\u' n : x? . . f . ' L 3.41.“ ....1....L‘T“‘ , INN.“ ..4.:.‘......... t a. . 515.17... I .H..u .: ,-‘ Y‘ .1 . .s. :3 ~ n. u ~hv~~§V I‘lethAl; 1.: 72‘" ‘ ..;‘....~...\~ ....‘ . .u W H “H H 908 1930 IllllfllllllzlfllljlfmWu This is to certify that the thesis entitled Preferences among reference systems in the use of the spatial term above presented by Laura Anne Carlson has been accepted towards fulfillment of the requirements for Master of Artsdegree in Psychology pgaw/ 5&4me Major professor Date Wig/4! I MSU is an Affirmative Action/Equal Opportunity Institution _, WWI"... I triummum. i l I 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 c :\clrc\deteduer Drnapt PREFERENCES AMONG REFERENCE SYSTEMS IN THE USE OF THE SPATIAL TERM ABQLE By Laura Anne Carlson A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of MASTER OF ARTS Department of Psychology 1991 65¢—6077 ABSTRACT PREFERENCES AMONG REFERENCE SYSTEMS IN THE USE OF THE SPATIAL TERM ABQLE by Laura Anne Carlson Spatial terms such as am may be used and interpreted within three different reference systems: within the efirinsic system, the term uses an environmental feature, such as gravity, as a reference point; within the deictic system, the term uses the location of the speaker or the listener as a reference point; and within the mimsic system, the term uses a predefined side of an object as a reference point. Six experiments are reported that use rating and production tasks to discover the preferences for above within these systems by examining its use in descriptions of objects in pictures. When in agreement, the deictic and extrinsic systems were overwhelmingly preferred over the intrinsic system. When the deictic and extrinsic systems were put in conflict with a head tilt and a body tilt manipulation, an extrinsic preference was found. These results indicate that the gravitational definition of M dominates its use. ACKNOWLEDGEMENTS i would like to extend my thanks and gratitude to my chairperson, Dr. David E. Irwin, for his unfailing guidance, constant support, and superior editing skills. I would also like to thank my other committee members: Barbara Abbott, Kathryn Bock, Tom Carr and Lauren Harris, for theirthoughtful comments and suggestions on this project. In addition, i would like to acknowledge Steven Weede for his able assistance in the data collection. Finally, I would like to extend extra thanks to Tom Carr for helping me to get started a year sooner than would have otherwise been possible. TABLE OF CONTENTS Page LIST OF TABLES .................................................. viii LIST OF FIGURES .................................................. ix CHAPTER I. INTRODUCTION .............................................. 1 Employment of deictic/extrinsic and intrinsic systems ........ 5 Semantic theories of spatial terms ......................... 9 Miller and Johnson-Laird's (1976) Propositional Account . 11 Jackendoff's (1983) Conceptual Semantic Approach ..... 14 Theories of linguistic reference systems ..................... 16 Levelt's Principle of Canonical Orientation .......... 16 Garnham's Framework Vertical Constraint .......... 20 II. THE EXPERIMENTS ........................................... 22 Ill. EXPERIMENT 1 .............................................. 25 Method ................................................. 25 Subjects ........................................ 25 Materials ........................................ 25 Procedure ...................................... 28 Results .................................................. 28 Discussion .............................................. 29 IV. EXPERIMENT2 ............................................... 32 Method .................................................. 32 Subjects ........................................ 32 Materials ........................................ 32 Procedure ...................................... 34 Results .................................................. 34 Discussion .............................................. 37 V. EXPERIMENT3 ............................................... 39 Method .................................................. 39 Subjects ........................................ 39 Materials ........................................ 39 Procedure ...................................... 39 Results .................................................. 39 Discussion ............................................... 42 VI. EXPERIMENT4 ............................................... 45 Method ................................................. 45 Subjects ........................................ 45 Materials ........................................ 45 Procedure ...................................... 45 Results .................................................. 46 Discussion .............................................. 46 VII. EXPERIMENTS ............................................... 51 Method ................................................. 57 Subjects ........................................ 57 Materials ........................................ 57 Procedure ...................................... 58 Results .................................................. 59 Discussion ............................................... 65 VIII. EXPERIMENT6 .............................................. 69 Method ................................................. 69 Subjects ........................................ 69 Materials ....................................... 69 Procedure ...................................... 69 Results .................................................. 70 Discussion ............................................... 88 IX. GENERAL DISCUSSION ....................................... 90 APPENDICES ..................................................... 96 A. List of reference objects used in Experiment 1 ................. 96 B. List of primary and secondary objects used in Experiment 2 ..... 97 C. Sixteen objects in Placement1 pictures and sixteen objects in Placement2 pictures ........................................ 98 LIST OF REFERENCES ............................................ 99 LIST OF TABLES Table Page 1 Experiment 1: mean ratings for sentence acceptability by picture condition ...................................... 30 2 Experiment 1: frequency distribution of preference ratings for noncanonical pictures ................................ 3O 3 Experiment 2: mean ratings for sentence acceptability by picture condition ...................................... 36 4 Experiment 3: mean ratings for sentence acceptability by picture condition ..................................... 41 5 Frequency distribution of preference ratings for Experiments 2 and 3 ............................................. 43 6 Experiment 4: frequency table for fly placements by picture condition ............................................ 47 7 Experiment 5: proportions of descriptions in no tilt condition for noncanonical pictures and for all tilt conditions for canonical pictures .................................... 60 8 Experiment 5: proportion of picture descriptions by tilt , placement, and description condition ................... 61 9 Experiment 6: questionnaire and responses ............. 