Bryant's ideas represent a promising direction for research if a real collaboration among specialists on linguistic semantics, human perception, and mental representation will result. However, I have several concerns about the links proposed between perceptual and linguistic representations as well as those concerned with representations of space and location. My hope is that serious progress will be made and that future research should include the possibility that the most deeply entrenched of our linguistic tools are also relevant to the issue of the representation of space.
1.2 Bryant makes the following claims and assumptions:
(1) The discourse level, and no lower level of linguistic
structure, serves as input to the SRS, and hence must
provide categories translatable into this modality-
independent, presumably universal, system.
(2) The representation of information at the discourse level
is propositional in nature.
2.1 I will concentrate on the first of these points here, with a brief comment on the second. Why should the discourse level be presumed to be the most relevant to the SRS? I suggest that the most deeply conceptual categories of spatial relations are contained at the lexical/grammatical level. Second, since the provisions of the language at this level constrain the content at the discourse level to a certain extent, even if (1) above is true, attention to the lexical/grammatical level will ultimately give structure to the hypothesized content of representations in the SRS. The translation system between the linguistic representation and the SRS must, I suggest, take into account considerations from the lowest level (i.e., the categories of spatial relations which are conventionalized in the lexicon) to the highest level (i.e., the patterns of inferences resulting at the discourse level).
2.2 The studies Bryant mentions discuss the consequences for the SRS of spatial relations as described over long sequences of sentences. Bryant writes that the complex relations expressed at the discourse level are a product of those found at the lexical level but he claims that the level relevant to "the construction of mental spatial models by the SRS" is the discourse level. Anyone interested in the actual contents of the SRS, however, must also be concerned with the possibilities for the representation of space at the level of the lexicon and the grammar -- the level at which the language conventionally provides categories. The discourse level, while providing constraints on production and expression which are often language-specific, is also a level at which many more choices for both the fact of expression and the manner of expression are provided to the speaker. By contrast, in any given language the lexical level provides a relatively small and highly constrained inventory of choices for the expression of any relation; and the categories of relations a language conventionally provides its speakers vary enormously. What is more directly important for Bryant's claims is how the discourse level of linguistic representation is built up from the level of sentences, which is in turn built up from lexical items and grammatical structures. Even if we share Bryant's assumption that the lexical level is not relevant to the structure of the SRS we must recognize that what is represented at the discourse level is a direct function of what is represented at lower levels of linguistic structure -- plus general knowledge.
2.3 The first question, then, is whether conceptual representation inheres in representations associated with the discourse level or some other level. Some linguists, myself included, would argue that the questions of representation which have the deepest implications for the study of cognition reside at the lexical and not the discourse level. Whorf, the linguist most famous for worrying about the relation between language and cognition (Whorf 1964), argued that the grammatical level is more telling than the lexical level. This possibility is reflected in my term "lexical/grammatical," which covers any resource, lexical or morphosyntactic, which speakers use in constructing complex expressions.
2.4 A simple thought experiment shows that there is much more freedom of expression at the discourse level than at the level of lexical choice. Imagine a situation in which a hundred people from the same linguistic community are each asked to produce independently a description of a scene involving the locations of a collection of items. The chance of any subset of the hundred people producing identical discourses is vanishingly small. The chance of any subset of the hundred people producing the same set of sentences, irrespective of their order, is very small. But the chance that the same lexical items will be used by a substantial subset of the group to describe a given spatial relationship is very great. If, for example, two items are arrayed in a line running parallel to the speaker's left/right axis at some distance but visually accessible, the speaker is likely to say one of the following:
A is NEXT TO Bryant (or Bryant is next to A) A is to the LEFT of Bryant (or Bryant is to the RIGHT of A) A is BESIDE Bryant (B is beside A),
or some more complex expression including one of these spatial relation terms, such as
A is standing/ lying/ [etc.] . . . next to B
2.5 Those hundred speakers have a quite limited inventory of possible lexical-grammatical expressions of this relationship. Speakers will neither make up an entirely new phonetic form to express this relationship nor will they use a complex expression such as the phrase "arrayed in a line running parallel to my left/right axis and at some distance from my ventral surface," if a lexical expression like BESIDE is available. Words categorize relations, and people will use words provided by their language; hence they utilize the semantic categories provided by their language. Why is this point interesting for the question of the SRS?
