Content uploaded by Barbara Landau
Author content
All content in this area was uploaded by Barbara Landau on Oct 21, 2021
Content may be subject to copyright.
Spatial language and spatial representation: a
cross-linguistic comparison
Edward Munnich
a,
*, Barbara Landau
b
, Barbara Anne Dosher
c
a
Department of Psychology, 220 Wolf Hall, University of Delaware, Newark, DE 19716, USA
b
Department of Cognitive Sciences, The Johns Hopkins University, Baltimore, MD, USA
c
Department of Cognitive Sciences, University of California, Irvine, Irvine, CA, USA
Received 30 March 1999; received in revised form 25 January 2001; accepted 23 February 2001
Abstract
We examined the relationship between spatial language and spatial memory by comparing
native English, Japanese, and Korean speakers' naming of spatial locations and their spatial
memory for the same set of locations. We focused on two kinds of spatial organization: axial
structure of the reference object, and contact/support with respect to its surface. The results of
two language (naming) tasks showed similar organization across the three language groups in
terms of axial structure, but differences in organization in terms of contact/support. In
contrast, the results of two memory tasks were the same across language groups for both
axial structure and contact/support. Moreover, the relationship between spatial language and
spatial memory in the two sets of tasks did not show a straightforward isomorphism between
the two systems. We conclude that spatial language and spatial memory engage the same
kinds of spatial properties, suggesting similarity in the foundations of the two systems.
However, the two systems appear to be partially independent: the preservation of particular
spatial properties was not mandatory across languages, nor across memory tasks, and cross-
linguistic differences in spatial language did not lead to differences in the non-linguistic
encoding of location. We speculate that the similarity in linguistic and non-linguistic repre-
sentations of space may emerge as a functional consequence of negotiating the spatial world.
q2001 Elsevier Science B.V. All rights reserved.
Keywords: Spatial language; Spatial representation; Cross-linguistic comparison
E. Munnich et al. / Cognition 81 (2001) 171±207 171
Cognition 81 (2001) 171±207
www.elsevier.com/locate/cognit
0010-0277/01/$ - see front matter q2001 Elsevier Science B.V. All rights reserved.
PII: S0010-0277(01)00127-5
COGNITION
* Corresponding author.
E-mail address: emunnich@udel.edu (E. Munnich).
1. Introduction
This paper concerns the relationship between spatial language and non-linguistic
spatial representations in memory. The nature of the correspondence between these
two systems of knowledge could take one of several forms: spatial language could
re¯ect universal non-linguistic spatial representations (Clark, 1973; Landau & Jack-
endoff, 1993; Talmy, 1983), non-linguistic representations could be molded by the
speci®c language which a person learns (Levinson, 1996; Lucy, 1992; Whorf, 1956),
or each system might play a role in shaping the other (Imai & Gentner, 1997).
Moreover, the correspondences between the two spatial systems might be complete
or only partial. In order to explore these issues, we ask how native speakers of
different languages express spatial relationships between pairs of objects, and
how they encode these same relationships in a non-linguistic memory task.
1.1. Background and rationale
At the heart of questions about the relationship between language and cognition
lie two contrasting views. One is that universals of non-linguistic spatial representa-
tion shape spatial language. This view assumes that the meanings expressed by
languages must re¯ect conceptual entities and relationships that are generally impor-
tant in human cognition. Foundational concepts ± such as space ± allow us to talk
about the world around us and our experiences in it. Because we all live in the same
physical world, our representations of space might well be universal. Furthermore, it
is plausible that spatial language has universal elements that rest on these represen-
tations (Clark, 1973; Jackendoff, 1983; Landau & Jackendoff, 1993; Talmy, 1983).
A second view is that differences in spatial language shape non-linguistic spatial
representations. This view is based on compelling observations of variability in how
languages encode space, and is often associated with the work of Benjamin Whorf
(1956) in the related domain of temporal cognition. Whorf claimed that the Hopi
language encodes the dimension of time quite differently from Western languages.
As a consequence, Whorf suggested, Hopi speakers possess a different concept of
time than that of speakers of Western languages (Lee, 1996).
1
Extending this argu-
ment to the domain of space, variation in spatial language might shape non-linguis-
tic spatial representations.
2
These two views ± that there must be universal non-linguistic foundations for
spatial language, and that there must be language-speci®c effects on spatial repre-
E. Munnich et al. / Cognition 81 (2001) 171±207172
1
Lee (1996), among others, has argued that the point made by Whorf (1956) was not that Hopi speakers
were unable to refer to temporal events at all, as some have said, but rather that the Hopi language invokes
a completely different concept of time than that of speakers of Western languages. Speci®cally, Whorf's
claim was that while Western languages refer to time on a continuum, Hopi makes reference to time in
discrete bundles, intimately tied to space (akin to the sense of time invoked in quantum physics).
2
Logically, there are two additional forms that the relationship between spatial language and non-
linguistic spatial cognition could take: universals of spatial language could lead to universals of non-
linguistic spatial cognition, or variation in non-linguistic spatial cognition could lead to variation in spatial
language. We will focus our attention on the views presented in the main text, because these two make
clearly opposed predictions, and have served as the main focus of recent research.
sentations ± have recently taken on new prominence in the face of discoveries that
languages vary widely in how they encode space (Bowerman, 1996; Brown &
Levinson, 1993; Lucy, 1992; Pederson et al., 1998). For example, the English
word ªonº encodes a relationship that is encoded by two separate words in German
(and other languages): German ªanº refers to instances of support involving attach-
ment, such as ªthe painting on the wallº or ªthe tab on the soda canº, whereas ªaufº
refers to support that can occur without attachment, as in ªthe cup on the tableº.
English (among other languages) does not draw this particular distinction in its
inventory of basic spatial terms,
3
and hence English differs from German in the
lexical encoding of these particular spatial relations. In other cases, English makes
distinctions not found in the basic spatial lexicons of other languages. For example,
English obligatorily distinguishes between relationships in which one object is ªonº
another and relationships in which one object is ªaboveº the other. However, both of
these relationships can be encoded by the single term ªueº (literally, ªtopº) in
Japanese.
Beyond these examples, there are cases in which the spatial distinctions that are
made in English appear to be orthogonal to the distinctions made by other languages.
In one study, Bowerman and Choi (1994) found that for a variety of joining actions,
English speakers focused on the distinction between support (ªput onº) and contain-
ment (ªput inº), while Korean speakers focused on tightness of ®t. For instance,
although English speakers differentiated ªputting a cap on a penº from ªputting a
cassette in a caseº, Korean speakers used the verb ªkkitaº ± meaning ªto ®t objects
together tightlyº ± to describe both. At the same time, while English speakers used
ªinº for both the latter example and ªputting a doll in a bathtubº, Korean speakers
contrasted ªkkitaº with ªnehtaº, which means ªto put objects together looselyº.
Thus, a natural encoding to the English speaker may be relatively unnatural in the
lexicon of another language, and vice versa. Another example in which languages
systematically covary is discussed by Talmy (1985, 1991), who distinguishes
between two broad typological patterns of describing motion events. Satellite-
framed languages (including English) canonically encode path (e.g. ªinº, ªoutº)
outside of the verb, while incorporating manner (e.g. ªrunningº) within the verb.
On the other hand, verb-framed languages (including Spanish and Turkish) place
path within the verb, and manner outside of the verb. So while an English speaker
would canonically say ªShe ran out of the roomº, a Spanish or a Turkish speaker
would say the equivalent of ªShe exited the room runningº. In studies of sponta-
neous production of motion verbs, Slobin and colleagues (Berman & Slobin, 1994;
Ozcaliskan & Slobin, 1999) have found that as early as 3 years of age, speakers of
E. Munnich et al. / Cognition 81 (2001) 171±207 173
3
In this paper, a basic spatial term is de®ned as a term with spatial content that is monolexemic, and that
is used obligatorily in certain situations in a given language. To be monolexemic, the term's meaning must
not be predictable from the meanings of its parts (Berlin & Kay, 1969). For example, ªtop leftº is not a
basic term, since it can be decomposed into ªtopº and ªleftº to discover its meaning. Our notion of basic
terms is related to the ªlexical conceptsº of Fodor (1981), which are monomorphemic predicates that he
contrasts with ªphrasal conceptsº, whose meanings are derived from several lexical concepts in conjunc-
tion (p. 260). For the purposes of this paper, we draw no distinction between an expression being
monolexemic and being monomorphemic.
satellite-framed languages focus more on the manner of motion for a given scene,
while speakers of verb-framed languages focus more on the path of motion for the
same scene (see also Ozyurek & Kita, 1999). Slobin (1996) has argued that as we
speak, the dominant patterns of our languages cause us to attend to certain aspects of
the world and not to others ± ªthinking for speakingº.
Such ®ndings have naturally led to the question of whether cross-linguistic differ-
ences lead to corresponding non-linguistic differences. One test-bed for such inquiry
has been the domain of color (Berlin & Kay, 1969; Brown & Lenneberg, 1954;
Davidoff, Davies, & Roberson, 1999; Heider, 1972; Heider & Olivier, 1972; Lantz
& Stef¯re, 1964; Stef¯re, Castillo Vales, & Morley, 1966). Heider and Olivier
(1972) carried out one often cited study, examining the structure of color naming
and the structure of color memory across two groups whose native color lexicons
differed. Native speakers of English, which has 11 basic color terms, were compared
to native speakers of Dani from Irian Jaya, Indonesia, whose language includes only
two basic color terms (ªmiliº and ªmolaº, corresponding roughly to ªdarkº and
ªlightº). Participants were given naming tasks in which they named color chips,
and memory tasks in which they remembered which colors they had seen. Despite
cross-linguistic differences in the color naming task, Heider and Olivier found that
the color memory categories were highly similar across language groups, indicating
that even quite substantial differences in the structures of the two color lexicons did
not result in changes to people's memory for color.
4
More recently, Davidoff et al. (1999) have reported evidence quite different from
that of Heider and Olivier. Davidoff and colleagues studied color naming and color
memory in the Berinmo of Papua New Guinea. They noted that although Dani
(studied by Heider and Olivier) has only one color distinction, it is consistent
with a distinction that English makes ± that is, Dani categories are supersets of
English categories. In contrast, the Berinmo vocabulary has categories that cut
across English category boundaries, thus possibly providing a stronger test of the
effects of language on color memory. Using Heider and Olivier's method, Davidoff
and colleagues found that Berinmo color memory was consistent with Berinmo
color terms, and accordingly not consistent with English speakers' color memory.
Thus, these ®ndings stand in direct opposition to those of Heider and Olivier, and
support the idea that cross-linguistic differences might change the organization of
color in memory.
This body of work on color has set the stage for recent research on cross-linguistic
differences in the coding of space, and the consequences of these differences for
spatial cognition. Spatial categories would seem to present a quite natural candidate
for shaping by language. First, spatial relationships are represented in multiple brain
systems, possibly organized by functional considerations such as perception and
action or by computationally different domains such as navigation and object recog-
nition (Gallistel, 1990; Milner & Goodale, 1996). Thus, spatial representation in
some systems could be affected by language, whereas in others it could be insulated
E. Munnich et al. / Cognition 81 (2001) 171±207174
4
Heider and Olivier (1972) de®ned color memory categories by patterns of memory errors. That is,
color tokens which were highly confusable were considered to be members of the same category.
from language. Second, spatial relationships play an important role in higher cogni-
tion, such as metaphor, analogy, and imagery (Gentner, 1977; Kosslyn, 1980), and
different kinds of spatial structure may be required for these different functions.