71 10 Experiment 6: proportion of picture descriptions by group, placement, and description condition ................... 76 1 1 Experiment 6: proportion of picture descriptions by subject classification, and description condition for nonrotators . . . .82 12 Experiment 6: proportion of picture descriptions by subject classification, and description condition for semi-rotators . . 83 13 Experiment 6: proportion of picture descriptions by subject classification, and description condition for rotators ....... 84 viii Figure 1 CDOTACON LIST OF FIGURES Page Illustration of the deictic, extrinsic and intrinsic reference systems ............................................. 2 John and the flies (from Levelt, 1984) .................. 18 Sample pictures from Experiment 1 ................... 26 Sample pictures from Experiment 2 .................... 33 Sample picture with Placement1 for Experiment 5 ...... 53 Sample picture with Placement2 for Experiment 5 ...... 54 CHAPTER I. INTRODUCTION Communication requires coordination. Consider the simple process of a homeowner telling a mover where to place a table, captured in the utterance, "Put the table to the left of the chair." In order for the act to be successful, the speaker (the homeowner) and the listener (the mover) must share a common language. They must have a common vocabulary; in other words, they both need to use the term chair to denote the object in question, rather than one labelling the object as a Locke; and the other labelling it as a recliner. They also must share a common knowledge base; that is, both must know that a chair exists in the room (Clark & Schaefer,1989). Additionally, many speech acts require the listener to understand the utterance in the context of the speaker's presentation. For example, the description of the spatial location of an object may depend on the location of the speaker, on the location and orientation of a second object, or on a feature of the environment. Each of these possibilities defines a linguistic reference system. When a spatial description is based on the location and orientation of the speaker, it is said to be deictic (Fillmore, 1975; Miller & Johnson—Laird, 1976). When a spatial description is based on the location and orientation of the object being described, it is said to be ILLIMLSLQ (Fillmore, 1975; Miller & Johnson-Laird, 1976). Finally, when a spatial description is based on properties of the environment in which the speaker and the object reside, it is said to be m (Olson & Bialystok, 1983; Schober, 1990; Shepard & Hurwitz, 1984). An example of each of these systems appears in Figure 1, an illustration of the mover and homeowner exchange. Figure 1a is a depiction of the mover, the homeowner, the table and the chair. V H Figure 1a. Viewpoints of the homeowner (Vh) and mover (Vm).a v. i—riEl. VH Figure 1b. Deictic and extrinsic placement of the table, according to Vh. V” flT—‘F V H Figure 1c. Intrinsic placement of the table. Figure 1. Illustration of the deictic, extrinsnc and intrinsic reference systems. aNote that the homeowner' IS standing before the chair, looking across and down at it, while the mover is standing beside the chair. The placement of the chair, the location of the homeowner (VH) and the location of the mover (VM) appear in Figure 1a. The top of the "V" indicates the direction that the person is facing (think of the two points of the "V" as the eyes of the person). In the utterance, "Put the table to the left of the chair," the table is called the lacatad object and the chair the Lejaranga object. The mover's actual placement of the table depends upon his understanding the reference system used by the homeowner. If the mover interprets the utterance according to the perspective of the homeowner (i.e., deictically), he will place the table to the left of the chair from the homeowner's viewpoint, as illustrated in Figure 1b. This interpretation is deictic because it is based on the location and orientation of the speaker. If the mover interprets the utterance intrinsically, he will place the table next to the Chair's left side, as illustrated in Figure to. This interpretation is intrinsic because it is based on the predefined sides of the reference object. Finally, if the mover interprets the utterance according to properties of the environment such as the walls of the room, he will place the table so that it is to the left of the chair from the homeowner's point of view. This interpretation is extrinsic because it uses the room and left and right walls as reference points. Note that for this example, an extrinsic placement of the table agrees with a deictic placement of the table, as illustrated in Figure 1b. The agreement between extrinsic and deictic descriptions is pervasive for terms that locate an object vertically; when a speaker or listener is upright, spatial terms defined by the body often agree with spatial terms defined by the environment. For example, use of the term angle for a speaker who is upright means 'in a vertical upward direction'; this coincides with the use of the term as defined by gravity. An object is gravitationally above a second object if it is located in a vertical upward direction from it. Due to this correspondence, most 4 research on the use of such spatial terms disregards the distinction between the deictic and extrinsic systems (e.g., Garnham, 1989; Levelt, 1984). One goal of these experiments is to dissociate this correspondence in order to examine the preferences for each system. Therefore, within this proposal, when the deictic and the extrinsic systems offer agreeing descriptions, the term daimjgzaxtfiusig will be used. When these terms differ, daiatja and extrinsic will be used to mark each separate system. A number of methods have been used to study a speaker's selection of a reference system. One approach has been to simply document the ways in which speakers describe various spatial relationships. These studies have typically involved having subjects provide apartment descriptions (Ullmer- Ehrich, 1982) and route directions (Klein, 1983), respond to placement instructions (Cox & lsard, 1990), or describe the spatial lay-out of sets of objects (Levelt, 1982; Schober, 1990). A second approach has been to examine the linguistic properties of the spatial terms themselves. These investigations have typically focussed on detailing a framework imposed on perceptual space by a set of spatial terms, as exemplified by Miller & Johnson-Laird's (1976) propositional account of spatial locatives and Jackendoff's (1983) conceptual structure of semantics. Whereas the aim of the first approach is to document specific instances of a particular term and the reference system in which the term is employed, the aim of the second approach is to provide a general framework which incorporates the interaction between perceptual space and linguistic terms, as governed by the semantic constraints imposed by the whole set of spatial terms. A third approach has been to fuse the natural preferences for the use of a term in a given context together with the linguistic properties of the spatial term into a theory that predicts the use of the term in a variety of conditions. In other words, this