LINGUIST List 10.1950

Fri Dec 17 1999

Review: Bloom et al. Language and Space

Editor for this issue: Andrew Carnie <>

What follows is another discussion note contributed to our Book Discussion Forum. We expect these discussions to be informal and interactive; and the author of the book discussed is cordially invited to join in. If you are interested in leading a book discussion, look for books announced on LINGUIST as "available for discussion." (This means that the publisher has sent us a review copy.) Then contact Andrew Carnie at


  1. Anne Reboul, Review of Bloom et al., Language and space

Message 1: Review of Bloom et al., Language and space

Date: Thu, 16 Dec 1999 17:17:25 +0100
From: Anne Reboul <>
Subject: Review of Bloom et al., Language and space

Bloom, P., Peterson, M.A., Nadel, L. & Garrett, M.F. (eds) (1996), Language
and space, Cambridge, Mass. MIT Press, 597 pages.

Reviewed by Anne Reboul, ISC-CNRS, France

This book corresponds to the edition of the papers given by participants in
the Language and Space conference (Tucson, 16-19 may 1994). It is very wide
ranging and strongly multidisciplinary. It would be hard to find one domain
of the research in spatial cognition not represented in one or the other
contribution. Though there are some dissenting papers, the main trend is in
favour of non-linguistic, conceptual primitives from which more complex
concepts - some of them underlying the linguistic representation of space -
are built. On the whole, Jackendoff's model is present, explicitly or
implicitly, with approval or criticism - though mostly with approval -, in
most papers, which gives some sort of unity to the chapters. All in all
this book could have been called "Everything that you always wanted to know
about space... ".

The preface describes the conditions in which the conference took place in
humorous details. It apparently originated in a discussion between a
"language researcher" (Bloom) and a "perception researcher" (Peterson) on
the genesis of left-right confusion.
1. The architecture of the linguistic-spatial interface (Ray Jackendoff)
According to Jackendoff, the aim of his talk is to elaborate the connection
between language and vision. His general hypothesis is Representational
Modularity, the notion that encoding in the brain is done in different
formats (languages of the mind), the different modules thus obtained being
individuated through their representations, rather than through
input/output conditions. There are also interface modules, which are
domain-specific and informationally encapsulated. The relevant interface is
between conceptual structure (CS) and spatial representation (SR). The
lexicon is one component of the interface modules in the language faculty.
CS encodes propositional representation and SR image schema or mental
models. CS encodes linguistic meaning, though it is not language dependant.
Its representations are built out of primitive features and functions.
Though it supports inference, it is relative to our conceptualization of
the world and it is not entirely digital as it has to account for
stereotype and family resemblance.
Jackendoff enumerates a few features which CS must include, such as
pointers to all sensory modalities, the distinction between token and type,
the encoding of quantifier and quantifier scope, the capacity of
abstracting actions from individuals, taxonomic relations, social
predicates (i.e. "uncle", "friend", etc.), and modal predicates. The
interface between CS and syntax preserves embedding, though CS has no
linear order and though syntax ignores most of conceptual structure, baring
argument structure. Language specific features are represented in the
interface between syntax and CS.
Jackendoff comes back to spatial representation and indicates it must solve
the problem of object constancy, encode spatial knowledge of invisible
parts of objects (hollowness for instance), degrees of freedom relative to
shape for objects whose apparence can change, shape variations, spatial
layout and be independant of spatial modality, to be able to take into
account haptic perception and proprioception. SR is geometric, but not
imagistic (no point of view), it encodes image schemas (2 1/2-D sketches).
It interfaces with CS through three criteria, economy (not double encoding
of information), interfacing (commensurability of representations),
grammatical effect (semantic distinctions which imply syntactic differences
must be encoded in CS and not only in SR).
Jackendoff then comes to a central point, the discussion of frames of
reference, of which there are, according to him, eight, four which are
intrinsic (geometric frame, motion frame, canonical orientation, canonical
encounter) and which determine the axis proper to an object, and four which
are environmental (gravitational frame, geographical frame, contextuel
frame, observer frame). He then turns to the relation between the axial
vocabulary and the SR/CS distinction. Relying on an experiment by
Narasimhan, he proposes a mostly SR hypothesis, according to which most
specifically spatial information is encoded in SR, rather than in CS.
2. How much space gets into language ? (Manfred Bierwisch)
The question which Bierwisch aims to answer is : "which components of
natural language accomodate spatial information and how ?" (31). Directing
his enquiry mainly on lexical semantics, Bierwisch points out that there is
no clear cut distinction between spatial and non-spatial vocabulary. He
places lexical and compositional semantics in I-language (in Chomsky's
sense) and endorses Jackendoff's hypothesis of a rich internal structure
for lexical items, with basic elements called "semantic primes". He,
however, does not endorse Jackendoff's view on modularity, prefering that
of Chomsky's where modules are (sub-)systems of tacit knowledge and not
levels of representation. This implies that the notion of interface is
relative to the modules interfacing and, on Bierwisch's view, phonological
form (PF) and semantic form (SF) are interfaces between I-language (mainly
syntax) and articulation-perception (A-P) on the one hand and
conception-intention (C-I) on the other.