Finally, many aspects of spatial representations appear to undergo substantial devel-
opment. The use of spatial representations in tasks ranging from symbolic use to
explicit understanding of metric systems develops substantially over the ®rst 6 years
and possibly beyond (DeLoache, 1987; Huttenlocher & Presson, 1979; Newcombe
& Huttenlocher, 1992; Silberstein & Spelke, 1995). The protracted time course for
development and the extensive use of spatial representations in higher cognitive
functions suggests the plausibility of learning effects and, particularly, effects of the
learning of spatial language on the organization of non-linguistic spatial representa-
tions. Several recent studies provide intriguing evidence on this front.
1.2. Non-linguistic spatial representations correspond to spatial language
Hayward and Tarr (1995) sought evidence that foundational aspects of non-
linguistic spatial representations could be re¯ected in spatial language, resulting
in a correspondence between the two systems in the spatial properties they encode.
They noted that an object's location is de®ned with respect to a reference object, in
both linguistic and non-linguistic representation. Furthermore, the axes of the refer-
ence object appear to play an important role in assigning location, in tasks ranging
from strictly linguistic judgments of the acceptability of terms such as ªaboveº and
ªbelowº (Carlson-Radvansky & Irwin, 1993) to attentional tasks involving the
detection of objects relative to each other (Logan, 1995). In order to determine
the importance of axial structures for language and memory, Hayward and Tarr
examined the way native English speakers label an object's location, and they
compared this to people's accuracy in remembering those same locations. In a series
of experiments, they found that those locations which were most consistently named
by the basic spatial terms of English (e.g. ªaboveº, ªbelowº, ªrightº, ªleftº) were
also the locations that were remembered best.
In their language tasks, Hayward and Tarr asked participants either to describe the
locations of ®gure objects (the objects that are located) relative to a reference object,
or to rate the applicability of a set of basic spatial terms as descriptors of those
locations. They found that vertical terms (such as ªaboveº and ªbelowº) were most
often elicited and received the highest applicability ratings along the vertical axis of
the reference object; likewise, horizontal terms (such as ªleftº and ªrightº) were
most preferred along the reference object's horizontal axis.
5
In corresponding memory tasks, Hayward and Tarr assessed people's memory for
the same locations that had been used in the language task. Participants saw the same
series of scenes as in the language experiments, again depicting ®gure and reference
object relationships. After each scene disappeared, they were asked either to judge
E. Munnich et al. / Cognition 81 (2001) 171±207 175
5
In this experiment and in others presented in this paper, the scenes that participants saw did not have
axes drawn in; rather, axes are conjectured to be part of the participant's mental representation of the
scene.
whether a second scene showed the objects in the same spatial relationship as the
®rst, or to place the ®gure object in the location relative to the reference object where
it had appeared earlier. The results were analogous to those of the language task:
accuracy was highest when the ®gure object lay on one of the axes extending from
the reference object.
The parallel in the structures elicited by language and memory tasks among
English speakers is consistent with three contrasting possibilities, which could be
tested by carrying out a cross-linguistic replication of Hayward and Tarr's study.
First, non-linguistic representations might serve as an organizational basis for spatial
language. Second, the two systems might independently draw on the same set of
spatial properties (see Crawford, Regier, & Huttenlocher, 2000, for discussion of
this possibility). If either of these were true, we would expect to ®nd parallels in
spatial organization on non-linguistic memory tasks, even among speakers of a
language with a spatial lexicon quite different from English. The third possibility
is that spatial language shapes non-linguistic spatial representations. If so, cross-
linguistic differences in the spatial lexicon should lead to corresponding differences
in memory for location. The experiments described in this paper accordingly test the
third possibility against the other two ± if the third possibility is con®rmed, it would
be evidence for a profound effect of language on non-linguistic cognition; if it is
discon®rmed, further work would be necessary to decide between the ®rst and
second possibilities.
1.3. Spatial language may shape non-linguistic spatial cognition
A separate series of recent studies (Brown & Levinson, 1993; Pederson et al.,
1998) have sought to determine whether differences in spatial language give rise to
corresponding non-linguistic differences. In one series of experiments, Brown and
Levinson examined variation in the kinds of reference system used by speakers of
Dutch and Tzeltal. In Dutch ± as in English ± terms such as ªaboveº, ªbelowº, ªleftº,
and ªrightº are appropriate for use with object- or environment-centered frames of
reference, whereas ªnorthº, ªsouthº, ªeastº, and ªwestº are appropriate for use with
geographic frames of reference. Different terms are used depending on what frame
of reference is adopted by the speaker. For example, in English the position of a
particular bicycle may be described either as ªto the north of the treeº using an
ªabsoluteº (i.e. geographical) system, or ªto the left of the treeº using a ªrelativeº
(i.e. object- or environment-centered) system. However, these different reference
systems are generally used in different contexts. For small layouts, it is unacceptable
to use the geographic system, hence the oddity of ª*The bowl is to my eastº,
compared to ªThe bowl is to my leftº. Generally, the geographic reference system
in English ± and in Dutch ± is reserved for relationships on the scale of bicycles and
trees.
In contrast, speakers of Tzeltal, a Mayan language spoken in Chiapas, Mexico,
use an ªabsoluteº system in all cases except when two objects are contiguous; in that
case, speakers use the ªrelativeº system. Thus, the native speaker of Mayan would
®nd it perfectly natural to state the equivalent of ªThe bowl is to the eastº. Brown
E. Munnich et al. / Cognition 81 (2001) 171±207176
and Levinson asked whether this very different usage of reference frames between
Dutch and Tzeltal has an effect on the way that people encode spatial relationships in
non-linguistic tasks. If people speaking Dutch reserve the absolute frame of refer-
ence for large layouts, but people speaking Tzeltal use it for a much larger variety of
layouts, then there might be differences in the ways that these different groups
encode location in non-linguistic tasks. In particular, Brown and Levinson asked
whether speakers of Tzeltal might be more inclined to use the absolute frame of
reference to encode tabletop arrays, whereas speakers of Dutch might be inclined to
use the relative frame of reference.
Brown and Levinson further administered recognition, recall, and transitive infer-
ence tasks that could be solved according to either an absolute or a relative frame of
reference. The tasks required people to observe the locations of objects and to make
judgments about what constituted ªthe sameº spatial arrangement, either by repli-
cating the spatial layout after they had moved or by choosing a drawing that repre-
sented the correct layout.
6
Across three experiments, Tzeltal speakers did in fact
favor an absolute solution (placing objects north/south of each other), while Dutch
speakers favored a relative solution (placing objects left/right of each other). Brown
and Levinson concluded that the frame of reference dominant in a participant's
language biases the conceptual coding employed by the person in non-linguistic
tasks. Pederson et al. (1998) have obtained similar results across a variety of
languages, using a close variant on these tasks: participants from languages that
pattern with Dutch performed the non-linguistic task like Dutch speakers, while
participants from languages that pattern with Tzeltal performed the non-linguistic
task like Tzeltal speakers (but see Li & Gleitman, 2000 and our Discussions).
1.4. The present experiments
Several issues arise from the preceding studies. The ®rst concerns the evidence for
universal properties of non-linguistic spatial representations and their relationship to
spatial language. We ®rst ask whether non-linguistic spatial representations show
parallels to spatial language across different languages. If they do, these parallels
could re¯ect the shaping of spatial language by non-linguistic spatial representa-
tions. Alternatively, any parallels between the systems could be due to their sepa-
rately engaging the same spatial properties. Although Hayward and Tarr's studies
clearly draw on a plausible aspect of spatial representation ± the axial structure of
reference objects ± it remains to be seen whether this structure, in memory and in
language, appears consistently across tasks and across languages that differ some-
what in their lexicon of basic spatial terms.
A second issue is whether fundamental aspects of spatial representation such as
axial structure might arise universally, but still co-exist with cross-linguistic effects
E. Munnich et al. / Cognition 81 (2001) 171±207 177
6
The Brown and Levinson (1993) tasks were designed to elicit results characteristic of the reference
frame used. Their tasks all involved looking at an array, then turning around to look at a second array:
when one turns around, the relative frame is reversed (left becomes right), but the absolute frame is
unchanged (north remains north). Responses could follow either of two opposite patterns, according to the
frame of reference a participant uses.
of the kind shown by Brown and Levinson. It is possible that spatial language might
shape representations invoked by non-linguistic memory tasks; therefore, we ask
whether cross-linguistic differences in spatial language lead to differences in spatial
memory.
In order to test these possibilities, we examined people's language and memory
for the locations of objects in contexts virtually identical to those used by Hayward
and Tarr. The Language task required people to label the ®gure object's location
when it appeared in a range of locations around the reference object. The Memory
task required people to judge the identity or difference of the spatial relationship
they observed over a short memory interval. We focused on two aspects of spatial
organization: axial structure (as in Hayward and Tarr), and the role of contact and
support provided by the reference object for ®gures adjacent to it. Using these two
properties, we carried out cross-linguistic comparisons between adult native speak-
ers of English and of Japanese (Experiment 1) or Korean (Experiment 2).
These languages form an interesting comparison since they share certain proper-
ties in their spatial language, but are quite different in other respects. In particular,
English, Japanese, and Korean are all similar in their ability to encode basic spatial
terms at locations lying along the four main half-axes of a reference object (ªaboveº,
ªbelowº, ªleftº, and ªrightº in English). The fact that all three languages do have
such basic terms raises the question of whether these terms are used in the same
distribution across languages. If the proposed parallel between language and space is
universal, we might expect similarities in the structure of the linguistic representa-
tions and in the memory representations that arise for these locations, across all three
linguistic groups.
English, Japanese, and Korean are also rather different in their resources for
encoding the second spatial property we examine ± contact with the reference
object. In particular, English differs from Japanese and Korean in the way that it
encodes contact relationships involving support ± either by gravity (in the case of an
object on the top of another), or by adhesion (in the case of an object attached to the
side or bottom of another). In English, support in all cases can naturally be expressed
with the basic term ªonº. In the case of one object resting on another, as soon as the
®gure object moves even slightly upwards from the reference object, the ®gure
object must be described as ªaboveº the reference object. Similarly, in the other
(adhesion) contexts, once adhesion breaks, the ®gure object is no longer ªonº the
reference object; rather it is ªto the right/leftº, or ªbelowº, and so forth. In contrast,
the distinction between immediate support and non-support is not obligatorily
encoded in Japanese or Korean; the same term can be used for locations along an
axis that are either in or out of a contact relationship with the reference object. In
both Japanese and Korean, there are a number of verbs which can express support,
but these are only used in cases for which context speci®cally calls for such a
distinction.
The following sentences describe the scenes in Fig. 1a,b, illustrating the fact that
the distinction is obligatory in English (sentences 1a and 1b) but optional in Japanese
(sentences 2a and 2b) and Korean (sentences 3a and 3b):
E. Munnich et al. / Cognition 81 (2001) 171±207178
1a The cup is ABOVE
the table.
1b The cup is ON the table.
2a Chawan-wa tsukue-no UE-ni [UITE iru]/ aru
(cup-TOPIC) (table-GEN.)
a
(top-LOC.) (be ¯oating)/ (be)
2b Chawan-wa tsukue-no UE-ni [NOTTE iru]/ aru
(cup-TOPIC) (table-GEN.) (top-LOC.) (be on)/ (be)
3a cup-i thakca WI-e [TTE] issta
(cup-TOPIC) (table) (top-LOC.) (¯oating) (be)
3b cup-i thakca WI-e [PUTTE] issta
(cup-TOPIC) (table) (top-LOC.) (sticking) (be)
a
ªGEN.º denotes genitive marker; ªLOC.º denotes locative marker.