third approach builds on the combined findings 5 of the first two approaches. The two main theories embodying this last approach are Levelt's (1984) Principle of Canonical Orientation and Garnham's (1989) Framework Vertical Constraint. The purpose of this thesis is to review the evidence and ideas generated by the research on natural preferences and theories detailing the linguistic properties of individual terms, show how they form the foundation for two theories that predict the use of particular spatial terms in given situations, and experimentally test between these two theories. Specifically, research that documents the way in which native speakers use spatial terms will be reviewed. This will be followed by a review of Miller & Johnson-Lalrd's (1976) and Jackendoff's (1983) semantic theories. Next, Levelt's (1984) Principle of Canonical Orientation and Garnham's (1989) Framework Vertical Constraint will be presented. Finally, a number of experiments will be reported that test the predictions offered by Levelt and by Garnham in order to differentiate these theories. Employment of deictic/extrinsic and intrinsic systems Evidence for the use of deictic/extrinsic and intrinsic reference systems comes from route direction studies in which subjects are asked to give directions to a particular location (Klein, 1983), room descriptions in which subjects are asked to describe the lay-out of their rooms, including the arrangement of furniture (Ullmer-Ehrich, 1982), and path descriptions in which subjects are asked to describe a set of objects arranged along a number of lines (Levelt, 1982). While these studies all differ in their methods and tasks, in each study subjects' responses may be reduced to those that are dependent on the context of the situation and the location of the speaker (deictic/extrinsic) and those that are independent of context (non-deictic or intrinsic). Moreover, each 6 study showed a differential preference for the two systems, with the deictic/extrinsic system generally more preferred than the intrinsic system. For example, Levelt (1982) had subjects describe the arrangement of different colored circles on different pathways. He defined two factors that accounted for the pattern of descriptions: a linear factor that described the direction and pattern of movement of the description, and an orientation factor that described the relationship between individual objects. The orientation factor had two levels, one corresponding to a deictic/extrinsic interpretation and one corresponding to an intrinsic interpretation. It the subject used a deictic/extrinsic interpretation, the description of the relationship among the objects would be from an external perspective, as though looking in on the pattern from the outside. An example of such an ego-oriented description would be, "To the left of the blue circle is a pink circle". In contrast, if a subject used an intrinsic interpretation, the description of the relationships among the objects would be an internal perspective, as though the subject were inside the pattern. An example of such a pattern-oriented description would be, "Straight ahead of the blue circle is a pink circle". As Levelt admits in a footnote, this interpretation of the intrinsic system is still somewhat dependent on the perspective of the speaker. Once the speaker "enters" the pattern, directions within the pattern revert to being deictically described. The clearest distinction he offers, then, is that the deictic system is an unambiguous description whereas the intrinsic system is dependent on interpretation, on the way in which a speaker proceeds down a pathway. Based on this distinction, Levelt found that in the cases in which one could differentiate between the two systems, the deictic/extrinsic descriptions were chosen over intrinsic descriptions roughly two-thirds of the time. In the present proposal, a more rigid distinction between the intrinsic and deictic/extrinsic reference systems will be adopted; objects with predefined 7 sides will be used, so that an intrinsic description may be defined as one that is based on the properties of the object. In a different study, Ullmer—Ehrich (1982) had subjects describe the layout of their apartments. She found that subjects generally began with a deictic/extrinsic perspective and maintained this throughout the description. An intrinsic perspective was used only when a sub-section was described, as an addition to the whole description. In addition, in situations in which the deictic/extrinsic and intrinsic interpretations conflicted, subjects tended to avoid the ambiguity by switching to a temporal relational expression, based on the tour of the apartment, rather than a spatial relational expression. ("After the sofa you come to the table" versus "To the left of the sofa is the table"). This study makes the important point that the selection of a particular system is not exclusive; rather, certain circumstances seem to result in deictic/extrinsic descriptions and others in intrinsic descriptions. For the apartment descriptions, the intrinsic system was used intermittently, while the deictic/extrinsic system provided the overall framework of the description. In contrast to these studies that found a preference for the deictic/extrinsic reference system, Cox and lsard (1990) found that children interpreted an utterance either intrinsically or deictically/extrinsically, depending on the information provided in the utterance. Their study was a developmental investigation of the use of 1010.01.91 and bahjag in a young group of children (mean age 5;10) and an older group of children (mean age 9;11). The task was to place a doll on a certain side of a car according to an instruction provided by the experimenter. The car faced away from the child. Cox and lsard either emphasized the child‘s viewpoint by telling the child to place the doll so that the child could pretend to take a photograph of the doll in front of the car, or did not emphasize the child's viewpoint, but simply told the child to place the doll in 8 front of the car. They also manipulated whether the spatial term was provided for the child by either telling the child to pretend to take a picture of the doll in front of the car or telling the child to pretend to take a picture of the doll and the car. They found that children in both age groups used a non-deictic reference system when their viewpoint was not emphasized. In other words, when the experimenter said "Put the doll in front of the car", the child tended to place the doll intrinsically to the front of the car. When the child's viewpoint was emphasized and the spatial term was omitted, the child placed the object deictically. Thus, when the observer asked the child to place the doll so that a picture could be taken of the doll and the car, the child placed the doll in between the child and the car. Finally, a split between the two systems was found when the child's viewpoint was emphasized and the spatial term was included. Finally, evidence for a preference of the intrinsic system over the deictic/extrinsic system has also been suggested by some researchers. For example, Wunderlich (1981) found a preferred use of the intrinsic system