Bierwisch wants to deal with what he calls I-space by analogy with
I-language. I-space draws information from several modalities of
perception, selecting relevant information and integrating it in a system
of spatial representation (SR) which is an interface between perception and
C-I. Bierwisch gives his own list of the conditions which SR should meet :
it is based on a set of locations with topological and metrical structures
; locations can be occupied by spatial entities though the assignation of a
location to an object is time-dependent, allowing for motion ; locations
are subject to conditions of orientation relative, notably, to verticality,
and to observer or intrinsic orientation. SR is distinct from CS because it
is domain-specific and analogous. SR and CS conjointly are two levels of
representation in C-I. The correspondances existing between the two systems
allows SR to account for the shape, size , place and path of objects. The
commonsense ontology of objects, places and events is common. SR deals with
relational properties which can be employed in other (non-spatial)
conceptual systems, such as time, social hierarchies, etc.). This leads
Bierwisch to the hypothesis that the basic structures of space constitute a
general schema of conceptual knowledge.
Bierwisch then turns to transfer of spatial structure of which there are
two varieties, implicit transfer and explicit transfer, the first one being
the more common. He proposes a few cross-classifications of concepts,
concerning types of concepts (aspatial, extrinsically, intrinsically or
strictly spatial), types of color-relatedness (no relation, extrinsic,
intrinsic, strict), and types of time relatedness (idem). The same
classification can applied to properties. Bierwisch also introduces a
distinction between rich concepts ("record") and spare concepts ("circle"),
the second often being strictly spatial while the first are either
intrinsically or extrinsically spatial. Explicit and implicit spatial
concepts combine in CS through object schemata.
Basic spatial terms (BSTs) are those which answer the following criteria :
they are lexical items, belong to the basic core of the lexicon, and
identify strictly spatial units. They include various subsystems depending
on syntactic category and aspect of space involved. Research on BSTs has
three aims : identification of conceptual repertoire, of basic patterns and
of systematic options. The condition for grammaticalization is that a
configuration of primes in SR is grammaticalized if there is a
morphological category to which it is related through rules. This leads
Bierwisch to a rejection of the Jackendoff/Landau hypothesis on the
one-to-one relation between the What and the Where systems on the one hand
and open and closed classes of lexical items. In an appendix, Bierwisch
discusses locative prepositions and dimensional adjectives.
3. Perspective taking and ellipsis in spatial descriptions (Willem J.M. Levelt)
Levelt introduces a distinction between macroplanning and microplanning,
the second including spatial perspective taking, which is an abstraction
from the visual scene. Perspective taking basically involves three
operations : focusing on some portion of the scene (the "referent"),
focusing on some portion of the field where the referent is located (the
"relatum"), relating referent and relatum. Perspective being linguistically
free, there is a pragmatics of perspective systems. Spatial representations
are language-independent, but perspective-taking maps them onto semantic
representation. The question raised by Levelt is whether spatial ellipsis
is generated from a perspective-free or perspectivized representation. The
first possibility would support the hypothesis of the language-independence
of spatial representation. Levelt distinguishes, classically, three
perspectives, deictic, intrinsic and absolute, which differ on the point of
inferential potential. Converseness holds for the deictic and, most often,
for the absolute system, but not for the intrinsic one. Transitivity holds
for both the absolute and deictic system, but not for the intrinsic one.
Most perspective taking is implicit and misunderstandings may occur. This
is all the more the case given that quite a few spatial terms are common
between deictic and intrinsic systems. Another difficulty is that the
deictic system is viewer-dependent - absolute and intrinsic systems are not
-, and a given point of view can never be entirely shared. The intrinsic
system raises another problem in that it imposes the recognition of the
relatum orientation, which depends on interpretative rather than purely
visual properties. Conflict between perspectives can be solved by the
principle of canonical orientation which implies that the relatum is in a
normal position. If canonical orientation does not hold, the intrinsic
system is preempted by either the deictic or the absolute systems.
Levelt then comes to ellipsis. Ellipsis originating before perspective
taking is deep ellipsis, while ellipsis occurring after perspective taking
is surface ellipsis. His conclusion is that ellipsis occurs before
perspective on the basis of a visual representation (2- or 3-D). The only
perspective system which allows for the distinction is the intrinsic
perspective, which is not touched by Brown and Levinson's whorfian
hypothesis. Levelt concludes by a discussion of that hypothesis. His
conclusion is that languages, by favoring one perspective rather than
another, create attentional biases, which then lead to a facilitation
regarding the spatial features most commonly involved in that perspective.