Japanese and Korean speakers typically use the base form of sentences 2a and 2b and
3a and 3b, respectively, adding verbs or adverbs such as those in square brackets only
when they intend to emphasize that support or lack of support is important to the
scene. The key is that the distinction is obligatory in English, so English speakers'
attention may be drawn to this distinction every time they use one of these sentences,
while Japanese and Korean speakers' attention may only be drawn to the distinction
when it is necessary. These differences ± like the differences between the Dutch and
Tzeltal in the distribution of use of different reference frames ± might shape people's
memorial representations. In particular, one might expect different memory structures
to arise for the native English speakers as compared to either native Japanese or
Korean adult speakers. On the other hand, memory structures might remain constant
over the three language groups, despite differences in language. If so, this would rule
E. Munnich et al. / Cognition 81 (2001) 171±207 179
Fig. 1. (a,b) Examples of relationships which are obligatorily distinguished in English but only optionally
distinguished in Japanese and Korean.
out effects of spatial language on non-linguistic spatial memory, at least in this task
context. At the same time, it would be consistent with universal non-linguistic repre-
sentations that exist side-by-side with linguistic differences.
2. Experiment 1
2.1. Participants
Thirty native Japanese and 30 native English speakers participated, with each
group roughly balanced for gender. Within each language group, no participant had
been exposed to languages other than their native language before the age of 12.
7
Native English participants were undergraduate and graduate students at the Univer-
sity of California, Irvine, while native Japanese participants were current students or
graduates of universities in Japan who were studying at the University of California,
Irvine, or the University of Delaware.
2.2. Design and materials
Ten participants from each language group participated in the Language task, and
the remainder participated in the Memory task. The design of each task was a
modi®ed replication of ones used by Hayward and Tarr (1995), in which a ®gure
object was placed at various distances and angles with respect to a reference object.
The principal change was in the stimulus arrays, which were broadened in the
present study in order to more closely examine any axial or contact effects. In
brief, while Hayward and Tarr's stimuli incorporated only one location adjacent
to each side of the reference object, our stimuli were designed so that the ®gure
object could occupy a range of locations adjacent to each of the sides of the refer-
ence object.
In both tasks, participants viewed a computer screen; a reference object was
displayed at the center of the screen, and a ®gure object was displayed in another
position on the screen. Both objects were positioned according to an overlaid 9 £9
grid, in which the reference object occupied the central 3 £3 cells and the ®gure
object occupied one of the 72 other cells (see Fig. 2, which shows a sample display
superimposed on a grid to indicate target locations; note that participants never saw
the grid). Each grid space was 0.5 inches (1.27 cm) square and the area in which
objects appeared was 4.5 inches (11.43 cm) square.
E. Munnich et al. / Cognition 81 (2001) 171±207180
7
All of our Japanese and Korean participants were recent arrivals in the US who had not been exposed
to English until the age of 12; all reported that they rarely used English in their free time, even while living
in the US. We set the cut-off at age 12 because this age corresponds to the end of what is generally
considered to be the critical period for language acquisition (cf. Johnson & Newport, 1989). Therefore,
even with substantial exposure to English, participants would not be expected to be able to gain a native-
like pro®ciency in English. In addition, Lucy (1996) has demonstrated that language-speci®c semantic
patterns in naming arise around the age of 8; after the age of 8, participants pattern strongly with the adults
of their native language group, suggesting that language-speci®c patterns may well be in place even
earlier than the age of 12.
For each task, there was a total of 72 different ®gure±reference object relation-
ships, each presented with two different ®gure objects, a circle and a square, for a
total of 144 different scenes. The two different ®gure objects were presented in
separate blocks, counterbalanced for order. The reference object was always a
square, as shown in Fig. 2. The 144 scenes were presented in randomized order
within blocks. All stimuli appeared on a Macintosh color monitor.
2.3. Procedures
2.3.1. Language task
Participants were asked to view each display and describe the relationship
between ®gure and reference objects by ®lling in the blank for a sentence of the
following form:
The small circle is __________ the large square:
8
Speci®cally, participants were instructed to describe the location using a ªsimple
word or phraseº, such as ªonº, ªaboveº, ªto the leftº, or ªon top ofº, and to avoid
using compass, clock face, or degree of angle answers. These instructions were
given in order to elicit the basic terms of each language. After writing their response
on an answer sheet, participants pressed a key to bring up the next scene.
E. Munnich et al. / Cognition 81 (2001) 171±207 181
8
The equivalent sentence was used in the Japanese version of the task: ªchisai en-wa ookii shikaku
__________.º (Actual stimuli were printed in Japanese script.) The ªbeº verb was left out of the Japanese
answer sheets because there are two possible be-verbs: ªaruº is used alone, while ªiruº is used in
progressive tenses, as in the English ªto be touchingº. The difference in position of the blank is due to
the fact that Japanese follows subject-object-verb order, while English follows subject-verb-object order.
Fig. 2. Example of a display used in Experiment 1. A 9 £9 grid is superimposed to indicate possible
positions of objects (this grid was not present in the display shown to participants).
2.3.1.1. Coding of responses. Responses were coded separately for the presence of
axial terms and the presence of contact terms. Axial terms were de®ned as single
words, which describe either vertical or horizontal orientation of the ®gure object
with respect to the reference object. These included ªaboveº, ªoverº, ªbelowº,
ªtopº, ªbottomº, ªunderº, ªleftº, ªrightº, and ªnext (to)º, and their equivalents in
Japanese (see Appendix A for a listing of expressions given by each language
group). While it was possible that the basic spatial terms of Japanese would not
map directly onto those of English, we found that the dictionary equivalents of the
English axial terms listed above were in fact the ones that were used by Japanese
speakers in greater than 99% of their responses, and that these particular terms were
the only means used by Japanese speakers to describe axial relationships, indicating
their primary status in the language.
Contact terms were de®ned as those terms expressing contact or support, and
included English ªonº (obligatory for many cases of contact) as well as verbs of
contact such as ªsitº and ªtouchº, which are optional. Japanese contact terms that
occurred included equivalents to the latter verbs as well as additional terms such as
ªnotteº (progressive form of the verb ªnoruº, meaning ªto be onº). Appendix A
contains an exhaustive list of contact terms that were used. Note that contact and
axial terms were coded independently, so expressions such as ªon the top/bottom/
left/rightº were coded as both contact (ªonº) and axial (ªtopº/ªbottomº/ªleftº/
ªrightº). Coding was done by the ®rst author, and reliability checks were done by
a native Japanese informant for 100% of the Japanese data and by another native
English speaker for 20% of the English data. Intercoder agreement was 98% for the
Japanese data and 97% for the English data.
While at least one axial term was used in almost all responses across languages,
it turned out that responses often contained more than one axial term, e.g. ªto the
top left ofº. Hayward and Tarr (1995) had coded such responses by assuming that
the ®rst term mentioned would have priority (in this case, ªtopº, a vertical term),
and this assumption gave rise to orderly results in their study, as well as in
preliminary analyses on our data (see Munnich, 1997). However, since lexical
ordering rules in Japanese are different from English, applying this rule across
the two languages would have resulted in some distinctly spurious patterns.
9
Therefore, the primary analysis of axial terms across both languages considered
only simple expressions ± those expressions in which a single axial term was
E. Munnich et al. / Cognition 81 (2001) 171±207182
9
Hayward and Tarr (1995) considered the ®rst (or only) axial term used in a response, based on the
assumption that the ®rst term given is primary. In fact, it is often the case that terms that come later in the
response are of equal or greater importance than the ®rst term. An example of this is that Japanese
participants often described the location of the ®gure object with compound expressions such as ªhidari
ueº (ªleft topº), but never with the reverse order. When queried, participants universally agreed that ªue
hidariº (ªtop leftº) is unacceptable in Japanese, and that placing the horizontal term ®rst has nothing to do
with its importance. Each language has rules ®xing the order of axial terms within complex expressions.
For a review of semantic and phonological rules that might be involved, see Cooper and Ross (1975). In
any case, it is incorrect to assume that the ®rst term in linear order is primary; hence, we elected to code
the pertinent term only when it occurred alone.
applied.
10
As will be seen, the results for English generally replicated Hayward and
Tarr using this method.
2.3.2. Memory task
Participants were told that they would view two scenes, separated by a mask, and
that they were to judge whether the second scene displayed the same spatial relation-
ship as the ®rst. At the beginning of each trial, participants saw a plus sign (100 ms)
for ®xation, followed by the ®rst scene (500 ms), a pattern mask (500 ms), the
second scene (500 ms), and ®nally a blank screen, at which point participants
pressed either the ªzº key with their left hands if the relationships between objects
were the same, or the ª.º key with their right hands if the relationships were differ-
ent. The key press activated the next trial. Within each trial, the second scene was
always displaced by 0.5 inches (1.27 cm) from the absolute position of the ®rst, in
order to eliminate any visual persistence.
The 144 different scenes were each presented eight times to each participant. The
spatial relationship between the ®gure and reference objects was identical in half the
trials and different in the other half. The four samples in which the ®gure was moved
relative to the reference object were created by translating the ®gure object one-
fourth of a grid space (0.125 inches, 0.32 cm) from its initial position in each
diagonal direction. On trials in which the ®gure object initially contacted the refer-
ence object, two of the ªdifferentº trials were produced by translating the ®gure
object along the edge of the reference object by the same distance as the other
distractors.
2.4. Results and discussion
2.4.1. Language task
The results of the coding were separately analyzed for effects concerning axial
terms and those concerning contact terms.
2.4.1.1. Axial terms. Table 1(a±d) shows the proportions of use of basic axial terms
for each of the 72 possible ®gure object locations, with a total possible of two per
location (one response for each of the two ®gure objects). Table 1(a,b) displays
proportions of use of vertical terms by Japanese and English speakers, respectively;
Table 1(c,d) displays the corresponding proportions of use of horizontal terms. Both
E. Munnich et al. / Cognition 81 (2001) 171±207 183
10
Our decision to analyze simple expressions is based on the fact that the most focal instances of
categories tend to be described with the monomorphemic expressions of the language (see, for example,
Fodor, 1981; Heider, 1972). Nevertheless, in order to determine whether a different pattern of spatial
structure would emerge if we also considered complex expressions, we carried out an analysis of vertical
and horizontal terms wherever they were used ± whether in simple or in complex expressions. In this
analysis, participants showed a complementary pattern of naming to the one reported for simple expres-
sions; that is, they almost always used expressions involving some vertical term (e.g. ªabove and to the
leftº, ªto the left and aboveº, or simply ªaboveº) in all locations except those on and adjacent to the
horizontal axis. Similarly, they almost always used expressions involving horizontal terms in all locations
except those on or adjacent to the vertical axis. Since this pattern is merely the complement of the one we
report, it would have been redundant to present it as well.
