer(s) The permission from the teacher(s) The traffic near the exit ramp(s) The homework for the student(s) The furniture near the window(s) The information in the computer program(s) Bre APPENDIX B Summary of miscellaneous responses in all four experiments. akdown of miscellaneous responses in Experiment 1: 101 total miscellaneous responses 1. 76 preamble repetition failures: (JON 4. 5. a) 64 Change in number of head or local noun The appointment with the doctors-> The appointment with the doctor The appointment with the doctors-> The appointments with the doctors Each order for the cooks-> Each orders for the cooks The appointment with the doctor—> The appointment with the doctors b)12 change in a word of the preamble The memo from the accountants—~> The memo for the accountant The stamp on the envel0pe-> A stamp on the envelope The label on the bottle--> The label on the bottom . 10 multiple repetitions before completion . 9 verb did not immediately follow preamble The shipment from the warehouse never arrived The mistake in the program...um...l don’t know...was my fault I omitted verb 5 responses no longer had preamble as subject phrase One appointment with the doctor-> One appointment with the doctor and you’re scared Breakdown of miscellaneous responses in Experiment 2: 2 1 78 total miscellaneous scores 223 preamble repetition failures a)151 change in number of head or local noun The dates on the applications—> The dates on the application The boats for the expeditions—> The boat for the expeditions The stamps on the envelope—> The stamps on the envelopes The donations to the charity-> The donation to the charity The names on each sign—> The name on each sign The advertisements in every magazine~> The advertisements in every magazines b)59 change in a word of the preamble The shipments from the warehouse—> The shipments to the warehouse The movies in the theater-> The movies in a theater The boats for the expedition--> The boats for the exposition c)13 added words to preamble The uniforms for the employees-> The uniforms were for the employees 94 95 2. 20 multiple repetitions 3. 30 verb did not immediately follow preamble 4. 5 no verb or imperceptible Breakdown of miscellaneous resmnses in Experiment 3: 62 total miscellaneous scores 1. 51 preamble repetition failures a) 31 change in number of either head or local noun The entrance to the laboratories-> The entrance to the laboratory The stamp on the envelopes—> The stamps on the envelopes The boat for many expeditions--> The boat for many expedition The memo from several accountants-> The memos from several accountants b)15 change in word of the preamble c) 5 words added to preamble 2. 6 multiple repetitions 3. 5 verb did not immediately follow preamble Breakdown miscellaneous resmms in Exgriment 4: 76 total miscellaneous scores 1. 55 preamble repetition failures a) 33 change in number of either the head or local noun The applause after the announcements--> The applause after the announcement The song for the performers--> The song for the performer The ovation after the announcements—> The ovations after the announcements The homework for the students->The homework for the student b)3 words added to the preamble c)19 change in words of the preamble 11 multiple repetitions 3 verb did not immediately follow preamble 2 verb was ommitted 5 2. 3. 4. 5. responses no longer had preamble as subject phrase REFERENCES Allan, K. (1980). Nouns and countability. Language, 56, 541-567. Battistella, E. L. (1990). _Iearkegiess: The evaluative sumrstructure of language . Albany: SUNY Press. Bauer, L. (1988). Number agreement with collective nouns in New Zealand English. Australian Jgurnal gf Linggisties, a, 247-259. Bock, J. K., & Cutting, J. C. (1992). Regulating mental energy: Performance units in language production. Journal of Memory and Language, 3_1_,99-127. Bock, J. K., & Eberhard, K. M. (1993). The meaning, sound, and syntax of English number agreement . Language ang ngnitive Prgeesses, 8, 57-99. Bock, J. K., 8: Miller, C. A. (1991). Broken agreement. ngnitive Psycholmy, Q, 45-93. Bock, K. , Eberhard, K. M. and Cutting, J. C. (November, 1992). Controlling number agreement on verbs and anaphors. Paper presented at the 33rd Annual Meeting of the Psychonomic Society, St. Louis, M O. Brewer, W. F. and Harris, R. J. (1974). Memory for deictic elements in sentences. Journal of Verbal Learning and Verbal Behavior,13, 321-327. Clark, H. H., & Card, 5. K. (1969). Role of semantics in remembering comparative sentences. Journal of Experimental Psycholggy, 8;, 545-553. Clark, H. H., & Chase, W. G. (1972). On the process of comparing sentences against pictures. ngnitive Payphglggy, _3_, 472-517. Clark, H. H., 8r Chase, W. G. (1974). Perceptual coding strategies in the formation and verification of descriptions. Meme ang anitign, 2. 