for the spatial terms "in front of" and "behind" when subjects were asked to describe the relationship between two objects when one of the objects had an intrinsic front. Levelt (1984) similarly predicts an increased use of the intrinsic system with these spatial terms when an object has an intrinsic front or acquires an intrinsic front side through movement. Finally, although Harris and Strommen (1979) stated that the self-referential system (deictic) develops before the object-referential system (intrinsic), they found that when both types of cues were present, the object-referent cues dominated the self-referent cues. Altogether, these studies suggest that the use of a particular system varies with the properties of the situation and the information provided in the utterance. Under one set of circumstances a speaker or listener uses the 9 deictic/extrinsic system and under another set of circumstances, the intrinsic system is used. The situations and tasks above are all extremely different, so that it is difficult to determine the circumstances governing the selection of a particular system. lmportantly, however, these studies all aim at categorizing the use of a particular reference system in terms of situational factors. For example, the hypothesis that a moving object could result increased use of the intrinsic system (Levelt, 1984) suggests that movement is a mediator in the speaker's selection of a particular linguistic system. Under this approach, one could document preferences by examining the properties of the situations. Semantic theories of spatial terms A different approach is to examine the spatial terms themselves, with the goal of coming up with a semantic representation of these terms. This approach advocates directly connecting the properties of the spatial term to the perception of the spatial relationship that the term describes. This connection is based on the use of language as a reflection of the perceived spatial environment. Human's perception of the spatial environment typically involves three axes, similar to the Cartesian x, y, and z axes. There is a vertical axis, typically corresponding to the perceiver's head/feet dimension, and two horizontal axes corresponding to the perceiver's front/back and left/right dimensions (Clark & Chase, 1972, 1974; Lyons, 1977; Talmy, 1983 ). This view is based on the fact that a speaker is a perceiver who is bounded by gravity, by the location of his sense organs, and by his primary direction of movement. These factors assign different saliency and strength to the three axes. According to Clark (1973), the vertical axis is considered the most stable and salient, as it is defined by gravity, a relatively unescapable permanent force on the earth's surface. The horizontal axis parsing front and back is considered the second most salient axis, due to 10 the location of the sense organs and the direction of mobility. Finally, the horizontal axis parsing left and right sides is the least salient and stable, due to the relative symmetry of the human body. Semantic components of spatial terms are seen to incorporate these environmental and biological influences. Thus, the spatial terms abaya and high carry with them a sense of direction further upward from the earth's surface (Lyons, 1977; Talmy, 1983). The use of these terms in a description of a spatial relationship assists a listener in locating an object by serving to restrict the perceptual space that needs to be searched in order to locate the object. For example, the statement "The ball was under the tree" restricts the search for the located object bafl to the location am the reference object mama. The semantic components control the segregation of space by acting as a set of conditions of application (Clark, Carpenter & Just, 1973) on the use of the term to describe a spatial relationship. In other words, a spatial relationship must embody certain characteristics in order for a term to be used to describe the relationship. For example, use of the spatial preposition u_nd_e_[ requires that the reference object has a surface below which the located object may be situated. Therefore, one may describe a dog as resting under a tree, but not as resting under a brick wall, unless it is buried there. This second approach then strives to account for the particular selection of a spatial term by the way that it constrains the interpretation of the perceived spatial relationship, both by parsing the perceptual space surrounding the reference object and by constraining the type of reference object that may be used. lmportantly, theories that embody this approach don't assert an order to the linguistic and perceptual interpretations. Generally, the unidirectional influence of linguistic interpretation on perceptual interpretation that has been reported in research using a sentence verification paradigm (Clark & Chase, 11 1972; Clark, Carpenter & Just, 1973) is not assumed in these theories. Rather, it is simply assumed that the selection of a term is compared against the perception of the spatial relationship in order to check the term's semantic constraints against the properties of the perceived spatial relationship. For example, in attempting to describe the spatial relationship of a dog lying beside a brick wall, one might first consider the term andar, compare the constraints of the term (namely, that the reference object has a surface beneath which the located object may be situated) to properties of the reference object (the brick wall), reject um, and opt instead for besida or magma. Much work has been done to establish frameworks that structure the semantic components for a given term and the patterns that emerge from these components across a set of terms (e.g. Talmy, 1983; Bennett, 1968; Fillmore, 1975; Lyons, 1977; Jackendoff, 1983; Miller & Johnson-Laird, 1976). Two approaches will be discussed here, for each offers a different emphasis on the way that a term's constraints are used to parse up the perceptual space surrounding the reference object. These emphases govern the interpretation of an ambiguous description that may be interpreted either deictically/extrinsically or intrinsically. Specifically, Miller and Johnson-Laird (1976) present an account in which the reference object plays a strong role in disambiguating the selection and interpretation of a spatial term, whereas Jackendoff (1983) presents an account in which the perceptual component of the spatial term has the leading influence. Miller and Johnson-Laird's (1976) Propositional Account According to Miller and Johnson-Laird, the description of an object in space presupposes the use of a vertical axis and two horizontal axes. However, the set of axes employed in spatial descriptions does not have standard units or a fixed set of coordinates that extend endlessly in predetermined directions. Rather, aspects of the utterance serve to fill in these 12 variables and align the axes in accordance with the utterance. One way of locating an object thus involves using an objective system that remains constant for all people, such as North, South, East, West. For example, to understand the spatial description "The main office is at 123 East Forty-fifth Street", one