E. Munnich et al. / Cognition 81 (2001) 171±207184
Table 1
Experiment 1: Proportion of vertical and horizontal terms used by Japanese and English speakers
a
Row Column
1234 5 6 789
(a) Proportion of vertical terms used by Japanese speakers
1 0.10 0.10 0.14 0.25 1.00 0.25 0.10 0.10 0.10
2 0.10 0.10 0.10 0.21 1.00 0.20 0.15 0.10 0.10
3 0.10 0.10 0.10 0.25 1.00 0.15 0.10 0.10 0.10
4 0.00 0.00 0.00 0.05 0.00 0.00
5 0.00 0.05 0.00 Reference object 0.00 0.00 0.00
6 0.00 0.00 0.00 0.00 0.05 0.00
7 0.10 0.10 0.10 0.10 1.00 0.10 0.10 0.10 0.10
8 0.15 0.10 0.10 0.25 1.00 0.20 0.10 0.10 0.10
9 0.10 0.10 0.15 0.20 1.00 0.29 0.15 0.10 0.10
(b) Proportion of vertical terms used by English speakers
1 0.00 0.00 0.05 0.35 0.95 0.35 0.00 0.00 0.05
2 0.00 0.00 0.10 0.35 1.00 0.35 0.00 0.00 0.00
3 0.00 0.10 0.05 0.25 1.00 0.30 0.00 0.00 0.00
4 0.05 0.00 0.00 0.00 0.00 0.00
5 0.00 0.00 0.00 Reference object 0.00 0.00 0.00
6 0.15 0.00 0.00 0.00 0.00 0.00
7 0.00 0.00 0.00 0.25 0.85 0.30 0.00 0.00 0.00
8 0.00 0.00 0.00 0.35 0.90 0.25 0.15 0.00 0.00
9 0.05 0.00 0.00 0.45 0.90 0.44 0.05 0.00 0.00
(c) Proportion of horizontal terms used by Japanese speakers
1 0.00 0.05 0.00 0.05 0.00 0.05 0.00 0.00 0.00
2 0.05 0.05 0.05 0.00 0.00 0.05 0.00 0.00 0.00
3 0.05 0.10 0.00 0.05 0.00 0.00 0.10 0.10 0.00
4 0.40 0.30 0.30 0.15 0.45 0.40
51.00 0.95 1.00 Reference object 1.00 1.00 1.00
6 0.25 0.50 0.20 0.20 0.30 0.30
7 0.00 0.05 0.15 0.05 0.00 0.00 0.05 0.05 0.15
8 0.05 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
9 0.05 0.05 0.05 0.05 0.00 0.00 0.00 0.00 0.00
(d) Proportion of horizontal terms used by English speakers
1 0.00 0.00 0.05 0.00 0.00 0.00 0.00 0.00 0.10
2 0.05 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.05
3 0.10 0.15 0.00 0.00 0.00 0.00 0.00 0.10 0.10
4 0.45 0.40 0.30 0.31 0.40 0.45
50.90 1.00 0.90 Reference object 0.85 1.00 1.00
6 0.45 0.50 0.30 0.40 0.55 0.45
7 0.10 0.05 0.00 0.00 0.00 0.00 0.05 0.10 0.10
8 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.10
9 0.00 0.05 0.00 0.00 0.00 0.00 0.00 0.00 0.00
a
Data are listed by the location of the ®gure object relative to the reference object. The locations are
laid out in a 9 £9 grid, with the central 3 £3 spaces taken up by the reference object (see Fig. 2). Note that
locations where an effect is predicted are indicated in bold.
language groups used axial terms most frequently when the ®gure object lay on an
(imagined) axis extending from the reference object, and these responses fell off as
the ®gure object's location moved away from the axes. In addition, the use of these
basic axial terms appeared almost categorical: use of axial terms was close to 100%
along the projected axes, and fell off immediately thereafter. The pattern across
languages is extremely similar to that found by Hayward and Tarr (1995).
Separate but analogous analyses were carried out for the vertical and horizontal
axial terms. First, we analyzed the proportions of vertical terms elicited across the
nine columns, for those locations lying above or below the reference object (i.e.
excluding portions of the columns falling in rows 4±6, which intersect the reference
object). In order to code distance from the reference object, locations falling directly
on the axis (Column 5 in Table 1) were designated as ª0º (ªzeroº), locations one
column away (Columns 4 and 6 in Table 1) were designated as ª1º, locations two
columns away (Columns 3 and 7 in Table 1) were designated as ª2º, etc. (see Fig.
3a). The proportions of vertical terms used for these locations were submitted to a 2
(language) £5 (distance from axis, 0±4) mixed analysis of variance, with the second
factor within-subjects. For horizontal terms, an analogous design was used, with
locations falling directly on the horizontal axis designated as ª0º, locations one row
away designated as ª1º, etc. (see Fig. 3b), and a separate 2 (language) £5 (distance
from axis) analysis of variance was carried out.
Only the effects of distance from axis were reliable (vertical region:
E. Munnich et al. / Cognition 81 (2001) 171±207 185
Fig. 3. (a,b) Regions used in analysis of axial effects for both Language and Memory tasks in Experiment 1.
Regions tested for vertical effects are shown in (a), and regions tested for horizontal effects are shown in (b).
F4;7284:95; horizontal region: F4;72112:14; both P,0:01). Neither the
main effects of language nor the interactions of language with distance from axis
were signi®cant. Post-hoc tests were carried out between levels of distance from
vertical and horizontal axes to determine where the drop-off in use occurred. There
were reliable differences between locations on-axis (0) and all rows/columns off-
axis (1±4), as well as between distances 1 and 2, but there were no further differences
among distances 2, 3, and 4 (Tukey's HSD 0:15 for both vertical and horizontal
terms, P,0:05). Thus, the axial terms were used at the highest (ceiling) levels
along the reference object's axis, then dropped sharply until the reference object's
edge, after which they showed no further drop. These results suggest strong effects
of the reference object's axis, equally for speakers of both languages.
2.4.1.2. Contact terms. Table 2(a,b) displays the mean proportions of use of contact
terms by position of the ®gure object for Japanese and English speakers,
respectively. Across the two languages, contact terms were used predominantly in
locations adjacent to the reference object. However, there was also a difference in
the distribution of the terms across the languages. Japanese speakers' use of contact
E. Munnich et al. / Cognition 81 (2001) 171±207186
Table 2
Experiment 1: Proportion of terms indicating contact used by Japanese and English speakers (by position
of ®gure object)
a
Row Column
1234 5 6 789
(a) Proportion of terms indicating contact used by Japanese speakers
1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.05 0.00
2 0.05 0.00 0.00 0.00 0.05 0.00 0.05 0.00 0.00
3 0.05 0.05 0.40 0.85 0.85 0.90 0.40 0.05 0.00
4 0.00 0.00 0.90 0.89 0.05 0.00
5 0.00 0.00 0.95 Reference object 0.65 0.00 0.05
6 0.00 0.10 0.75 0.80 0.05 0.05
7 0.05 0.10 0.45 0.85 0.85 0.85 0.40 0.10 0.05
8 0.05 0.00 0.00 0.00 0.00 0.00 0.10 0.00 0.10
9 0.00 0.05 0.00 0.05 0.00 0.00 0.00 0.05 0.00
(b) Proportion of terms indicating contact used by English speakers
1 0.00 0.05 0.00 0.10 0.00 0.05 0.10 0.00 0.05
2 0.00 0.00 0.00 0.05 0.10 0.05 0.00 0.05 0.05
3 0.00 0.00 0.05 0.50 0.55 0.55 0.10 0.00 0.00
4 0.00 0.00 0.20 0.31 0.00 0.00
5 0.00 0.00 0.15 Reference object 0.20 0.00 0.00
6 0.00 0.00 0.20 0.15 0.00 0.00
7 0.05 0.00 0.10 0.15 0.10 0.20 0.10 0.00 0.00
8 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
9 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
a
Note that locations where an effect is predicted are indicated in bold.
terms (verbs only) appears to be roughly symmetrical across the four sides of the
reference object (Table 2(a)) while English speakers' use of contact terms is
asymmetrical, appearing predominantly on the upper side of the reference object
(Table 2(b)).
In order to evaluate these effects, the three ®gure object locations contacting each
side of the reference object were collapsed and compared to corresponding locations
further out in each direction, designating locations adjacent to the reference object as
ª0º, and those one or two grid spaces away from the reference object as ª1º and ª2º,
respectively (see Fig. 4 for groupings of locations). Thus, analyses were carried out
as 2 (language) £3 (distance from reference object, 0±2) £4 (side of reference
object) mixed analyses of variance, with the last two factors within-subjects.
The analysis of variance showed signi®cant main effects of language
(F1;1823:49), as well as distance (F2;3677:81), and side
(F3;544:42) (all P,0:01). Japanese speakers used more contact terms than
English speakers (29 vs. 10%, respectively, across all positions), and contact terms
were used only for locations adjacent to the reference object, shown by reliably
greater use of contact terms in these locations than those one or two grid spaces away
E. Munnich et al. / Cognition 81 (2001) 171±207 187
Fig. 4. Regions used in analysis of contact effects for both Language and Memory tasks in Experiment 1.
(Tukey's HSD 0:13 for Japanese, 0.22 for English, P,0:05; compare Table 2(a)
and (b)). Like the axial terms, use of contact terms was categorical, dropping sharply
in non-adjacent locations.
There were also two-way interactions of language with distance
(F2;3621:81), language with side (F3;544:21), and distance with side
(F6;1086:29), and a three-way interaction of language, distance, and side
(F6;1083:97) (all P,0:01). Planned comparisons were carried out examining
use of contact terms in adjacent vs. non-adjacent locations, for each language on
each side of the reference object. Japanese speakers used contact terms signi®cantly
more often in adjacent locations than in non-adjacent locations for each side (top:
F1;999:36; bottom: F1;9248:80; left: F1;9125:94; right:
F1;9107:93; all P,0:01). In contrast, English speakers used terms in adjacent
positions reliably more often than non-adjacent positions only on the top side
(F1;913:05, P,0:01). The distribution of use of contact terms in adjacent
positions was symmetrical around the reference object for Japanese speakers (top:
86.7%; bottom: 85%; left: 86.7%; right: 78%), but strongly biased in favor of the top
side for English speakers (top: 53.3%; bottom: 15.0%; left: 18.3%; right: 20.8%). It
is clear that neither group of speakers used contact terms obligatorily for the top side,
a point to which we return later.
To summarize the results of the Language task, speakers of both languages
showed the same strong categorical axial effects, with axial terms predominantly
falling directly on the axis. Participants showed somewhat weaker effects of contact/
support, and the distribution of response was somewhat different in Japanese than in
English.
2.4.2. Memory task
As in the Language task, separate analyses were conducted to examine possible
effects of the reference object's axial structure, and effects of contact with the
reference object. In particular, we sought to determine whether the effects emer-
ging in the Language task would also be found in this non-linguistic spatial
memory task.
2.4.2.1. Axial effects. The mean proportions correct for each location are shown in
Table 3(a,b). These show graded effects in accuracy, with the highest accuracy
across both language groups occurring for locations on or around the reference
object's axes. In order to analyze accuracy, locations were grouped into the same
rows and columns as in the Language task, and analyses were carried out on the
mean proportions correct, by distance from each axis, and separately for distance
from the vertical vs. horizontal axes. We employed two separate 2 (language) £5
(distance from axis, 0±4) mixed analyses of variance, each with the second factor
within-subjects.
Both analyses revealed signi®cant main effects of language (for the vertical region:
F1;386:00; for the horizontal region: F1;386:15; both P,0:02), re¯ect-
ing the generally lower accuracy of Japanese speakers compared to English speakers
(M percents correct 64 vs. 67%, respectively, for the vertical regions and 64 vs. 68%
E. Munnich et al. / Cognition 81 (2001) 171±207188
for the horizontal regions).
11
There were also main effects of distance from each axis
(vertical region: F4;152221:08; horizontal region: F4;152245:41; both
P,0:01) and an interaction of language with distance from the horizontal axis
(F4;1523:17, P,0:02) re¯ecting faster drop-off of accuracy from the horizon-
tal axis among Japanese speakers. Planned comparisons were carried out for each
language group between locations on the axis (0) and locations on the surrounding
columns/rows (as in the Language task, collapsing one, two, three, or four grid spaces
off-axis, i.e. columns 4 and 6, 3 and 7, 2 and 8, and 1 and 9). Accuracy was signi®-
cantly higher on-axis than off-axis among both language groups, and for both axes
(Vertical Axis, Japanese: F1;1961:80; English: F1;19179:06; Horizontal
Axis, Japanese: F1;1925:01; English: F1;1936:20; all P,0:01). This
effect replicates the ®ndings of Hayward and Tarr (1995).