2.6 First, even when dealing with a semantic area which has "objective" meaning directly expressing perceptual input, such as expressions of space, languages vary radically in the categories of relations which are provided by the lexical/grammatical system versus those that can be expressed using complex expressions (from the phrase level to the discourse level) versus those that habitually fail to be expressed at all by speakers. Some would argue that the language-specific level is the most deeply conceptual one. This means that the SRS must equally well represent information from a presumably species-specific perceptual input and from a linguistic input which varies wildly across the species. What does the SRS have to have in it in order to store both kinds of information equally well? I suggest below that it must be able to discriminate among all the categories of relation found lexicalized in any language.
2.7 Second, as Bryant notes, linguistic descriptions radically underspecify the information provided by the perceptual input. Bryant mentions that the SRS must represent underdetermined relationships but he does not discuss in detail how, when receiving linguistic input over a stretch of time, the SRS must constantly revise the representation -- not just by elaborating underdetermined information but often by correcting inaccurate inferences. General knowledge plays an important role in the elaboration of information which is not semantically expressed, as Bryant also notes; but it does not always fill in all missing information, and Bryant does not really elaborate under what conditions it kicks in. Finally, as I will suggest below, the contents of a discourse are in part constrained by the conventions of the lexical/grammatical system, so it is no small matter to construct a language-independent system which can take language-specific input and language-specific indeterminacies and construct conditional and in part flexible representations.
3.1 I would like to turn now to examples from a couple of languages. All the facts outlined below are vastly oversimplified, sometimes to the point of inaccuracy; I hope the inaccuracies do not detract from the overall point.
3.2 German has a preposition AN which translates into English variously as AT or ON. It can only translate as ON when there's a vertical-plane relation of contact, such as a painting hanging on a wall. It translates as AT when there is proximity, such as a person standing AT a window. That means that German conventionally encodes, by this preposition, vertical-plane proximity, but not the feature of contact (`ON') vs. mere proximity (`AT'). English, by contrast, encodes contact vs. non-contact -- ON vs. AT -- but it does not conventionally encode the plane of contact or proximity (we can encode horizontal-plane contact similarly as in "the phone is ON my desk" or vertical-plane contact as in "the poster is ON my wall". The two "kinds" of ON are differentiated lexically in German: AN vs. AUF. Thus the English sentence "The rug is hanging on the wall" is ambiguous: one could be talking about a vertical-plane relation where the rug is being used decoratively, like a tapestry, or a (partly) horizontal-plane relation in which a rug was hung over the top surface of a wall, say to air it out. (Notice that "general knowledge" will not help us with interpretation here.) If we take discourse as a real-time phenomenon in which information accrues in the understanding of the hearer/reader, we can easily imagine a discourse in which the English ambiguity is not explicitly resolved and a hearer has to construct and maintain two different representations or to choose the one most plausible to him; and at the presentation of information contradicting his inference he might have to backpedal into the alternate representation. Of course, the German speaker will in this case have the advantage of lexical differentiation to preclude ambiguity, but he might be found in a similar situation when encountering a sentence using AN and subsequent discourse.
3.3 A more exotic example comes from my own work on Mixtec, a language indigenous to Mexico (Brugman 1983). The Mixtec systems of spatial expression are different from those of English in important ways. It is possible to express the locations of objects without any semantic expression of their spatial relationships per se. One can produce a sentence which expresses only that there is a locative relation between two objects, with the specifics of that relation unexpressed and inferred through pragmatics. One can, for instance, say something like `The dish is-located face-table'. This sentence uses a general verb of location and a complex nominal which specifies a subsurface or subpart of the reference object. In this sentence, FACE-TABLE conventionally refers to the top surface of the table. (Depending on the object, FACE may refer to a front surface.) The fact that the dish is what we would call "on" (that is, it is making contact with that surface) is not expressed in this sentence (though it is expressible via verbs of location).
3.4 Another example: the sentence `It is-located HEAD-TREE' can imply `on top of the tree' or `above the tree' (among other relations) depending on other lexical information such as that provided by verbs like SIT or FLY AROUND or simply on the basis of general knowledge. Notice that I said "imply": For most of the system, ABOVE vs. ON is simply not expressed, and a discourse may continue without specification of such a relation. Here we may invoke general knowledge, but general knowledge will simply narrow down the possibilities of interpretation and can do so only probabilistically. From this I would argue that the lexical-level underspecifications must be available to the SRS so that constant revision may take place. I would also argue that there will have to be some kind of representational distinction between those properties and relations which are taken to be reliable and those which will be subject to confirmation or disconfirmation by further information. So the SRS must provide not only for the possibility of indeterminacy, but also for the dimensions of indeterminacy which are ultimately traceable to the lexical/grammatical system via discourse-level linguistic input.