needs to impose the compass directions onto the set of axes in order to understand the location, such that one axis is aligned with the North/South direction and a second axis is aligned with the East/West direction. A second way of using these axes involves locating an object relative to a single point as defined by the location of a speaker, a reference object, or a salient feature in the environment, as with deictic, intrinsic and extrinsic descriptions. To embody this approach, Miller and Johnson-Laird (1976) f present a propositional account of spatial terms, in which a spatial term is treated as a two-place predicate, represented as TERM (target, relatum), where TERM stands for the spatial term, target refers to the located object (e.g., the table in the mover example discussed earlier), and Lelatum refers to the reference object (e.g., the chair). The location of the table with respect to the chair would thus be represented as LEFT(table, chair) within this system. In this approach, an interpretation assigned to a two-place predicate could be deictic/extrinsic or intrinsic, as seen in the mover example (LEFT(table, chair)). It the interpretation were intrinsic, the spatial designation of the target would depend on the referent or a property of the referent, such as an intrinsic side. In this case, the vertical axis would be aligned with the top/bottom axis of the referent object and the two horizontal axes would be aligned with the object's front/back and left/right sides. It the interpretation were deictic/extrinsic, the speaker would provide the point of reference. In this case, the vertical axis would be aligned with the head/feet axis of the speaker, and the two horizontal axes would be aligned with the speaker's front/back and left/right. To resolve 13 this ambiguity, Miller and Johnson-Laird (1976) state that in ambiguous circumstances, the predicate represents an intrinsic description when the relatum has intrinsic sides. Therefore, since the chair has a predefined front and back and therefore a left and right side, the mover in the original example should infer that the speaker employed the intrinsic reference system, and should place the table intrinsically to the left of the chair, as in Figure 1c. In situations in which the relatum does not have intrinsic sides, the predicate represents a deictic/extrinsic description. According to Miller and Johnson- Laird, these latter cases are typically marked by the speaker by the addition of a qualifier such as "From my point of view", and are marked in the predicate with a subscript d. as in LEFTd (table, chair). The use and interpretation of these two-place predicates to describe static spatial relationships predicts that the use of the spatial term depends on the properties of the relatum or the reference object. As shown in the Cox and lsard (1990) study, when children were told to place the man in front of the car, an object with a predefined front and back, they adopted an intrinsic interpretation. Miller and Johnson-Laird's (1976) theory would predict that if the children had been asked to place the man in front of an object that did not have intrinsic sides, such as a brick, they would have placed the man deictically in front, in between themselves and the brick. Indeed, Talmy (1983) has suggested that objects without intrinsic sides somehow adopt a front and back side based on their relationship with the speaker within a given utterance. Thus, although a brick does not have a predefined front or back, it assumes a front when one is facing it. If one were told to place the man in front of the brick, the man would then be placed between the viewer and the brick, in accordance with the viewer's position. This system then reduces spatial relationships to 14 two-place predicates whose interpretations are governed by the properties of the second argument, the relatum. Jackendoff's (1983) Conceptual Semantic Approach In contrast , Jackendoff (1983) follows Bennett (1968) in presenting a theory in which the spatial relationship is treated as an entity in and of itself; that is, it doesn‘t depend on properties of the relatum. According to Jackendoff, the relationship between an object and a location is represented by one of two elements: places or paths. A place is an element that represents a term that refers to a static location, whereas a path represents a term that stands for a dynamic location. For example, the preposition in would be represented by a place element, whereas the preposition into would be represented by a path . element because jam implies movement and a change of location. Places will be focussed on here, as the preceding discussion has dealt with static locations. A place takes the following phrase structure-like form: [placeX] > [place] or [placeX] > [place place-function([ThjngY]) The subscript serves as a reminder of the element type. An element of type [place] specifies a specific point or region; it is static, and is usually occupied by an element of type [thing]. The first term represents single-point locatives such as "here" and "there" which represent places in and of themselves. These places are deictic, in that correct interpretation of these terms requires a knowledge of the speaker's location. The second form includes the relatum of the phrase. For example, placelN([thing(KlTCHEN)]) means mm. According to Jackendoff (1983), a place function imposes conceptual constraints on its reference object through a series of preference rules, a list of typicalities that describe the necessary but not sufficient conditions for the use of 15 a spatial term. For example, in requires that its reference object be a bounded area or volume. Ambiguity occurs within this theory when a spatial term has two different senses, such as an meaning 'supported from below', as in ambitabla, and an meaning 'location at a surface', as in Qn_th_9_c_e_ili_ng. The intended sense is disambiguated by the application of the preference rules. In this case, the preference rule stating that the on requires a notion of support would be applied to the description. If an interpretation with this preference rule fails, the preference rule stating that 9_rt requires contact with an adjacent surface would be applied. The ambiguity present in the use of spatial terms are treated similarly within Jackendoff's theory, although in these cases, interpretation of the spatial term is governed by the perceptual, environmental and biological constraints subsumed within these terms (Shepard & Hurwitz, 1984; Shepard, 1981 ; Jackendoff, 1987; Olson & Bialystok, 1983). Thus, a preference rule for spatial term up would indicate that it refers to height in a direction above the earth's surface. Indeed, Shepard and Hurwitz (1983) found that statements such as 'She came right up to me' employed this sense of up. Extending this account of other spatial terms, then, Jackendoff's theory would predict that a term should be interpreted within a reference system whose interpretation agrees with the constraints (perceptual, environmental) imposed on it by its preference rules. The difference in emphasis between these two theories in offering a reconciliation of ambiguity in the use and interpretation of a spatial