4. Frames of reference and Molyneux's question: crosslinguistic evidence
(Stephen C. Levinson)
Levinson begins by a summary of the most generally adopted position and of
his own, rather different, one. Most research on space is conducted on
roughly Jackendoffian lines, with language-independant primitives used in
the representation of spatial information. However, according to him, the
use of this or that frame of reference is language-specific and implies a
preference among the speakers for the same frame of reference in
non-linguistic tasks. This raises Molyneux's question, i.e. the existence
and extent of "cross-modal transfer of spatial information" (110). Again,
Levinson distinguishes between three frames of reference (= Levelt's
perspectives): absolute, intrinsic and deictic or relative. He comes back
to Tzeltal, where the relevant frame of reference is absolute. He reports
on experiments devised to test memory and inference and visual recall and
gesture in nonlinguistic spatial tasks, with a rather robust result, in
which Dutch subjects use a relative conceptual coding, while Tenejapan
subjects (speaking Tzeltal) use an absolute coding system. According to
Levinson, the experiments exclude the use of linguistic mnemonic means. The
general conclusion is that the dominant linguistic frame of reference
biases the choice of a frame of reference for the encoding of nonlinguistic
conceptual representations. This, Levinson notes, in the current
theoretical set up, is surprising for three reason: the anti-whorfian
tendencies of most researchers, the modularity hypothesis which militates
against cross-modal effects, the apparent incoherence of having the same
frame of reference accross different modalities.
Levinson adresses the last question and notes that distinction between
frames of reference can be reduced to distinctions between coordinate
systems. After a short historical overview of the notion, he comes to
linguistic frames of reference, which he wants to examine in a
cross-linguistic perspective. He redefines the three main frames of
reference, intrinsic, absolute and relative (i.e. deictic). Though these
three frames of reference are in effect in language, not all must be found
for a given language. The choice of frame of reference is language
Levinson aligns frames of reference, grouping intrinsic and absolute
together as allocentric frames and absolute and relative together as
orientation-bound. He then points out that frames of reference are
translatable which leads back to Molyneux's question. Levinson's own way of
approaching the problem is to say that the question is whether the same
frame of reference can be used in all modalities. His answer to that
question is clearly positive.
5. The confluence of space and language in signed languages (Karen Emorey)
Emorey's main thesis is that, given that sign languages rely on space to
linguistically express locations, orientation and point of view, they are
especially relevant to the investigation of the relation between space and
language. Emorey notes a few areas where space enters sign languages:
notably morphological inflexion, coreference and anaphora, locative
expressions, frames of reference - intrinsic and relative can be expressed
simultaneously -, and narrative perspectives. She then turns to a few
experiments on the representation of space and resolution of spatial
puzzles in both signers and speakers. In one of them, centering on mental
rotation, it was shown that signers react faster than speakers in forming
representations in such conditions. The general conclusion is that though
signed languages are languages as any others and as such obey much the same
constraints as other languages, they are specific in that they can use
space and movement to linguistically represent space and movement in the
6. Fictive motion in language and "ception" (Leonard Talmy)
Talmy's aim is to propose a unified account of cognitive representations of
nonveridical phenomena in both linguistic representation and visual
perception. His theoretical framework is the "overlapping systems" model
which "sees partial similarities and differences accross distinct cognitive
systems in the way they structure perceptual, conceptual, or other
cognitive systems" (211). The main problem with nonveridical phenomena is
that there can be two contradictory representations, of which one is
assessed as more veridical than the other. The first he calls "factive",
while the second is called "fictive". The general pattern involved is
called "general fictivity" and can concern occurrence - presence/absence or
existence/inexistence -, change - change/stability in an object or state of
affairs -, motion - motion/stativity. Talmy's paper concentrates on fictive
motion coupled with factive stationariness. The linguistic pattern is that
where the literal meaning of a sentence describes the motion of a referent
which is believed to be stationary.
Talmy introduces the notion of palpability to account for general fictivity
in vision, where one more palpable percept (factive) is entertained
simultaneously with a less palpable percept (fictive) and where both
percepts conflict. The palpability parameter correlates with other
cognitive parameters which cover different domains of perception and
conception. This whole range is called "ception". Considering a
cross-domain correspondance, language and vision, for instance, there is a
chance to find component parallels, one between factive representations,
for instance, and the other between fictive representations.