101-111. Clark, H. H., & Stafford, R. A. (1969). Memory for semantic features in the verb. Jggrnal gf Experimental Psycholggy, @, 326-334. Cowart, W., & Calms, H. S. (1987). Evidence for an anaphoric mechanism within syntactic processing: Some reference relations defy semantic and pragmatic constraints. Memgty a ngnitign, 1_5_, 318-331. 96 97 Croft, W. (1991). Syntactic categories and grammatical relations: The cognitive organization of inforrration . Chicago: The University of Chicago Press. Francis, W. N. & Kucera, H. (1982). Freguency analysis of English usage: Lexicon and grammar. Boston: Houghton Mifflin Company. Gemsbacher, M. A. (1991). Comprehendlng conceptual anaphors. Language and Cognitive Processes, 6_, 81-105. Gordon, P. (1985). Evaluating the semantic categories hypothesis: The case of the count/mass distinction. Cognition, 2Q, 209-242. Greenberg, J. (1966). Language universals . The Hague: Mouton. Harris, R. J. and Brewer, W. F. (1973). Deixis in memory for verb tense. Journal of Verbal Learning and Verbal Behavior. 12. 590-597. Jakobson, R. (1939). Zero-sign. Russian and Slavic grammar studies (pp. 151-60). Jakobson, R. (1957). Shifters, verbal categories, and the Russian verb. Russian and Slavic grammar studies (pp. 135-40). Jakobson, R. (1984). Russian and Slavic grammar studies . Berlin: Mouton. Jakobson, R., & Halle, M. (1956). Fundamentals of Language. The Hague: Mouton. Kaplan, R. M., & Bresnan, J. (1982). Lexical-functional grammar. A formal system for grammatical representation. In J. Bresnan (Ed.), The mental representation of grammatical relations (pp. 173-281). Cambridge, MA: MIT Press. Lapointe, S. G. (1988). Toward a unified theory of agreement. In M. Barlow, 8r C. Ferguson (Ed.), Agreement in natural language: Approaches. theories. aee descriptions Stanford, CA: Center for the Study of Language and Information. Lapointe, S. G., & Dell, GS. (1989). A synthesis of some recent work in sentence production. In G.N. Carlson & M.K. Tanenhaus (Eds), Linguistig stetcture in language processing (pp.107-156). Dordrecht: Kluwer. Nixon, G. (1972). Corporate-concord phenomena in English. Studia Neo hilolo ica 5.4;. 120-126. Pollard, C., & Sag, I. A. (1991). Lnformation-based syntax and semantics: AgreemenL bindim and control, vol. 2. Stanford, CA: Center for the Study of Language and Information. Quirk, R., Greenbaum, S., Leech, G., & Svartvik, J. (1985). A comprehensive grammar of the Eeglish laeguage. London: Longman. 98 Roberts, C. (1987). Distributivity. In the Proceedings of the Sixth Amsterdam Colloquium. Schriefers, H. (1990). Lexical and conceptual factors in the naming of relations. ngnitive Psychology, 22, 111-142. Sparkes, l. G. (1985). DictionarL of collective nouns and group terms. Detroit, MI: Gale Research Company. Stemberger, J. P. (1985). Bound morpheme loss errors in normal and agrammatic speech: One mechanism or two? Brain and Language, 2_5, 246-256. Stemberger, J. P., & MacWhinney, B. (1986). Form-oriented inflectional errors in language processing. Cognitive Pachglggy, lb, 329-354. Stemberger, J. P. (1991). Apparent anti-frequency effects in language production: The addition bias and phonological underspecification. Journal of Memorv a_nd Language, 30, 161 -1 85. Strang, B. M. H. (1966). Some features of S-V concord in present-day English. In I. Cellini, & G. Melchiori (Ed.), English studies today: Fourth series (pp. 73-87). Rome, Italy: Edizioni di Storia e Letteratura. Waugh, L (1982). Marked and unmarked: A choice between unequals in semiotic structure. Semiotica, _3_8_, 299-318. Zandvoort, R. W. (1961). Varia syntactica. Language and society: Essays presented to Arthur M. Jensen on h_is seventieth birthday (pp. 193-203). Copenhagen, Denmark: Det Berlingske Bogtrykkeri. ”Illlllllllllilllllllllli