3.5 Size and shape specifications are not a requirement of the lexical/grammatical system for expressions of space in English or German. But in Mixtec, size and shape specifications are crucial in the expression of space, for those specifications are what provide the information necessary to infer the relations themselves. (Talmy's (1972, 1985) work on Atsugewi elucidates a still different system.) So, as Franklin (1992) notes, other information about the objects described is something the SRS will have to be sensitive to. Bryant's hypotheses about the SRS allow such linguistic differences to be accounted for, and I do not dispute his claims, but we cannot infer that the discourse level is the linguistic level closest to cross-linguistic (universal) representation. Bryant's account makes a bare beginning at specifying what we need to investigate: We must look at what relations the lexical/grammatical level provides conventionally, even if, like Bryant, we are primarily concerned with the discourse level.
4.1 A secondary point concerns the assumption that the discourse level of representation is propositional. Talmy and others (see Hawkins 1984, Vandeloise 1984, Casad 1982, Talmy 1972, Brugman 1988, inter alia) have argued or assumed that spatial relation terms are most naturally represented nonpropositionally ("(image-)schematically") and that sentences are elaborations of these schemata; why shouldn't language be so represented at the discourse level as well? Certain linguists in the "cognitive" tradition (see Langacker 1987 and Lakoff 1987) have indeed assumed that the schematic representations in any given language represent a subset of the (universal cognitive) inventory of spatial relations which something like a SRS might exploit. Rt is therefore at least possible that such nonpropositional accounts of word meaning provide more direct claims about the contents of the SRS than any account which relies on a propositional representation; hence the burden of proof would rest on those who claim that propositional representation is either necessary or superior at the discourse level.
4.2 This commentary is not intended to impugn Bryant's proposals in any way. His ideas represent a promising direction for research if a real collaboration among specialists on linguistic semantics, human perception, and mental representation will result. My hope is that serious progress will be made and that future research will be constructed to include the possibility that the most deeply entrenched of our linguistic tools are also relevant to the issue of the representation of space.
Brown, P. (1991) Spatial Conceptualization in Tzeltal. Working paper No. 6., Cognitive Anthropology Research Group, Max Planck Institute for Psycholinguistics.
Brugman, C. (1988) Story of OVER. Outstanding Dissertations in Linguistics. New York: Garland Press.
Brugman, C. (1983) The use of body-part terms as locatives in Chalcatongo Mixtec. In Schlichter et al. (Eds.) Studies in Mesoamerican Linguistics. Report No. 4, Survey of California and Other Indian Languages, pp. 235-290.
Brugman, C. & Macaulay M. (1986) Interacting Semantic Systems: Mixtec expressions of location. In Proceedings of the Thirteenth Annual Meeting of the Berkeley Linguistics Society. 315-328. Berkeley Linguistics Society.
Bryant, D.J. (1992) A Spatial Representation System in Humans. PSYCOLOQUY 3 (16) space.1
Casad, Eugene H. (1982) Cora Locationals and Structured Imagery. Ph.D. Dissertation, U.C. San Diego.
Franklin, N. (1992) Inquiring into the Spatial Representation System. Commentary on Bryant on Space. PSYCOLOQUY 3 (16) space.1
Langacker, R. (1987) Foundations of Cognitive Grammar, vol. 1. Stanford, California: Stanford University Press.
Lakoff, G. (1987) Women, Fire, and Dangerous Things. Chicago: University of Chicago Press.
Leon, Lourdes de (1991) Space Games in Tzotzil: creating a context for spatial reference. Working paper No. 4., Cognitive Anthropology Research Group, Max Planck Institute for Psycholinguistics.
Levinson, S. (1991) Relativity in Spatial Conception and Description. Working paper No. 1., Cognitive Anthropology Research Group, Max Planck Institute for Psycholinguistics.
Lindner, S. (1981) A Lexico-Semantic Analysis of Verb-Particle Constructions with UP and OUT. Ph.D. Dissertation, U.C. San Diego.
Talmy, L. (1972) Semantic Structures in English and Atsugewi. Ph.D. Dissertation, U.C. Berkeley.
Talmy, L. (1985) Semantic Typology. In T. Shopen, Ed., Language Typology and Syntactic Description, Vol. 3. Cambridge: Cambridge University Press.
Vandeloise, C. (1984) Description of Space in French. Ph.D. Dissertation, U.C. San Diego.
Whorf, B. L. (1964) Language, thought and reality. Cambridge MA: MIT Press