Talmy then turns to fictive motion and notes that it concerns a number of
categories: emanation, pattern paths, frame-relative motion, advent paths
(both site manifestation and site arrival), access paths and coverage
paths. What distinguishes between them is the presence or absence of some
features: the presence of some elements in factive motion for fictive
effects, the factivity or fictivity of the moving entity, the
observer-neutral/observer-based nature of the fictive effect and the fact
that it is an entity or the observation of an entity which is conceived as
Talmy then turns to the varieties of fictive motion, subdividing some of
them: emanation paths subdivide in orientation paths, - themselves
subcategorized as prospect paths, alignment paths, demonstrative paths,
targeting paths, lines of sight -, radiation paths, shadow paths and
sensory paths. He then proceeds to show the common principle behind all of
them, the active-determinative principle, which allows for the
determination of the source of the emanation. It states that of two objects
that which is taken to be the most active is conceptualized as the source.
The foundation of that principle, according to Talmy, is a basic cognitive
system, i.e. agency. He then discusses the issue of fictive motion in
language and other cognitive system, insisting on overlapping between
systems rather than strict modularity as well as the relation between
fictive motion and such cultural beliefs as ghosts, spirits, the evil eye,
mana, etc.
Talmy quickly presents the other categories of fictive motion, pattern
paths, frame-relation motion, advent paths, access paths and coverage paths
before turning to the more general problem of "ception", that is a
generalization accross perception and conception. The next step is the
enumeration of the parameters which play over ception, of which palpability
is the most central. The other twelve parameters depend on it to such an
extent that he calls them "palpability-related parameters". Palpability
directly relates to the opposition concreteness/abstractness. Four
different reference points can be identified: (fully) concrete,
semiconcrete, semiabstract, (fully) abstract. The parameter of clarity
again is a gradient with clearness, distinctivness and definiteness at one
end, while vagueness, indistinctness, or murkiness inhabit the other end.
The parameter of strength opposes vividness to faintness, while that of
ostension deals with the "overt substantive attributes" (247) of an entity,
the parameter being the mesure up to which these overt attributes are
perceived or conceived as present in that entity. The parameter of
objectivity opposes reality or objectivity to subjectivity or imagination.
The parameter of localizability deals with the degree to which a given
entity is experienced as having a specific localization. The parameter of
identifiability opposes categorization to absence of categorization through
uncertainty. The content/structure parameter opposes the perception of an
object through its content to that of the same object through its
structure. The type of geometry parameter opposes the perception of the
object as amenable to euclidian geometry and metrical quantification to
that of the object as topologically geometric and qualitatively
perceivable. The parameter of accessibility to consciousness opposes the
fact for an object to be the center of attention to that of its being
utterly inaccessible to consciousness. It should be noted that at all but
the last level, the object can be accessed. The parameter of certainty
opposes certainty as to the perception/conception of an object to
uncertainty about it. The parameter of actionability opposes the ability of
agency relative to an object to the ability of receptive experience toward
it. Finally, the parameter of stimulus dependence relies on the opposition
between the presence and the absence of stimuli for the experience.
Talmy then comes back to vision and language, noting that both have two
subsystems, one for content and one for structure, and that palpability may
depend on which of these two subsystems is foremost in a given "ceptive"
situation. The content subsystem is on the concrete side, while the
structure subsystem is on the abstract side. There is a direct link with
language, in as much as open-class words are on the content side, and
closed-class items are on the structure side. These two subsystems are
complementary in both language and vision. This leads Talmy to the strong
hypothesis that both structure subsystems (in vision and in language)
connect with a single neural system. Talmy also sees the visual structure
subsystem as the origin of "abstract" pictorial representations such as
stick figures.
In conclusion, Talmy compares his position with others, begining by the
distinction between the What and the Where systems, advocated, among
others, by Jackendoff and Landau, concluding that the Where system does not
cover all structure. He continues with Lakoff's theory of metaphor,
explaining that it is superseded by the general fictivity theory.
7. The spatial prepositions in English, vector grammar, and the cognitive
map theory (John O'Keefe)
In this paper, O'Keefe is concerned with an extension of the notion of
"cognitive map", notably to include in it episodic information and to
account for its lateralization in the human brain. The cognitive map theory
includes entities, relationally located, spatial relations being specified
through places, directions - specified as a set of parallel vectors -, and
distances. Paths are ordered sequences of places and translation vectors
linking place of origin and place of destination.
In his paper, O'Keefe wants to pursue the idea that language is inherently
spatial and to lay down the foundations of "vector grammar", seeing spatial
prepositions as providing the spatial relations among sets of objects and
places and their modification through motion. On this view, the location of
an object is determined by a vector determining distance and direction
relative to a known location. The function of the locative prepositions
consists in identifying distance and direction.
O'Keefe then turns to the analysis of specific prepositions, beginning with
the "vertical" ones ("below", "down", "under" and "beneath"), concluding
that they can be easily accomodated in his vector system and that direction
in this case is obtained through the "universal gravity system". Horizontal
prepositions ("beyond", "behind", "beside", "by") differ from vertical
prepositions in that direction is essentially determined locally and that
vectors need to be coordinated through a global directional system.