term has been stressed here. Miller and Johnson-Laird (1976) claim that the interpretation is based on properties of the reference object or relatum, whereas Jackendoff (1983) claims that the interpretation is based on the semantic constraints (perceptual, environmental) imposed on the term itself. It is important to point out that outside of this difference, these theories are 16 remarkably similar. Each stems from the use of a spatial term as a parser of the perceptual space surrounding the reference object. Also, each necessitates that the semantic components of the spatial term match the perceptual arrangement of objects in space. They differ only in their reconciliation of ambiguity. This difference in disambiguating an interpretation has been emphasized as a way to introduce a third approach to studying the use of spatial terms. This third approach builds on the preferences found in the empirical studies described earlier, and on the theories just presented. It centers on predictions about the use of specific terms in situations in which the deictic/extrinsic and intrinsic systems offer competing descriptions. In other words, the two theories that embody this third approach largely differ along the r same lines as the Miller and Johnson-Laird and Jackendoff theories differ. Given a spatial term that could be interpreted deictic/extrinsically or intrinsically, Levelt (1984) predicts that the term's uses will be governed primarily by properties of the reference object; in contrast, Garnham (1989) predicts that the term's uses will be governed by properties of the spatial term itself. Theories of linguistic reference systems Levelt's Principle of Canonical Orientation As with the studies reviewed above, Levelt (1984) is interested in documenting the uses of the different reference systems. Moreover, based on these uses, Levelt has formulated a model that outlines the conditions under which specific systems will be used. Within this model, a speaker's decision to employ a deictic/extrinsic or intrinsic reference system is governed by the Principle of Canonical Orientation. This principle is based on three factors: perceived verticality, the orientation of the reference and located objects, and the dimension linking them. The principle states that, in situations in which the deictic/extrinsic and the intrinsic systems offer different descriptions of the 17 relationship between two objects, the intrinsic system is allowable only when the located object is in "canonical" orientation with respect to the intrinsic parts of the reference object, along the dimension that is being described. For example, consider Figure 2, taken from Levelt. A man, John, is lying, on his side, with his head in the rightward direction (deictic/extrinsically). Fly1 is buzzing about his head, further in the rightward direction. A deictic/extrinsic description of this situation is given below the figure in Sentence 1: "Fly1 is to the right of John's head". Compare this with an intrinsic description of the picture, provided as Sentence 2: "Fly1 is above John's head". According to Levelt's Principle of Canonical Orientation, Sentence 2 is an acceptable description of the figure for the following reasons: The dimension that is being described is the vertical above/below dimension. The reference object, John's head, provides the organization of the above/below and left/right axes. Because John is supine, the above/below dimension becomes horizontal, aligned with John's head. According to Levelt, as long as the located object is aligned with this axis, an intrinsic description is acceptable. In order to use this rotated axis intrinsically, the object to be located must be aligned with the axis. In Figure 2, Fly1 is placed farther along the above/below axis than John's head; thus, this fly is in canonical orientation with respect to the rotated reference object. In sum, the use of the reference object (John) as the frame in which to locate the fly renders Sentence 2 acceptable. Now consider Fly2. Fly2 also is located along the rotated vertical axis, aligned with John's head. However, Fly2 is positioned much further along the axis, at a great distance from the reference object, John's head. Here, according to Levelt (1984), it seems less acceptable to allow the description ‘. qu3 Fiqz. Sentence 1. Fly1 is to the right of John's head. (Deictic/extrinsic) Sentence 2. Fly1 is above John's head (Intrinsic) Sentence 3. Fly2 is above John's head (Intrinsic - questionable acceptance). Sentence 4. Fly3 is above John's head (Deictic/extrinsic) Figure 2. John and the flies (from Levelt, 1984). given by Sentence 3: "Fly2 is above John's head". An addendum to the principle explains that the intrinsic system is not allowed in such an instance (and therefore Sentence 3 is not an acceptable description of the figure) because the reference object does not provide a compelling perceptual reference framework for Fly2. The large distance between the located object and the reference object results in Fly2 being interpreted within its own visual frame and its own alignment of axes. Consequently, the intrinsic system is not allowed. Of course this gets quite fuzzy; Levelt is vague about the specific parameters that govern when an object will be perceived within the visual frame of a second object and when it will be perceived within its own visual frame. Until such parameters are explicitly investigated, uncontroversial predictions about the use of the intrinsic system according to Levelt's Principle remain undisclosed. Factors such as distance and the functional relationship between the located object and the reference object may mediate the use of the intrinsic system; however, these will not be addressed extensively in this proposal. instead, attempts will be made to control these possible mediating factors. Later research will focus extensively on the intrinsic system, with the goal of specifying the relative contributions of these and other factors in eliciting the use of the intrinsic system; unfortunately, that set of experiments is beyond the scope of this thesis. These extensions will be further discussed in the General Discussion. For the present experiments, it will be assumed that it is possible for the located object to be recognized within the visual frame provided by the reference object. Given this assumption, it is necessary only to understand that according to Levelt the intrinsic system may be used in situations in which the located object lies in canonical orientation along the appropriate dimension of measure, as provided by the intrinsic parts of the reference object. 