However, closer inspection suggested a more complex picture than just a strict
superiority for locations directly on the reference object's axes. For example,
accuracy for locations directly on the vertical axis did not appear to be different
from accuracy on the pair of directly adjacent columns (one off-axis, or columns 4
and 6 in Table 3(a,b)). We therefore conducted post-hoc tests between levels of
E. Munnich et al. / Cognition 81 (2001) 171±207 189
Table 3
Experiment 1: proportion correct for Japanese and English speakers in the Memory task (by position of
®gure object)
a
Row Column
1234 5 6 789
(a) Proportion correct for Japanese speakers in the Memory task
1 0.52 0.54 0.60 0.59 0.66 0.59 0.58 0.52 0.51
2 0.50 0.51 0.61 0.73 0.71 0.69 0.57 0.58 0.55
3 0.59 0.56 0.85 0.93 0.91 0.93 0.84 0.62 0.55
4 0.64 0.75 0.93 0.90 0.73 0.69
50.60 0.62 0.86 Reference object 0.84 0.63 0.60
6 0.64 0.76 0.92 0.95 0.73 0.67
7 0.57 0.59 0.85 0.89 0.90 0.93 0.85 0.60 0.56
8 0.49 0.53 0.65 0.73 0.67 0.69 0.60 0.56 0.57
9 0.50 0.56 0.57 0.58 0.66 0.65 0.57 0.53 0.56
(b) Proportion correct for English speakers in the Memory task
1 0.52 0.51 0.58 0.70 0.77 0.62 0.57 0.53 0.56
2 0.55 0.54 0.66 0.78 0.80 0.72 0.61 0.53 0.56
3 0.63 0.65 0.89 0.97 0.93 0.93 0.89 0.67 0.60
4 0.72 0.80 0.95 0.96 0.84 0.74
50.60 0.71 0.93 Reference object 0.92 0.64 0.60
6 0.72 0.83 0.98 0.97 0.83 0.74
7 0.60 0.66 0.91 0.95 0.95 0.93 0.90 0.69 0.59
8 0.54 0.57 0.64 0.73 0.74 0.73 0.66 0.57 0.53
9 0.51 0.53 0.61 0.71 0.70 0.63 0.57 0.53 0.53
a
Note that locations where an effect is predicted are indicated in bold.
distance from both the vertical and horizontal axes. Considering ®rst the vertical
axis, there were no reliable differences between accuracy on-axis and that of the
directly adjacent columns (one off-axis, i.e. columns 4 and 6). However, there were
reliable differences in accuracy between locations on-axis and subsequent adjacent
columns (two, three, and four off-axis, i.e. columns 3 and 7, 2 and 8, and 1 and 9),
with accuracy dropping in each case except the last (HSD 0:04 for Japanese,
0.03 for English, P,0:05). One possible explanation for the lack of drop between
the axis and locations one off-axis is that there may have been an effect of align-
ment with the reference object, rendering all locations in this region equally
memorable. In fact, a number of participants in the Memory task reported that
they found it easier to remember the position of the ®gure object when it was
aligned with imagined extensions of the reference object's edges.
Parallel analyses for the horizontal axis showed a similar pattern. Accuracy on the
axis was reliably lower than that of rows adjacent to the axis (one off-axis, or rows 4
and 6), equal to the accuracy for locations two off-axis (rows 3 and 7), but reliably
higher than accuracy for subsequent rows (three and four off-axis, or rows 2 and 8,
and 1 and 9; HSD 0:04 for both English and Japanese, P,0:05). The increased
accuracy in positions ¯anking the horizontal axis is again suggestive of an effect of
alignment with the reference object's edges.
In summary, the accuracy analyses revealed clear evidence of some priority of the
locations on-axis or one off-axis, compared to the surrounding locations. However,
the results were not categorical as they had been in the Language task, where the
locations on-axis were uniformly named by the basic spatial terms but use of these
terms fell off sharply away from the axes. Accuracy in the Memory task, by contrast,
revealed a more graded pattern, with possible effects of alignment with the reference
object, which were not seen at all in the Language task.
12
2.4.2.2. Contact effects. As Table 3(a,b) shows, contact between the ®gure and
reference objects was very important to people in the Memory task: locations
where the ®gure object was adjacent to the reference object were remembered
much more accurately than those not adjacent. A 2 (language) £3 (distance from
reference object) £4 (side of reference object) mixed analysis of variance with the
last two factors within-subjects showed signi®cant effects of language
(F1;386:69, P,0:02) and distance (F1;38984:24, P,0:01). Japanese
speakers were overall less accurate than English speakers (75 vs. 80%, respectively),
as in the analysis of axial positions. Post-hoc tests showed that participants were
reliably more accurate for positions adjacent to the reference object than for those
one grid space away, and were more accurate for these than for positions two grid
E. Munnich et al. / Cognition 81 (2001) 171±207190
11
Despite efforts to match language groups by level of education, English-speaking participants in the
Memory task of Experiment I had a higher level of education on average than did their Japanese-speaking
counterparts. For this reason, separate analyses were carried out for a subset of our participants, who were
more rigorously matched for education level. These analyses revealed effects of both horizontal and
vertical axes and an effect of contact, which were comparable to those found for the whole sample;
however, for this matched subset, there were neither main effects of language group nor any interactions
with language group.
spaces away (Tukey's HSD 0:02, P,0:05). There was no effect of side, nor any
interactions.
This pattern suggests an effect of distance which is graded, much like the results
of the Axial analyses for the Memory task. Again, these effects of contact are quite
different from those in the Language task, which were categorical.
2.5. Summary of results from Experiment 1
The results showed both similarities and differences across language group and
task. In the Language task, both native English and Japanese speakers showed
striking categorical use of axial terms, as well as categorical use of terms expressing
contact with the reference object. The axial effect for English speakers is virtually
identical to that found by Hayward and Tarr (1995), and the extension to Japanese
speakers suggests that the organization of these terms is quite similar in both
languages. The effects for contact terms were categorical for both languages, but
there were also differences in application of the terms: whereas English speakers
used more contact terms on the top side of the reference object, Japanese speakers
used contact terms symmetrically across all sides of the reference object.
The similarities in use of language across English and Japanese mapped clearly ±
but not exactly ± onto patterns of performance in the Memory task: both language
groups were most accurate in their memory for those locations named most consis-
tently by the basic spatial terms, both axial and contact. However, the results for the
Language task showed categorical effects whereas the results for the Memory task
showed more graded effects. Moreover, differences in language use across English
and Japanese did not map directly onto the patterns found in the Memory task. The
only cross-linguistic difference in use of spatial terms ± the distribution of contact
terms around the reference object ± was not re¯ected in analogous differences across
language groups in the Memory task.
However, the lack of such an effect may have been due to the nature of the scenes
we used. The scenes we used did not elicit obligatory coding of contact by native
English speakers. Obligatory coding by English speakers along with non-obligatory
coding by Japanese speakers are prerequisites for demonstrating any linguistically-
based difference in memory. Anecdotally, many participants thought of the scenes as
two-dimensional, which is incompatible with construal of support relationships. Lack
of a compelling support relationship could have weakened the difference between
English and Japanese speakers' linguistic coding of the scene, and thereby obscured
E. Munnich et al. / Cognition 81 (2001) 171±207 191
12
One more analysis was conducted to examine the possibility that people's memory for location can
ªmigrateº towards canonical spatial organizers (as found by Huttenlocher, Hedges, & Duncan, 1991; also
see Rosch, 1973). If this was a factor in our experiment, and if axes were organizers in memory, then
distractors moving towards the axis might produce more errors (i.e. more erroneous ªsameº) than those
which moved away from the axis. We therefore compared error rates across the two types of distractor for
each language group. However, there were no systematic patterns of migration. This null result may have
been due to the diagonal direction along which the distractors moved. We return to the issue of error
patterns in Experiment 2.
any effects in the non-linguistic task. We examined this possibility in Experiment 2,
by using scenes that much more clearly depicted support relationships. In addition, we
tested the generality of the ®ndings of Experiment 1 by extending our comparison to
that between native English speakers and native speakers of Korean which, like
Japanese, does not have an obligatory contrast between ªonº and ªaboveº.
3. Experiment 2
Using scenes containing balls and cups resting on or hovering above a table, we
tested participants' naming and memory for locations in these regions just as we
had in Experiment 1. Clear portrayal of such support (and non-support) relation-
ships should induce a sharp linguistic contrast for English speakers, but not for
speakers of Korean. Such a difference between the lexicons of the two languages
might be accompanied by correlated effects on memory, speci®cally with different
memorial structures arising for speakers of the two languages in a non-linguistic
task. Alternatively, speakers of the two languages might differ largely on the
linguistic task alone, with memorial structures remaining constant despite years
of language use.
3.1. Participants
Twenty native Korean speakers and 20 native English speakers participated, with
each group roughly balanced for gender. Within each language group, no participant
had been exposed to languages other than their native language before the age of 12.
English participants were undergraduate students at the University of Delaware, and
Korean participants were current students or graduates of universities in Korea who
were studying at the University of Delaware.
3.2. Design and materials
Ten participants from each language group participated in the Language task, and
the remainder participated in the Memory task. The design and tasks were identical
to those in Experiment 1, with the following exceptions.
First, the stimulus scenes displayed a cup or a basketball as the ®gure object and
a table as the reference object (see Fig. 5). Pre-testing showed that these objects
elicited a strong perception of support for the appropriate locations (i.e. those in
which the cup or ball were ªonº the table). For each task, participants viewed a
total of 25 different ®gure±reference object relationships, each presented with the
cup and the ball, for a total of 50 scenes. The ®gure objects occupied one of the 25
positions in a 5 £5 grid (with each grid space 0.5 inches (1.27 cm) square) which
lay adjacent to the top surface of the reference object. The grid spaces were the
same size as in Experiment 1, with a new column or row of squares overlapping
every 0.25 inches (0.64 cm). This design was used to check whether the same
distinctions would emerge in the Memory task with more ®ne-grained positions. In
the Language task, the 50 scenes were presented once each. Participants were
E. Munnich et al. / Cognition 81 (2001) 171±207192
again asked to view each display and describe the relationship between ®gure and
reference objects, but this time were given sentences of a slightly different form:
The cup is __________ the table:
13
In the Memory task, each scene was presented for twelve trials (six ªSameº, six
ªDifferentº), for a total of 600 trials. There were also several modi®cations made to
the distractors used in the Memory task. ªDifferentº locations were displaced hori-
zontally (left, right) and vertically (up only) from their initial position, rather than
diagonally as in Experiment 1. This change was made so that we could more directly
examine patterns of errors pertinent to possible cross-linguistic effects on memory.
In particular, this made it possible to examine whether obligatory marking of contact
in English would lead to sharper contact contrasts in the Memory task. Distractors
were displaced by 0.125 inches (0.32 cm, as were used in Experiment 1) for half of
the trials, and by 0.0625 inches (0.16 cm) for the remaining half of the trials
(randomly intermixed). While performance was better overall for the larger displa-
cements, the patterns of axial and contact effects were the same in both cases. We
therefore combined data from the two levels of displacement.
Results for the Language task were coded using the same scheme as in Experiment
1. A native English speaker and a native Korean speaker coded the data in their
respective languages. Twenty percent of the responses in each language group
were independently coded by another native English and native Korean speaker,
and the intercoder agreement scores were 99 and 97%, respectively. As was the
case with Japanese speakers, Korean speakers used dictionary equivalents of the
English axial terms widely, invoking these axial terms in 97% of their descriptions
of axial relationships. In addition, as with Japanese speakers, these particular terms
were the only means used by Korean speakers to describe axial relationships, indicat-
ing their primary status in the language. Finally, like the Japanese speakers, Korean
speakers do not have an exact equivalent of ªonº at their disposal, but they used
equivalents of English verbs of contact and support, which are listed exhaustively
in Appendix A.