Omnidirectional prepositions ("at", "about", "around", "between", "among",
"along", "across", "opposite", "against", "from", "to", "via", "through")
can be used irrespective of any direction. Temporal prepositions imply the
introduction of the fourth dimension, time being represented through a set
of vectors in that fourth dimension, which is conceived as "at right angles
to the three spatial ones" (301). They also imply the presence of a
reference point relative to which the vectors are oriented. The main
hypothesis is that "the temporal prepositions (...) specify the location,
order and direction within the fourth dimension of the entities and events
of the other three dimensions" (302). The difference among them is due to
the number of reference points which they imply.
With the introduction of time, the vector system can begin to account for
changes of both location and state on the semantic map. Both types of
change are, not surprisingly, represented by vectors. For location, the
vector links the place of origin at a given time with the place of
destination at a different time. For state, the vector links a given object
at a particular time to itself at a later time but at the same location.
O'Keefe then comes to the metaphorical use of vertical prepositions in the
domains of (social) influence and social status. This leads him to the
introduction of a fifth dimension for status and control and to the
insertion in vector theory of control and causation and the relation
between them as represented on that fifth dimension.
8. Multiple geometric representations of objects in languages and language
learners (Barbara Landau)
Landau begins by noting that understanding the spatial language of young
children implies understanding how they represent objects. She insists on
the fact that spatial prepositions generally imply two objects (the figure
and the ground) and their relation and that the use of spatial prepositions
imply the ability of representing objects in a fairly abstract way. It also
implies the ability of recruiting different aspects of representations,
considering them globally or partially, this last selecting surface,
orientation and other features of objects. This leads her to the hypothesis
that learning spatial prepositions is only possible when representational
abilities have been mastered.
Noting that topological properties, though they are necessary, are not
sufficient, Landau concentrates on the accessibility of nontopological
properties to young children and on three kinds of representations, coarse,
axial and fine-grained. Though there is variability in the prepositional
systems of different languages, there nevertheless seems to be universals
of geometric schematization of objects and of spatial relations. One of
them is the coarse representation of the figure object relative to the much
richer representation - notably axial - of the reference object.
>From these considerations, Landau and her team have designed experiments to
test experimentally the treatment of objects by young children. The
findings seem to justify the distinction between shape and location. This
raises the question of the combination of the two and led to further
experiments. They showed that children were readier than adults to conflate
shape and location. Landau then turns to the question of axial
representation. Though prepositions presupposing axial representations are
typically mastered later, there nonetheless seems to be evidence that
children who haven't mastered them can represent axes. This motivated
experiments regarding both the presence of axial representation in young
children and the influence of the structural properties of objects in
triggering axial representation. The results of these suggest that young
children do represent axes and that, though structural properties play a
role in the early years, young children are nevertheless able to set axes
even when perceptual clues are weak. The next thing was children's ability
to form fine-grained representations, and the evidence suggests that this
ability is present but mainly emerges when children learn the names of
objects, where object shape plays a crucial role. Indeed, the "shape bias"
gets stronger with age and is correlated with vocabulary. However, it is
context-dependent for both children and adults. How exactly informations
regarding shape are represented is still a mystery and Landau and team have
tried to pursue this question through experiments which involved change in
shape, size and malleability. What the results of these suggest is that
representation of shape seems to imply parts, their relations and their
range of relative motion.
Landau then turns to Tzeltal and more generally to cross-linguistic
evidence. She notes that "crosslinguistic differences point to a strong
role for early learning, but they do not invalidate the search for
universals that underlie the expression of spatial language" (351) and
shows that Tzeltal is indeed a good example for her thesis.
Finally, she examines the asymetry of representations of figure and
reference objects, which she explains in terms of the difference between
the What and the Where system, the Where system, concerned with location,
not needing the same kind of representations as does the What system.
9. Preverbal representation and language (Jean M. Mandler)
Mandler's main argument in this paper is to the effect that preverbal
conceptual representation is largely spatial and constrains what is
learnable. What is more, according to her, "language is structured in
spatially relevant ways" (365). Quite a lot of representational and
inferential abilities are acquired during the first year, notably concepts
of animals and vehicles. However, Mandler wants to defend a rather
different picture of the emergence of concepts from the one currently
available. She contests, for instance, the idea, taken from Rosch, that the
first concepts to appear are basic level : her experiments have shown that
7-months-old children distinguish global categories much earlier than
basic-level categories. Another widespread view, related to Empiricism,
claims that the first concepts are acquired on the basis of perception,
abstract concepts coming much later. This however does not explain how
abstract concepts are derived from the concrete concepts built from
perception. Indeed it makes difficult to understand how abstract
properties, such as "animate" or "interacts with me" can be represented.