20 Note that this is very similar to Miller and Johnson-Laird's (1976) reconciliation of conflicting deictic/extrinsic and intrinsic interpretations. According to their theory, it the reference object has intrinsic sides, the axes should be aligned according to these sides, and an intrinsic interpretation should be made. Similarly, Levelt (1984) claims that an object located with respect to the intrinsic dimensions of a reference object allows intrinsic interpretations to be made. This idea conflicts sharply with Garnham's Framework Vertical Constraint. Garnham's Framework Vertical Constraint Like Levelt (1984), Garnham (1989) is interested in outlining the conditions under which specific reference systems will be employed. Garnham's model differs from Levelt's in that it rests on the primacy of the spatial term above rather than on the position and orientation of the reference object. Specifically, Garnham's Framework Vertical Constraint states that in situations in which the deictic/extrinsic and the intrinsic reference systems offer different evaluations of a situation, the use of a spatial term is not permissible if it conflicts with the deictic/extrinsic interpretation of angle. This constraint is based on the fact that the earth is subject to the force of gravity, and that this force governs the definition of abm and DM- Since people may not (easily) escape the force of gravity while on earth, gravity will always define the interpretation of alum and balm; situations in which the intrinsic and deictic/extrinsic reference systems suggest different usages of the term apple will default to the usage that coincides with the vertical as defined by gravity. Garnham's claim is that gravity is the controlling factor on the selection of a reference system, such that a reference system is permissible as long as it 1 doesn't conflict with the gravitationally defined use of 312912 and ham. 21 Therefore, based on the superiority of the term apple as defined by gravity, in contrast with Levelt (1984), Garnham's (1989) theory would judge Sentences 2 and 3 in Figure 2 as not acceptable. Gravity calls for the deictic/extrinsic use of abme. Thus, Garnham's constraint would only permit Sentence 4 as an acceptable description of the relation, "The fly is above John's head". Note that this approach follows closely the semantic theory put forth by Jackendoff (1983) in which the spatial term is considered an element in and of itself. Here properties of the spatial term govern its use. It is not dependent on properties of the relatum. This approach than would predict that use of the term Ma should be limited to situations in which the located object is gravitationally am (that is, farther from the earth's surface) than the reference object. CHAPTER II. THE EXPERIMENTS Levelt's (1984) Principle of Canonical Orientation and Garnham's (1989) Framework Vertical Constraint offer a basis for examining people's preferences for particular reference systems. These two theories offer differing predictions about the use of spatial terms in situations in which the two reference systems offer conflicting descriptions. These situations may be examined to determine how a speaker or listener decides which term and which reference system to use. Faced with a choice of more than one system, a speaker or listener may decide to only use the deictic/extrinsic reference system and treat descriptions according to his/her own point of view. Alternatively, a speaker or listener may decide to rely solely on the use of the intrinsic system, and thus treat descriptions based on object relationships, independent of perspective. Finally, a speaker or listener may consider both reference systems, and then select one based on the situational constraints, such as plausibility or location of the listener. According to this last method, one would expect to see differing selection of the deictic/extrinsic and intrinsic reference systems based on contextual properties of the situation. In order to test between these possibilities, situations must be created in which the deictic/extrinsic and intrinsic systems offer different spatial terms as descriptions of the relationship between two objects. Such situations were created for the following experiments by rotating a single object ninety degrees, and positioning a second object so that it appeared either deictically/extrinsically or intrinsically above the rotated reference object. Note that this rotation serves to dissociate these two systems so that am intrinsically does not equal am deictically/extrinsically, as shown in the 22 23 "John and the flies" example in Figure 2. The following experiments examined the use of the term apple in such situations. The spatial term abm was chosen over other spatial relational terms such as Left, right, Damon orinjanQt because the term best brings out the differences between the theories proposed by Levelt (1984) and by Garnham (1989). Specifically, Garnham states that the use of above may only be used as defined by gravity. Levelt states that angle may also be used when a located object is located along the vertical dimension aligned with the reference object's top and bottom sides. Therefore, when the reference object is rotated 90 degrees, Levelt permits the use of angle when the located object is further along the axis in the same direction as the top of the reference object, even though this is deictically/extrinsically Wthe ‘i reference object. While this research examines the use of angle within these contexts, the same questions and tests may be (and eventually will be) applied to other relational terms. The preference for a particular linguistic reference system in conditions in which the deictic/extrinsic and intrinsic reference systems offer conflicting descriptions for the spatial term abo_ve was tested in Experiment 1. Conditions in which the reference object was rotated and the located object was placed either deictically/extrinsically or intrinsically above the reference object were depicted. Subjects rated the acceptability of deictic/extrinsic and intrinsic descriptions of these scenes. Experiment 2 further examined this issue by manipulating the number of rotated objects and the selection of the reference object. Experiment 3 examined whether a response bias caused the preference of one system over the other. Experiment 4 used a production task in which subjects drew in the located object in accordance with their interpretation of the scene description. Experiment 5 used another production task in which subjects provided their own verbal descriptions of a scene, in 24 order to determine whether deictic/extrinsic or intrinsic utterances were preferred. In addition, Experiment 5 also examined the effect of dissociating the use of the deictic reference system from the extrinsic reference system by having subjects provide their descriptions in a no tilt, a head tilt or a body tilt condition. Experiment 6 replicated the body tilt condition of Experiment 5 but included a questionnaire designed to ensure that the body tilt manipulation was successful in rotating the subject's deictic reference system. These experiments will now be presented in detail. CHAPTER III. EXPERIMENT 1 The purpose of Experiment 1 was to examine whether viewers preferred a deictic/extrinsic or intrinsic description of the spatial relationship between a reference object and a located object when the reference object appeared rotated, in noncanonical orientation. As discussed earlier, the rotation of the reference object results in a dissociation of the deictic/extrinsic and intrinsic reference systems. Levelt's (1984) Principle of Canonical Orientation predicts use of both deictic/extrinsic and intrinsic descriptions as long as the located object appears along the vertical (above/below) axis as defined by the