3.3. Results and discussion
3.3.1. Language task
3.3.1.1. Axial effects The mean proportions of axial terms were computed as in
Experiment 1 (see Fig. 6 for coding groups), and are shown in Table 4(a,b) for
Korean and English, respectively. The data show a strong categorical effect of axis,
as in Experiment 1, with dense use of axial terms along the axis, but not in the off-
axis positions. Analyses were carried out on these data using a 2 (language) £3
(distance from axis, 0±2) mixed analysis of variance with the second factor within-
E. Munnich et al. / Cognition 81 (2001) 171±207 193
13
The equivalent sentence was used in the Korean version of the task: ªball-i thakca __________.º
(Actual stimuli were printed in Korean script.)
subjects. There was a reliable main effect of distance from axis (F2;3645:32,
P,0:01), but no main effect of language (F1;182:72, P.0:05) nor any
interaction of language with distance. Locations on-axis elicited reliably more
axial terms than either locations one or two grid spaces away for English and Korean
separately and for the data collapsed. There were no differences between locations
one and two grid spaces from the axis (Tukey's HSD 0:18, P,0:05). These
effects for axial terms replicate those found in Experiment 1.
3.3.1.2. Contact effects. The mean proportions of contact terms were computed as in
Experiment 1 (see Fig. 7 for coding groups), and are shown in Table 5(a,b) for Korean
and English, respectively. A 2 (language) £5 (distance from reference object, 0±4)
mixed analysis of variance was carried out with the second factor within-subjects,
revealing reliable main effects of language group (F1;1834:13) and distance
(F4;72111:88) (both P,0:01). English speakers used more contact terms than
Korean speakers overall (20 vs. 6%). Post-hoc tests between levels of distance taken
pairwise showed reliably greater use of contact terms in the adjacent locations than in
any others, and no differences among the non-adjacent positions (HSD 0:04 for
English, 0.21 for Korean, P,0:05).
There was also a reliable interaction between language and distance
(F4;7235:96, P,0:01). Planned comparisons evaluating the difference
between language groups in their use of contact terms showed a difference only
for adjacent positions, with English speakers providing more contact terms than
Korean speakers (97 vs. 27%, respectively) (F1;1836:09, P,0:01). In fact,
the distinction was categorical among English speakers, moving from ceiling in the
adjacent locations to zero in non-adjacent locations. Notably, while English speak-
ers used contact terms uniformly in contact positions, only ®ve Korean speakers
E. Munnich et al. / Cognition 81 (2001) 171±207194
Fig. 5. Example of a display used in Experiment 2.
used them at all. Thus, the scenes of balls and cups contacting or suspended above a
table appeared to elicit the obligatory ªonº/ªaboveº distinction in English, but not in
Korean.
Finally, there were relatively few axial terms used by either language group in
contact positions. Instead of using contact terms, English speakers often refer to
contact locations as simply ªonº with no axial component (e.g. simply ªonº, rather
than ªon top ofº). This is consistent with the general pattern expected among English
speakers of expressing support obligatorily ± in this case, the importance of support
apparently trumps any consideration of axial status.
3.3.2. Memory task
3.3.2.1. Axial effects The mean proportions correct by position of the ®gure
object are shown in Table 6(a,b) for Korean and English, respectively.
14
These
E. Munnich et al. / Cognition 81 (2001) 171±207 195
Fig. 6. Regions used in analysis of axial (vertical only) effects for both Language and Memory tasks in
Experiment 2.
14
The overall proportion correct in the Memory task of Experiment 2 was signi®cantly lower than that of
Experiment 1 (F1;58111:97, P,0:01). This appears to be due to differences in the distractors that
we used in each experiment. In Experiment 1, the distractors were always diagonally displaced from the
target, whereas in Experiment 2 they were displaced horizontally (left or right) or vertically upwards (see
Section 3.2 for rationale). An inspection of the proportions correct by direction reveals that accuracy was
much greater when the distractor was displaced vertically. Since we eliminated the downward distractors
in Experiment 2, it is therefore not surprising that proportions correct would fall. In accord with the
explanation, this drop in accuracy occurred throughout the space we sampled.
data were submitted to a 2 (language) £3 (distance from axis, 0±2) mixed analysis
of variance, with the second factor within-subjects. In contrast to the Memory task in
Experiment 1, there were no reliable effects of axis. However, as discussed with
respect to Experiment 1, the axial effects may partly re¯ect enhancement of the
entire region aligned with the reference object. Because we did not test locations
outside of the reference object's edges, we cannot rule out this type of regional
effect. The results of the Memory task also differed from those of the Language task,
which showed axial effects for speakers of both languages.
3.3.2.2. Contact effects. Mean proportions correct were submitted to a 2
(language) £5 (distance from reference object, 0±4) analysis of variance with the
second factor within-subjects, revealing only an effect of distance
(F4;7233:15, P,0:01). Post-hoc comparisons revealed that location 0
(adjacent to the reference object) elicited reliably greater accuracy than any other
locations; non-adjacent locations did not differ from each other (Tukey's
HSD 0:04, P,0:05). Unlike the results of Experiment 1, which showed that
accuracy declined in a graded fashion with increasing distance from the reference
object, in this case the effect of distance appears to be entirely accounted for by the
drop in accuracy between adjacent and immediately non-adjacent positions.
Further analyses were carried out to determine whether distractors that crossed a
category boundary elicited higher accuracy than those that did not cross a category
boundary. Speci®cally, we compared four cases (shown in Fig. 8). For targets that
E. Munnich et al. / Cognition 81 (2001) 171±207196
Table 4
Experiment 2: proportion of vertical terms used by Korean and English speakers
a
Row Column
12345
(a) Proportion of vertical terms used by Korean speakers
1 0.05 0.30 0.95 0.30 0.05
2 0.05 0.25 1.00 0.25 0.05
3 0.00 0.20 0.75 0.20 0.00
4 0.00 0.25 0.90 0.35 0.00
5 0.05 0.20 0.65 0.30 0.00
Reference object
(b) Proportion of vertical terms used by English speakers
1 0.45 0.35 0.95 0.40 0.35
2 0.30 0.40 1.00 0.50 0.30
3 0.35 0.50 0.95 0.50 0.35
4 0.45 0.45 1.00 0.55 0.40
5 0.40 0.40 0.45 0.30 0.25
Reference object
a
Data are listed by the location of the ®gure object relative to the reference object. The locations are
laid out in a 5 £5 grid, with the bottom side of the grid touching the reference object. Note that locations
where an effect is predicted are indicated in bold.
were initially in contact with the reference object, distractors could have moved (a)
vertically out of contact or (b) horizontally, maintaining contact. For targets that
were initially out of contact with the reference object, distractors could have moved
(c) vertically, remaining out of contact or (d) horizontally, remaining out of contact.
The critical question is whether there was a difference between English and Korean
speakers in their memory for locations that broke the contact/non-contact boundary
(distractor a), compared to distractors that did not break this boundary (b, c, or d). If
there is an effect of linguistic coding on memory, then the obligatory linguistic
marking of the contact/non-contact distinction by English speakers would be
expected to enhance this distinction in the Memory task.
The results of a 2 (language) £2 (initially in contact/out of contact) £2 (vertical
vs. horizontal distractor movement) mixed analysis of variance with the last two
factors within-subject showed main effects of contact (F1;1891:46) and direc-
tion of distractor movement (F1;18107) and an interaction between the two
(F1;18127:25) (all P,0:01). There was neither an effect of language group
nor any interactions with it. Planned comparisons showed greater accuracy detect-
ing distractors that broke the contact/non-contact boundary (i.e. the a distractors)
than for each distractor type that did not break this boundary (a vs. b, a vs. c, and a
vs. d; F1;18125:59, 138.25, and 138.55, respectively, all P,0:01).
E. Munnich et al. / Cognition 81 (2001) 171±207 197
Fig. 7. Regions used in analysis of contact effects for both Language and Memory tasks in Experiment 2.
3.4. Summary of results from Experiment 2
The Language task showed strong and categorical effects of the reference object's
axes among both English and Korean speakers. This replicates and extends the
language ®ndings of Experiment 1. However, unlike Experiment 1, these axial
effects were not re¯ected at all in the results of the Memory task. The Language
task also showed cross-linguistic differences in the coding of contact/support rela-
tionships: English speakers obligatorily coded the distinction between contact/
support and non-contact, but Korean speakers did not. No such difference appeared
in the Memory task, indicating that obligatory coding of support does not lead to
better memory for contact/support positions.
4. General discussion
Across the two experiments, we found evidence for cross-linguistic similarities
and differences in various aspects of spatial language. We also found similarities in
spatial memory across different language groups. These similarities emerged despite
the cross-linguistic differences in spatial language, suggesting that spatial memory is
not affected by differences in how languages encode location. As a whole, the
®ndings suggest that both language and memory draw on the same kinds of spatial
properties, including axial structure and contact/support. However, these properties
do not appear to be invoked in mandatory fashion by all languages in all memory
tasks, suggesting that the relationship between the two systems is not a simple one.
E. Munnich et al. / Cognition 81 (2001) 171±207198
Table 5
Experiment 2: proportion of terms indicating contact used by Korean and English speakers (by position of
®gure object)
a
Row Column
12345
(a) Proportion of terms indicating contact used by Korean speakers
1 0.00 0.00 0.00 0.00 0.00
2 0.00 0.00 0.00 0.00 0.00
3 0.00 0.00 0.00 0.00 0.00
4 0.00 0.00 0.00 0.00 0.05
50.30 0.25 0.20 0.30 0.30
Reference object
(b) Proportion of terms indicating contact used by English speakers
1 0.00 0.05 0.00 0.00 0.00
2 0.00 0.00 0.00 0.00 0.00
3 0.00 0.00 0.00 0.00 0.00
4 0.00 0.00 0.00 0.00 0.00
51.00 1.00 0.90 0.95 1.00
Reference object
a
Note that locations where an effect is predicted are indicated in bold.
4.1. The structure of spatial language and spatial memory: similarities and
differences
Two aspects of spatial language were studied ± axial structure and contact/support.
In both experiments, and across three languages, axial terms were used quite simi-
larly, providing a cross-linguistic replication of the key language ®ndings reported by
Hayward and Tarr (1995). In English, Japanese, and Korean, axial terms were applied
most consistently along the axes of the reference objects, and the terms were used
categorically, dropping sharply outside of the region of the axis. These ®ndings show
that axial structure plays an important role in a variety of languages. Further, the fact
that this structure was present in all three languages suggests the possibility that it may
be an obligatory property of spatial language, encoded in all languages.
Axial structure was also re¯ected in the results of the Memory task of Experiment
1, suggesting that spatial memory engages some of the same spatial properties as
spatial language. Axes of the reference object played a crucial role in both talking
about location and remembering location. This ®nding is consistent with that of
Hayward and Tarr, as well as numerous other ®ndings that show a strong role for
axes in tasks ranging from attentional tasks to naming tasks (Carlson-Radvansky &
Irwin, 1993; Logan, 1995). However, in our results, the parallel between memory
and language was far from perfect. First, whereas the effects in the Language task
were categorical, those for the Memory task were graded in both English and
Japanese. Second, the axial effects appeared to be modulated by a number of factors.