Mandler's aim is to distinguish carefully between perceptual (sensorimotor)
categories and conceptual categories, showing that their respective
developments are quite different. The main difference is that conceptual
categories involve, and perceptual categories do not involve, meaning. The
source of conceptual knowledge is indeed perception, but perception
mediated by very early distinctions, such as that, for instance, between
biological and nonbiological motion, leading to a conceptual division
between "things that move in animate and inanimate ways" (370). Most of the
properties which lead to the establishment of such distinctions are
spatial/kinetic and it is Mandler's hypothesis that babies have a mechanism
geared to the abstraction of such spatial properties leading to the
construction of image-schemata, which "bridge the transition from
prelinguistic to linguistic representation" (373). Where language comes in
is in dividing the global classes of objects thus obtained (e.g. animate
vs. inanimate) in smaller classes (e.g. dogs vs. cats).
Mandler then turns to spatial expressions, crosslinguistic evidence, and
the whorfian question. She notes that though containment, contact and
support may be divided differently by spatial expressions in different
languages, this does not undermine the importance of such preverbal
concepts in the nonlinguistic representation of space as well as in the
acquisition of language.
10. Learning how to structure space for language: a crosslinguistic
perspective (Melissa Bowerman)
Bowerman's thesis is that the extent to which languages differ in their
representation of space makes nonlinguistic spatial knowledge insufficient
to account for linguistic acquisition. Though she recognizes that
prelinguistic children have a wide set of spatial abilities, she claims
that the categorization of spatial patterns of relations between objects
are heavily constrained by language which cuts across preexisting spatial
concepts. In other words, "different languages structure space in different
ways" (402). This leads her to posit a distinction between conceptual and
semantic knowledge and to try to assess the importance of each in the
acquisition of spatial language by children in different languages.
Bowerman turns to the description of the experiments which she and her
colleagues conducted on English- and Korean-speaking children. The main
differences between English and Korean are that, where paths are usually
expressed in English through closed-class words (prepositions), they are
typically expressed in Korean through open-class expressions (verbs).
Korean also takes more account of object shape than does English. The
results of the experiments show that as early as twenty months of age there
is a difference in structuration of space between English and Korean
children regarding paths. The next experiment involved English-, Korean-
and Dutch-speaking children, the result being that the youngest children's
characterization of space was more similar to that of adults speaking their
language than to that of peer-age children of other languages. This
evidence, according to Bowerman comes in stark contradiction with the
hypothesis that spatial terms are learned through a mapping onto
preexisting spatial categories. Bowerman discusses Mandler's hypothesis of
image schemata, trying to show that it is uneconomical.
She then criticizes the hypothesis of a set of spatial nonlinguistic
primitiveswhich would be universals and would constrain language and its
11. Space to think (Philip N. Johnson-Laird)
Johnson-Laird opens his paper by a definition of perception, thinking and
action: "perception is the transformation of local information at the
sensorium into a mental model of the world at a distance, thinking is the
manipulation of such models, and action is guided by its results" (437).
His paper is geared to showing that spatial reasoning is based on models
and that those models underlie other types of thinking. He begins by a
description of models in terms of a similarity of structure between the
model and what it represents (this, indeed, is what distinguishes models
from propositional representations). In the same way, inference is
different in models and in propositional representations. The main idea is
that testing the validity of an inference depends on testing its conclusion
against a model, given that mental model represents many situations. The
advantage of a mental model lies in the fact that it selects all and only
relevant information. This, however, does not mean that a mental model
eliminates the necessity for propositional information: indeed, it is
"essential to have a representation of the meaning of an assertion
independent of its particular realization in a model" (442). It is used to
update models. The first advantage of mental models is that they give a
decision procedure - the validity of an inference is checked against the
holding of its conclusion in ALL possible models of its premises. Another
advantage is that mental models help to account for nondemonstrative
inferences, through a unified account of inference, via a grading of the
modal strength of the conclusion, strongly reminiscent of that in possible
worlds theory: necessity corresponds to holding in all possible models of
the premises, probability in most models, possibility in some model, while
probability corresponds to holding in a few models and impossibility in no
Johnson-Laird then turns to space, insisting on the fact that the main
operations on mental models are mainly procedural (i.e. do not reduce to
formal rules of inference), with special algorithms for space. Spatial
models are "functionally organised on spatial axes" (445) and make
different predictions from the models devised on formal rules. This allows
for experimental testing which runs in favour of mental models.