orientation of the reference object. Garnham's (1989) Framework Vertical Constraint predicts that deictic/extrinsic descriptions will be preferred. Method Subjects. Two hundred and ninety-two Michigan State University undergraduate introductory psychology students participated as part of an in- class experiment in partial fulfillment of a course requirement. The data from four subjects were deleted because they filled in multiple responses on several questions. Matarials. The stimuli were based on forty-eight pictures taken from the Snodgrass and Vanderwart (1980) norms that were expanded to create scenes. Each scene contained a reference object, a fly (a small filled circle) as the located object, and scene markings such as a horizon line or a wall or floor border. Four sample pictures appear in Figure 3. A list of all of the objects used in Experiment 1 appears in Appendix A. In sixteen of these pictures the 25 26 3a. 3b. The fly is above the cat. The fly is above the pineapple. Canonical correct Canonical incorrect 3c. 3d. "hm-s. r r . The fly is above the cake. The fly is above the donkey. Noncanonical intrinsic Noncanonical deictic Figure 3. Sample pictures from Experiment 1. 27 reference object appeared in canonical orientation (the canonical pictures). Eight of these objects had a canonical horizontal orientation and eight had a canonical vertical orientation. Of the eight vertical, four faced to the left and four faced to the right. Of the eight horizontal, three faced left and five faced right. In thirty-two pictures, the reference object was rotated ninety degrees so that it appeared in noncanonical orientation (the noncanonical pictures). Sixteen of these objects had a canonical horizontal orientation and sixteen had a canonical vertical orientation. Sixteen of the rotated objects were rotated to the left (eight horizontal and eight vertical) and sixteen were rotated to the right (eight horizontal and eight vertical). The scene markings for the rotated reference objects varied according to the object, and provided a context aimed at making the noncanonical orientation of the object plausible. For example, in a picture in which the reference object was a car rotated ninety degrees so that the front side of the car pointed toward the ground, a crane was hooked up to the back of the car, so that the car was suspended. Of the thirty-two noncanonical pictures, sixteen pictures had the fly placed deictically/extrinsically above the reference object and sixteen pictures had the fly placed intrinsically above the reference object. Of the canonical pictures, eight pictures had the fly appearing correctly (both deictically/extrinsically and intrinsically) above the reference object and eight pictures had the fly appearing to the side of the reference object, four to the left and four to the right. Below each picture appeared a sentence of the form "The fly is above the " where the blank line was replaced with the name of the reference object. Below the sentence appeared a rating scale, ranging from 1 to 5 with positions 1, 3 and 5 on the scale labelled "Not at all acceptable", "Moderately acceptable" and "Perfectly acceptable", respectively. 28 These pictures were copied onto transparencies for presentation in the experiment. The order of the presentation of the pictures was random. 2mm Subjects were told that they would be shown a series of pictures containing several objects, including a fly (a small filled circle). A sentence would appear below each picture, and their task was to rate the acceptability of the sentence as a description of the picture. The experimenter placed the picture on an overhead projector for thirty seconds and pointed to the position of the fly on the picture. Subjects marked their ratings on computer score sheets. The session lasted forty-five minutes. Resuhs The pictures were separated into four conditions according to the position of the fly: noncanonical deictic/extrinsic, noncanonical intrinsic, canonical correct and canonical incorrect. These were collapsed across the counterbalancing factors of canonical orientation (horizontal or vertical) and facing direction (left or right). The mean acceptability rating for each condition is provided in Table 1. The analyses in this and the other experiments were conducted with a significance level of .05 unless othenivise noted. A one-way (condition) with four levels repeated measures ANOVA was conducted with subjects as the random variable. A significant difference in ratings was found across the conditions (F1 (3,861) = 2839.4, MSE = .308. Three planned comparison tests were conducted to clarify the nature of this effect. There was a significant difference between the noncanonical deictic/extrinsic and intrinsic conditions, by subjects (F (1,287) = 2004.348, MSE = 1.093) and between the noncanonical deictic/extrinsic and canonical correct condition, by subjects (F (1 ,287): 149.7, MSE = .225). In addition, there was a significant difference between the noncanonical intrinsic and canonical incorrect conditions (F (1 ,287) = 10.94, MSE = .606). Independent t-tests conducted with items as the 29 random variable yielded a significant difference between the deictic/extrinsic and the intrinsic condition means for noncanonical pictures, T(15) = 35.98, and between the correct and the incorrect condition means for canonical pictures, T(7) = 22.18. Using a pooled error term to compare across canonical orientation of pictures, there was a significant difference between the deictic/extrinsic and the correct means, but not between the intrinsic and the Incorrect means. Finally, a frequency distribution of the ratings was calculated for the intrinsic and the deictic/extrinsic condition of pictures. These distributions appear in Table 2. The distributions offer confirmatory evidence of the ANOVA results. In short, 239 subjects rated the acceptability of the deictic/extrinsic pictures as greater than or equal to 4 while 269 subjects rated the acceptability of the intrinsic pictures as less than 3. Discussion This experiment showed an overwhelming preference for a deictic/extrinsic interpretation of the spatial relation between the reference object and the located object. This offers evidence against a strong interpretation of Levelt's (1984) Principle of Canonical Orientation that would maintain that the ratings for the intrinsic description should be as acceptable as the ratings for the deictic/extrinsic descriptions. However, the fact that the intrinsic placement of the fly resulted in significantly higher ratings than the incorrect placement suggests that subjects were able to recognize a use of the intrinsic system that would result in an acceptable description. In other words, subjects did not reject the intrinsic statements outright, as predicted by Garnham's (1989) Framework Vertical Constraint. 30 Table 1. Experiment 1: mean ratings for sentence acceptability by picture condition. Noncanonical Deictic/extrinsic 4.513 Intrinsic 1.755 Canonical Correct 4.855 Incorrect 1.603 Table 2. Experiment 1: frequency distribution of preference ratings for noncanonical pictures. Frequencies Ratings Range Deictic/extrinsic Intrinsic 1 < x < 2 0 190 2 < X < 3 2 79 3 < x < 4 41 16 4