For example, while we observed increased accuracy on the axis itself relative to
E. Munnich et al. / Cognition 81 (2001) 171±207 199
Table 6
Experiment 2: proportion correct for Korean and English speakers in the Memory task (by position of
®gure object)
a
Row Column
12345
(a) Proportion correct for Korean speakers in the Memory task
1 0.52 0.54 0.55 0.55 0.53
2 0.54 0.56 0.57 0.53 0.53
3 0.55 0.55 0.56 0.55 0.53
4 0.55 0.55 0.57 0.56 0.57
50.67 0.65 0.69 0.66 0.69
Reference object
(b) Proportion correct for English speakers in the Memory task
1 0.51 0.52 0.56 0.56 0.52
2 0.52 0.54 0.56 0.57 0.54
3 0.55 0.56 0.55 0.55 0.55
4 0.52 0.53 0.56 0.56 0.54
50.66 0.64 0.71 0.61 0.66
Reference object
a
Note that locations where an effect is predicted are indicated in bold.
locations in the periphery, there were also enhancing effects for locations that were
off-axis but aligned with the reference objects' edges ± these effects of alignment
were not found by Hayward and Tarr. This difference is directly attributable to
differences in the testing arrays: in Hayward and Tarr's experiments, locations
directly on-axis were also aligned with the two edges of the reference object,
confounding the two effects. As another example, we found no axial effects at all
in the Memory task of Experiment 2. We speculate that this re¯ects the three-
dimensional context of balls, cups, and tables, which strongly elicited the perception
of support, thereby inviting people to encode location in terms of contact and
support rather than the axial structure of the reference object. In short, although
axial structure seems to play an important role in spatial memory and shows parallels
to the structures engaged for spatial language, the underlying spatial organizations
that can be used in memory tasks are clearly quite ¯exible, varying depending on the
conceptual context and corresponding functional requirements of any given task.
Contact and support were also re¯ected in the Language and Memory tasks,
although, as we expected, there were clear cross-linguistic differences in the obli-
gatory linguistic use of these terms. For example, in Experiment 1, both Japanese
and English speakers used contact terms largely in locations that did in fact contact
the reference object. However, whereas Japanese speakers used contact terms
symmetrically around the reference object, English speakers concentrated the
terms more heavily on the top side than on other sides. In Experiment 2, when
people's attention was focused strongly on support, English speakers obligatorily
E. Munnich et al. / Cognition 81 (2001) 171±207200
Fig. 8. Contrast used in analysis of categorical boundaries in Experiment 2. See text for further explana-
tion.
encoded contact and support using ªonº, and lack of contact/support using ªaboveº.
In contrast, Korean speakers encoded this distinction only rarely, re¯ecting the fact
that contact is not encoded obligatorily in their language. The Memory tasks of both
experiments showed strong effects of contact in all language groups, suggesting that
contact and/or support provide a strong organization for spatial memory. Thus,
contact and support appear to play important roles in both systems. Yet, this spatial
property appears to be optionally encoded by languages, as seen by the difference
between English and Korean.
Did the cross-linguistic differences in the use of these terms affect people's perfor-
mance in the Memory task? If differences in the basic distinctions of a language result
in differences in non-linguistic organization, then the differences in the language of
contact and support should have resulted in corresponding differences in the Memory
task. However, no such effects were found. The strongest test of this possibility
occurred in Experiment 2. There, English speakers categorically and obligatorily
distinguished between locations in which the ®gure object was supported by the
reference object and those in which it was located above the reference object. Korean
speakers did so only sporadically. These cross-linguistic differences were not,
however, mirrored by corresponding differences in the sharpness of the contact effects
in the Memory task. Rather, the differences in memorial accuracy for contact vs. non-
contact locations appeared consistently across the two language groups. The same
pattern of results for memory was found in Experiment 1, despite cross-linguistic
differences in the distribution of contact/support terms. Across the three language
groups, there were equally strong effects of contact with the reference object, with
contact enhancing memory for location. The lack of cross-linguistic differences in
memory suggests that memory, as measured in our task, is not susceptible to long-
term exposure to a particular spatial lexicon.
4.2. Effects and non-effects of cross-linguistic difference on spatial cognition
These ®ndings suggest that cross-linguistic differences in spatial language need
not have direct causal effects on the organization of spatial memory. How does this
conclusion square with other existing ®ndings on the language±thought relation-
ship? First, our non-effects are consistent with the results of two recent studies by
Malt, Gennari and colleagues. Malt, Sloman, Gennari, Shi, and Wang (1999) exam-
ined the domain of artifacts with Chinese, Spanish, and English speakers: partici-
pants named a variety of containers in a linguistic task, and sorted the same set of
containers into categories based on perceived similarity in a non-linguistic task.
Despite clear cross-linguistic differences in how the set of artifacts was carved up
in naming, Malt and colleagues found no corresponding differences in how the same
artifacts were carved up in the sorting task. Thus, linguistic differences did not cause
changes in perceived similarity. Subsequently, Gennari, Sloman, Malt, and Fitch
(2000) carried out parallel linguistic and non-linguistic tasks involving manner and
path of motion in a verb-framed language (Spanish) and a satellite-framed language
(English). They found the expected cross-linguistic differences in descriptions of
motion scenes, but no corresponding differences in patterns of recognition memory
E. Munnich et al. / Cognition 81 (2001) 171±207 201
or categorization of scenes. Again, linguistic differences could not have been the
basis of patterns observed in non-linguistic tasks.
As discussed in Section 1, other investigators have reported effects of cross-
linguistic differences on corresponding non-linguistic tasks. In the color domain,
Davidoff et al. (1999) found that memory for color varies in accord with the distinc-
tions made by a speaker's color lexicon, suggesting causal effects of language on
cognition. Similarly, Brown and Levinson (1993) and Pederson et al. (1998) report
that people can develop distinct biases for using particular reference frames in non-
linguistic spatial tasks if there are correspondingly strong biases in their language
(but see Li & Gleitman, 2000, for evidence that these biases can be induced in a
range of circumstances not related to language differences).
How can we reconcile reports of both effects and non-effects of cross-linguistic
differences on cognition? One possibility is that signi®cant task differences might be
responsible for these discrepancies in outcome. For example, the experiments of
Brown and Levinson, Pederson et al., and Davidoff et al. all used non-linguistic
tasks that allowed participants ample time to encode the stimuli, which could have
invited verbal encoding even in the absence of verbal responses. If people did verbally
encode location, this would naturally have led to results consistent with the preferred
linguistic categorizations. In contrast to these studies, our Memory task required
people to spatially encode objects' locations under very short exposure durations,
which made verbal encoding highly unlikely. Furthermore, our stimuli were followed
by visual masks and were displaced by a small amount in order to prevent visual
persistence. The spatial distinctions (between Same and Different locations) were
quite ®ne-grained, and thus ill-suited to making verbal distinctions. Finally, the
memory structures that were revealed in our data were not isomorphic to those elicited
in the Language task, indicating that they were not in fact mediated by verbal encod-
ing. These memory structures are clearly mental representations ± they were not
precise at every distance from the reference object, but rather, re¯ected mental orga-
nization in terms of the reference object's axes and support relationships. Cross-
linguistic differences appear not to have effects on these non-linguistic spatial repre-
sentations, suggesting a lower boundary on the locus of cross-linguistic effects.
A second possibility is that the particular spatial representations we investigated
are basic enough that they resist effects of language variation. Imai and Gentner
(1997) have recently suggested that effects of language on non-linguistic categor-
ization may only take place in the absence of strong universal tendencies. They offer
this argument as a way of understanding the simultaneous presence of positive and
negative effects of cross-linguistic differences on object categorization. Speci®cally,
Lucy (1992) and Imai and Gentner (1997) examined patterns of generalization in
object sorting tasks among English speakers vs. Yucatec-Mayan and Japanese
speakers. When given a sample object, and asked to categorize other objects on
the basis of their similarity to the sample, English speakers tended to generalize on
the basis of object shape, whereas Yucatec-Mayan and Japanese speakers sometimes
generalized on the basis of material. Both Lucy and Imai and Gentner attributed
these different patterns of non-linguistic sorting to a syntactic difference between
English and the other two languages, and in particular the obligatory English mark-
E. Munnich et al. / Cognition 81 (2001) 171±207202
ing of the distinction between objects and substances, which is not made in either
Yucatec-Mayan or Japanese.
15
In contrast, Imai and Gentner found that both Japa-
nese and English speakers tended to generalize on the basis of shape when grouping
ªcomplexº objects, i.e. rigid objects with complex shapes that are typical of most
artifacts. It was only for ªsimpleº objects ± simple geometric shapes with few parts ±
that the cross-linguistic difference emerged. This led them to propose the presence of
a universal bias towards grouping artifactual concrete objects by shape ± one which
would have dominated any cross-linguistic effects and resulted in shape-based judg-
ments for ªcomplexº (artifact-like) objects among both English and Japanese speak-
ers (see also Yoshida & Smith, 1999).
Both axial structure and contact/support are properties of spatial organization that
are likely to be foundational to both language and cognition. If these structures
constitute strong universals in spatial cognition, then linguistic variation on this
basic pattern might tend not to have any effects on non-linguistic spatial organiza-
tion. Similarly, perceptual similarity of containers, as well as manner and path of
motion events, as examined by Malt, Gennari, and colleagues, may also constitute
such strong universals. Just as the perception of complex object shape might
preempt any effects of cross-linguistic variation, these arguably universal dimen-
sions of spatial cognition might preempt such effects as well.
To conclude, our evidence reveals that spatial language and spatial memory
engage similar spatial properties ± axial structure and contact/support. At the
same time, where we did ®nd clear cross-linguistic differences, there was no
evidence of corresponding differences in the non-linguistic organization of space.
This indicates that, to the extent that there are similarities between spatial language
and non-linguistic spatial cognition, linguistic representations do not play a major
role in shaping non-linguistic representations. Finally, although similar properties
were re¯ected in both the Language and Memory tasks, these properties were not
mandatory across the Language or Memory tasks. Although axial structure emerged
in both language and memory in the ®rst experiment, it did not emerge across both
tasks in the second experiment. Similarly, although contact and support emerged in
both Memory tasks, it did not appear across speaker groups in the two Language
E. Munnich et al. / Cognition 81 (2001) 171±207 203
15
The complementary distribution of unitizers and pluralizers in English corresponds to the distinction
between count and mass nouns. Speci®cally, English requires use of the plural form to characterize
several animate entities or discrete objects (e.g. ªtwo turkeysº, ªtwo bananasº), but does not directly
pluralize nouns referring to materials (e.g. *ªtwo cottonsº, *ªtwo zincsº); instead, in order to quantify
materials English requires a ªunitizerº such as ªbasketº or ªjarº which takes on the plural form (e.g.
ªbasketsof cottonº, ªjarsof zincº) Yucatec-Mayan, on the other hand, uses unitizers for both objects and
materials, as in the following:
ka0a tuul uulum
twounitizerturkey
ªtwo turkeysº
(All examples borrowed from Lucy, 1992; note that the convention of denoting ungrammatical expres-
sions with an asterisk has been adopted here.)
tasks. This lack of a perfect correspondence between language and memory suggests
that the two systems draw somewhat independently on the same set of properties
(see Crawford et al., 2000, for more direct evidence in this direction). Although there
may be a universal base of properties for linguistic and non-linguistic tasks, a perfect
isomorphism clearly does not exist between the two systems.