Johnson-Laird claims that spatial reasoning underlies other types of
reasoning, for instance reasoning about time. His proposal is that such
reasoning rests on a static spatial model of the temporal sequence, where
one axis represents time. Again, experiments seem to validate this
hypothesis. Johnson-Laird then comes to the more general problem of the use
of diagrams in reasoning and pleads for the validity of spatial models in
12. Spatial perspective in descriptions (Barbara Tversky)
The main question that Tversky aims to answer in this paper is the
following: "given the freedom to select a perspective, what determines the
perspective selected?" (463). Tversky begins by a multidisciplinary account
of perspective taking, examining the various views taken by the different
disciplines. She then turns to her view on perspective, investigating
"determinants of perspective choice". She chooses to examine, not isolated
inferences, but extended discourse, which, as she points out, raises
different issues, mostly consistency of perspective and order in the
presentation of information.
As she points out, choice of perspective is probably pragmatically
determined, depending on ease of interpretation. She presents three models
of accessing objects around oneself. On the first, the "equiavailability
model", all areas of space are equally accessible: this predicts equal
reaction times to all directions. On the second, the "mental transformation
model", the position of objects in the (imaginary) field of vision plays a
major role and this predicts that reaction times will be different
depending on the spatial position of the object (with front the quickest
and back the slowest, all other positions coming in between). The third
one, the "spatial framework model", claims that subjects built mental
spatial representations, relying on the extension of the three body axes
and the positions of objects along them. Again different reaction times are
predicted, with the head/feet axis being foremost, the left/right axis
being the farthest, with front/back occupying the middle position.
Tversky then describes experiments where subjects were given choice of
perspective relative to different spatial layouts and where they were
tested relative to two different perspectives in the same narrative. On the
production side, people were given maps that they had to describe. The most
usual distinctions relative to such tasks are between "route descriptions",
where the adressee's deictic perspective is adopted and "survey tours"
which use extrinsic perspective. On the whole, the proportions of subjects
choosing route, survey or mixed perspectives were equal, but differences
were salient regarding the map which the subjects had to describe. As
Tversky puts it, "perspective seemed to depend on the environment" (482).
Tversky then describes a new experiment on the same principle but where
specific features varied in the different maps involved: the size of
environment, the uniformity of landmarks sizes, and the number of paths.
The result seems to indicate that it is the structure of the environment
rather than its size which is relevant to perspective choice. As Tversky
notes, extended discourses, unsurprisingly, tends to trigger switch in
In conclusion, she notes that there does not seem to be a default
perspective, that perspective is not necessarily consistent, and
classification of perspectives should include two levels: personal vs.
neutral perspectives, the first subdividing in speaker's and adressee's
perspectives, while the later subdivides between intrinsic and absolute
13. A computational analysis of the apprehension of spatial relations
(Gordon D. Logan and Daniel D. Sadler)
The authors begin by distinguishing three classes of spatial relations:
basic relations which relate an object's position to the viewer, deictic
relations, which take two or more objects as arguments and specify the
position of one (the located object) relative to the other(s) (reference
object(s)), in the reference frame of the viewer, and intrinsic relations
where the located object is determined relative to the intrinsic reference
frame of the reference object. The implications for computation are
twofold: the first task is to determine the reference frame; the second is
to identify the located object relative to the reference object(s).
Logan and Sadler's hypothesis is that regions of acceptability for spatial
relations are established through fitting a spatial template to the
objects. A spatial template is a two or three-dimensional field centered on
the reference object and its reference frame, and it determines three rough
regions of fit: good, acceptable and unacceptable. The cognitive process of
determining spatial relations would thus involve four kinds of
representations: a perceptual representation with objects and surfaces, a
conceptual representation with spatial predicates, a reference frame and a
spatial template. It also involves four type of processes: spatial indexing
which links arguments of the relations to objects in the perceptual
representation, reference frame adjustment, which extracts from the
reference object its reference frame, spatial template alignment, which
aligns the template on the reference frame, and computing goodness of fit,
which consists in ascertaining where the located object falls in the
spatial template (good, acceptable, unacceptable regions). However,
apprehension of spatial relations should be flexible, given that it is
crucially context-dependent. To account for this, the order of
representations and processes can be varied. An ordered sequence of
representations and/or processes is called a "program" or "routine". The
authors then review the relevant literature with a view of justifying their
hypotheses and describe their own experiments to that effect. Their
conclusions is that the data support the existence of spatial templates and
they make the further assumption that perceptual representation, reference
frame computation and spatial indexing are language independent, while
conceptual representation and spatial template are not.
14. The language-to-object perception interface: evidence from
neuropsychology (Tim Shallice)
The aim of Shallice's paper is the investigation of the relation between
accessing representations of space from visual perception and accessing
them from language. As Shallice points out, a number of pathological
syndromes are directly relevant to this issue. He distinguishes four
positions in the neuropsychological literature, one which is sympathetic
with Jackendoff's position, two which are not (they advocate more than one
semantic/conceptual system) and a fourth which is oriented toward
Shallice then turns to neuropsychological evidence, beginning with
syndromes which seem to support the existence of several subsystems, namely
those which involve category specificity or selective loss. One explanation
for that lies in the hypothesis that identification for some categories
(living things) mainly relies on sensory information while for others it
relies on functional information (artefacts). This has been investigated in
a connectionist model which seems to support the hypothesis.