How and why might language and memory have emerged to represent the same sets
of spatial properties? One possibility is that these similarities evolved by building
language ªon top ofº non-linguistic spatial representations. In this view, evolutionary
pressures inherent in the non-linguistic spatial domain might become internalized as
part of our representational systems (Shepard, 1984), and these might then give rise to
similarities in the structures of the language of space (see, for example, Landau &
Jackendoff, 1993; Pinker & Bloom, 1990). Another possibility is that functional
considerations pertaining to the task of locating objects ± either linguistic or non-
linguistic ± naturally give rise to the need for the same class of distinctions; this could
then lead to parallel emergence of systems that preserve the same kinds of spatial
distinctions (Freyd, 1983; Landau & Jackendoff, 1993). Whether searching for an
object, picking it up, or talking about its location, certain structures ± such as reference
systems ± and certain properties ± such as contact and support ± are likely to be
important. From this observation, it might come as no surprise that the same proper-
ties are preserved by the two systems: the task of negotiating space is, after all,
constrained by the physical space we live in, our perceptual capacities to detect spatial
information, and common evolutionary demands.
Acknowledgements
This research was partially supported by grant NIMH 55240 to B.L. We would
like to thank those who translated materials, assisted in coding data, and helped with
interpretation of results: Kippeum Kim, Meesook Kim, Masato Kobayashi, Jerome
(Barney) Pagani, Melissa Schweissguth, Tim Simpson, Kazue Takeda, Ryuhei
Tsuji, and Diamond Vu. This paper is based in part on work presented in the ®rst
author's Master's Thesis (Munnich, 1997). Much of the work discussed in this paper
was presented at the 23rd Annual Meeting of the Psychonomic Society, Philadel-
phia, PA, November 1997.
Appendix A. Expressions used in language tasks of Experiments 1 and 2
English Japanese Korean
AXIAL TERMS
above, over ue (-ni), jouhou (-ni) wi (-e)
below, under shita (-ni), kahou (-ni) ±
(to the) left hidari (-ni) oencho (-e)
(to the) right migi (-ni) oreuncho (-e)
next (to) tonari (-ni)
E. Munnich et al. / Cognition 81 (2001) 171±207204
(continued)
English Japanese Korean
above (and to the) left ue (-ni), hidari (-ni) wi (-e), oencho (-e)
below (and to the) right shita (-ni) migi (-ni) ± oreuncho (-e)
(to the) left (and) above hidari (-ni), ue (-ni) oencho (-e), wi (-e)
(to the) right (and) below migi (-ni) shita (-ni) oreuncho (-e) ±
(to the) top left ± ±
(to the) top right ± ±
(to the) bottom left ± ±
(to the) bottom right ± ±
(to the) left top hidari ue (-ni) oenchok wi (-e)
(to the) left bottom hidari shita (-ni) ±
(to the) right top migi ue (-ni) oreunchok
(to the) right bottom migi shita (-ni) ±
CONTACT TERMS
[being] on notte noye
sitting tsuite ±
touching sesshite ±
[sticking to] kuttsuite putte
TERMS DENOTING LACK OF CONTACT
hovering ± ±
¯oating uite tte
away from hanarete ±
Note: square brackets denote English translations of expressions used by Japa-
nese- or Korean-speaking participants, but which English speakers never used;
parentheses denote the functional morphemes that typically accompanied the lexical
morphemes with which we were concerned.
References
Berlin, B., & Kay, P. (1969). Basic color terms: their universality and evolution. Berkeley, CA: Univer-
sity of California Press.
Berman, R., & Slobin, D. (1994). Relating events in narrative: a crosslinguistic developmental study,
Hillsdale, NJ: Lawrence Erlbaum.
Bowerman, M. (1996). Learning how to structure space for language: a cross-linguistic perspective. In P.
Bloom, M. Peterson, L. Nadel & M. Garrett (Eds.), Language and space. Cambridge, MA: MIT Press.
Bowerman, M., & Choi, S. (1994, November). Linguistic and nonlinguistic determinants of spatial
semantic development. Paper presented at the Boston University Conference on Language Develop-
ment, Boston, MA.
Brown, P., & Levinson, S. (1993). Linguistic and nonlinguistic coding of spatial arrays: explorations in
Mayan cognition (Working Paper No. 24). Nijmegen: Cognitive Anthropology Research Group, Max
Planck Institute for Psycholinguistics.
Brown, R., & Lenneberg, E. (1954). A study in language and cognition. Journal of Abnormal and Social
Psychology,49, 454±462.
Carlson-Radvansky, L., & Irwin, D. (1993). Frames of reference in vision and language: where is above?
Cognition,46, 223±244.
E. Munnich et al. / Cognition 81 (2001) 171±207 205
Clark, H. (1973). Space, time, semantics, and the child. In T. E. Moore (Ed.), Cognitive development and
the acquisition of language. New York: Academic Press.
Cooper, W., & Ross, J. R. (1975). World order. In R. Grossman, L. San & T. Vance (Eds.), Papers from
the parasession on functionalism. Chicago, IL: Chicago Linguistic Society.
Crawford, L., Regier, T., & Huttenlocher, J. (2000). Linguistic and non-linguistic spatial categorization.
Cognition,73 (5), 209±235.
Davidoff, J., Davies, I., & Roberson, D. (1999). Colour categories in a stone-age tribe. Nature,402 (6762),
604±605.
DeLoache, J. (1987). Rapid change in the symbolic functioning of very young children. Science,238,
1556±1557.
Fodor, J. (1981). Representations. Cambridge, MA: MIT Press.
Freyd, J. (1983). Shareability: the social psychology of epistemology. Cognitive Science,7, 191±210.
Gallistel, C. R. (1990). The organization of learning. Cambridge, MA: MIT Press.
Gennari, S., Sloman, S., Malt, B., & Fitch, W. (2000, March). Language processing and perception of
motion events. Poster presented at the 13th Annual CUNY Conference on Human Sentence Proces-
sing, La Jolla, CA.
Gentner, D. (1977). Children's performance on a spatial analogies task. Child Development,48 (3), 1034±
1039.
Hayward, W., & Tarr, M. (1995). Spatial language and spatial representation. Cognition,55, 39±84.
Heider, E. (1972). Universals in color naming and memory. Journal of Experimental Psychology:
General,93 (1), 10±20.
Heider, E., & Olivier, D. (1972). The structure of the color space in naming and memory for two
languages. Cognitive Psychology,3, 337±354.
Huttenlocher, J., Hedges, L., & Duncan, S. (1991). Categories and particulars: prototype effects in
estimating spatial location. Psychological Review,98 (3), 352±376.
Huttenlocher, J., & Presson, C. (1979). The coding and transformation of spatial information. Cognitive
Psychology,11, 375±394.
Imai, M., & Gentner, D. (1997). A cross-linguistic study of early word meaning: universal ontology and
linguistic in¯uence. Cognition,62 (2), 169±200.
Jackendoff, R. (1983). Semantics and cognition. Cambridge, MA: MIT Press.
Johnson, J., & Newport, E. (1989). Critical period effects in second language learning: the in¯uence of
maturational state on the acquisition of English as a second language. Cognitive Psychology,21, 60±99.
Kosslyn, S. (1980). Image and mind. Cambridge, MA: Harvard University Press.
Landau, B., & Jackendoff, R. (1993). ªWhatº and ªwhereº in spatial language and spatial cognition.
Behavioral and Brain Sciences,16, 217±238.
Lantz, D., & Stef¯re, V. (1964). Language and cognition revisited. Journal of Abnormal and Social
Psychology,69, 472±481.
Lee, P. (1996). The Whorf theory complex: a critical reconstruction. Philadelphia, PA: John Benjamins.
Levinson, S. (1996). Frames of reference and Molyneux's Question: cross-linguistic evidence. In P.
Bloom, M. Peterson, L. Nadel, & M. Garrett (Eds.), Language and space. Cambridge, MA: MIT Press.
Li, P., & Gleitman, L. (2000). Turning the tables: language and spatial reasoning, IRCS Technical Report
00-03. University of Pennsylvania.
Logan, G. (1995). Linguistic and conceptual control of visual spatial attention. Cognitive Psychology,28,
103±174.
Lucy, J. (1992). Grammatical categories and cognition. Cambridge: Cambridge University Press.
Lucy, J. (1996). The scope of linguistic relativity: an analysis and review of empirical research. In J.
Gumperz & S. Levinson (Eds.), Rethinking linguistic relativity. Cambridge: Cambridge University
Press.
Malt, B., Sloman, S., Gennari, S., Shi, M., & Wang, Y. (1999). Knowing versus naming: similarity and the
linguistic categorization of artifacts. Journal of Memory and Language,40 (2), 230±262.
Milner, A., & Goodale, M. (1996). The visual brain in action. Oxford: Oxford University Press.
Munnich, E. (1997). Axial terms: a cross-linguistic comparison. Unpublished master's thesis, University
of California, Irvine.
E. Munnich et al. / Cognition 81 (2001) 171±207206
Newcombe, N., & Huttenlocher, J. (1992). Children's early ability to solve perspective-taking problems.
Developmental Psychology,28 (4), 635±643.
Ozcaliskan, S., & Slobin, D. (1999). Learning how to search for the frog: expression of manner of motion
in English, Spanish, and Turkish. In A. Greenhill, H. Little®eld, & C. Tano (Eds.), Proceedings of the
23rd Annual Boston University Conference on Language Development (Vol. 23(2), pp. 541±552).
Somerville, MA: Cascadilla Press.
Ozyurek, A., & Kita, S. (1999). Expressing manner and path in English and Turkish: differences in speech,
gesture, and conceptualization. In N. Hahn & S. C. Stoness (Eds.), Proceedings of the 21st Annual
Conference of the Cognitive Science Society (pp. 507±512). Hillsdale, NJ: Erlbaum.
Pederson, E., Danziger, E., Wilkins, D., Levinson, S., Kita, S., & Senft, G. (1998). Semantic typology and
spatial conceptualization. Language,74 (3), 557±589.
Pinker, S., & Bloom, P. (1990). Natural language and natural selection. Behavioral and Brain Sciences,
13, 707±727.
Rosch, E. (1973). Natural categories. Cognitive Psychology,4, 328±350.
Shepard, R. (1984). Ecological constraints on internal representation: resonant kinematics of perceiving,
imagining, thinking, and dreaming. Psychological Review,91 (4), 417±447.
Silberstein, C., & Spelke, E. (1995). Explicit vs. implicit processes in spatial cognition. Poster presented at
the meeting of the Society for Research in Child Development, Indianapolis, IN.
Slobin, D. (1996). From `thought and language' to `thinking for speaking'. In J. Gumperz & S. Levinson
(Eds.), Rethinking linguistic relativity. Cambridge: Cambridge University Press.
Stef¯re, V., Castillo Vales, V., & Morley, L. (1966). Language and cognition in Yucatan: a cross cultural
replication. Journal of Personality and Social Psychology,4, 112±115.
Talmy, L. (1983). How languages structure space. In H. Pick & L. Acredolo (Eds.), Spatial orientation:
theory, research, and application. New York: Plenum Press.
Talmy, L. (1985). Lexicalization patterns: semantic structure in lexical forms. In T. Shopen (Ed.),
Language typology and semantic description III.: Grammatical categories and the lexicon (pp. 57±
149). Cambridge: Cambridge University Press.
Talmy, L. (1991). Path to realization: a typology of event con¯ation. In L. A. Sutton, C. Johnson, & R.
Shields (Eds.), Proceedings of the 17th Annual Meeting of the Berkeley Linguistics Society (pp. 480±
519). Berkeley, CA: Berkeley Linguistics Society.
Whorf, B. (1956). Language, thought, and reality. New York: Technology Press of MIT and Wiley.
Yoshida, H., & Smith, L. (1999, November). Japanese syntax facilitates children's acquisition of names
for animals. Paper presented at the 24th Annual Boston University Conference on Language Devel-
opment, Boston, MA.
E. Munnich et al. / Cognition 81 (2001) 171±207 207