Shallice examines another type of neuropsychological syndromes, those in
optic aphasias, where there seems to be an inability to name objects given
their images but an ability to indicate their nature through miming. This
discrepancy might come from a disconnection between "visual semantics" and
"verbal semantics".
His general conclusion is that neuropsychological evidence is not in favor
of a single conceptual system, such as that advocated by Jackendoff.
15. Space and language (Mary A. Peterson, Lynn Nadel, Paul Bloom and
Merrill F. Garrett)
In this final chapter of the book, the editors both propose their own view
of the relations between time and space and provide a general summing up
and evaluation of the hypotheses proposed by various contributors. They
adopt Jackendoff's proposal with its three-fold distinction between
language, space and concepts. This raises some questions, notably about the
(lack of) transparency between space and language, the relations between
language and concepts and their mutual influence, and the use of language
in studying underlying concepts.
The authors review the existing neurological literature relative to spatial
representation, showing that it supports the existence of specific
representations for space ("cognitive maps") and concluding that the
evidence is not in favor of a single amodal spatial representation, though
it does favor a distinction between environmental space and spatial
structure of objects, despite the fact that no absolute separation can be
made between location systems and categorical systems for objects. Turning
to studies of perception, they note that experimental evidence supports
roughly the same set of hypotheses. They then turn to language, partly
rejecting the space/closed class and category/open class equivalence
defended by Jackendoff and Landau, proposing an alternative hypothesis
according to which nouns map onto essentially conceptual and nonspatial
information, though a link between spatial and conceptual information may
exist when the shape of an object is relevant to its categorization, while
both abstract conceptual information (e.g. the notion of containment) and
concrete spatial information may be involved in spatial prepositions. They
defend a non-whorfian view of conceptual representation, assuming it to be
This leads them to a discussion of Levinson's experiment on Tzeltal
defending the hypothesis that his data can be explained through the idea
that "exposure to different languages can engage certain aspects of spatial
cognition to a greater extent than others" (571), which does not support a
strong whorfian hypothesis.

I would like to conclude this already too long review by strongly advising
anyone interested not only in space but in the relation between conceptual
and linguistic knowledge to read this book. It does not merely give a
panoramic view of most if not all of the important work realized on space
in the past, and pointers toward future research in the same direction, it
also provides fascinating discussions on the interaction of different types
of representations, on their internal oganisation, on the type of
modularity which the experimental evidence supports, etc. Another quality
is that the multidisciplinary nature of the book ensures wide disagreement
among contributors, with Jackendoff's position as one defender of a
primitive set of nonlinguistic concepts from which language draws for its
semantic representations as the one around which debates occur.
Jackendoff's position is indeed discussed or used, either for refinment or
for criticism, in most if not all of the papers. The opposite position,
which is strongly whorfian, finds its most important proponent in Levinson
and his experiments on Tzeltal.
There is not enough space here to take up the whole debate but it does
seem that before reassuming the whorfian position to be right, some serious
consideration should be given to the doubts expressed by some participants
in this volume, for instance, the editors in the final chapter: granted
that some languages favor one perspective or frame of reference over
others, granted that speakers of these languages will tend to favor the
same frame of reference in nonlinguistic tasks, does it allow us to claim
that language is actively transforming their nonlinguistic conception of
the world? Shouldn't we merely think that the linguistic use of a given
frame of reference makes speakers more familiar with the perceptual clues
used in this frame of reference, thus making them receptive to these clues
more than to others even for nonlinguistic tasks? If this were the case,
then there is no need for the worfhian hypothesis and indeed, even if the
majority of subjects favor the most familiar frame of reference, a minority
does rely on less familiar frames, showing that they are still accessible
to them.

Reviewer : Anne Reboul, Research Fellow at CNRS (National Center for
Scientific Research), France. Ph.D. in Linguistics, Ph.D. in Philosophy,
currently working at The Institute of Cognitive Sciences in Lyon, in the
Linguistics team. Has written quite a few papers in French and English.
Author of Rh�torique et stylistique de la fiction (1992, Nancy, PUN),
co-author of Dictionnaire Encyclop�dique de Pragmatique (1994, Paris, Le
Seuil. English translation in preparation for Basil Blackwell, Oxford), of
La Pragmatique ajourd'hui (1998, Paris, Le Seuil) and of Pragmatique du
discours (1998, Paris, Armand Colin). Is currently involved in a research
project on space with her colleagues at ISC.

Anne Reboul
Institut des Sciences Cognitives-CNRS
67 bd Pinel
69675 Bron cedex
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue