Paul Kay Terry Regier - Institute for Mathematical Behavioral

advertisement
Relativity versus Universals: an obsolete dichotomy
Paul Kay
Terry Regier
International Computer Science Institute,
U. C., Berkeley
paulkay@berkeley.edu
Irvine
March 2008
U. of Chicago
regier@uchicago.edu
Major point of this talk: The opposition of
so-called relativism to so-called
universalism impedes our understanding
of the complex ways in which language
both influences perception and is
influenced by perception…
or
Half a Whorf is better than no Whorf at all.
Plan of the talk
• Influence of language on color
perception/discrimination
(Kay) Moral: Hedged “relativism”
• Color naming similarities and
differences across languages
(Regier) Moral: Hedged “universalism”
Previous work, going back as far as
1984, has demonstrated:
•(1) “Categorical Perception” (CP) for
color,
•(2) that CP occurs at boundaries of
linguistically encoded basic color terms,
and
•(3) that when basic color term
boundaries differ between languages,
color CP occurs at the linguistic
boundaries.
•(e.g., Kay & Kempton 1984, Levinson, 1996; Lucy, 1996; Roberson et al., 2000; Winawer
et al., 2003. For full references: paulkay@berkeley.edu.).
This part of the talk is about recent work,
showing that:
• Color CP is lateralized to the right Visual
Field (RVF) and therefore to the Left
Cerebral Hemisphere (LH).
• BACKGROUND:
(A) Visual fields project to
brain contralaterally;
(B) LH is dominant for
language.
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
Front
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
Lexical categories influence perception in the RVF. (a) Print-rendered versions of the four colors used. (b) Sample display
for the visual search task. Participants were required to press one of two response keys, indicating the side containing the
target color. (c) In the no-interference condition, RTs were faster for the between-category pair and slower for the withincategory pairs when targets appeared in the RVF compared with when they appeared in the LVF. (d) Effects were
reversed with verbal interference. *, P < 0.05, two-tailed t test, df = 10; ns, nonsignificant. Values are mean  SEM
BERKELEY EXP. 1
Fig. 1. Lexical categories influence perception in the RVF. (a) Print-rendered versions of the four colors
used. (b) Sample display for the visual search task. Participants were required to press one of two response
keys, indicating the side containing the target color. (c) In the no-interference condition, RTs were faster for
the between-category pair and slower for the within-category pairs when targets appeared in the RVF
compared with when they appeared in the LVF. (d) Effects were reversed with verbal interference. *, P <
0.05, two-tailed t test, df = 10; ns, nonsignificant. Values are mean ± SEM.
( Gilbert, A., Regier, T., Kay, P., & Ivry, R. (2006) PNAS)
BERKELEY EXP. 2
Fig. 2. Modulation of color-category effects in the RVF is specific to linguistic demands of the
interference task. (a) Trial events. Within a block of trials, the visual search task was interleaved with
blank displays, displays containing a color word, or displays containing a spatial grid. (b and c) Nointerference and verbal-interference results replicate those obtained in the first experiment. (d) For the
nonverbal-interference condition, performance followed a pattern similar to that observed in the nointerference condition. *, P < 0.05, two-tailed t test, df = 10; ns, nonsignificant.
( Gilbert, A., Regier, T., Kay, P., & Ivry, R. (2006) PNAS)
Elimination of lateralized Whorf effect
with verbal interference argues that
linguistic categories are activated online in this task (as against the learning
of linguistic categories having warped
the perceptual space itself). Whether
the effect is strictly speaking perceptual
or involves immediate post-perceptual
processing can’t be answered from
these data.
OK, but:
That’s just one study, in one lab,
on one color boundary.
Replicated, at University of Surrey,
at blue/green, blue/purple, purple/pink
boundaries. (Also: weaker Whorf in LFV,
probably due to trans-callosal transfer.)
(Drivonikou, Kay, Regier, Ivry, Gilbert, Franklin & Davies (2007). PNAS.)
SURREY EXP. 1
Fig. 1. The category effect is larger in the RVF than in the LVF in a reanalysis of the data from a color
identification task used by Daoutis et al. (16). (a) Stimuli in CIE coordinates. The within-category set contains
three hues of green: G1, G2, G3; the across-category set contains a blue (B), a purple (Pu), and a pink (Pi).
Perceptual distance is the same for all adjacent pairs across both sets (G1-G2, G2-G3, B-Pu, and Pu-Pi).
When a peripheral stimulus (e.g., G1) is the target, it is linearly separable from the distractors (G2 G3). (b)
Illustration of a target-present trial with 15 distractors. The target is indicated here by the arrow that, however,
was not present in the display itself. (c and d) Target detection times for within- and across-category targets
by LVF and RVF: collapsed across linear separability (c) and linearly separable targets alone (d). Error bars
show 95% confidence limits. (G. V. Drivonikou, P. Kay, T. Regier, R. B. Ivry, A. L. Gilbert, A. Franklin, and I. R. L.
Davies. PNAS 2007;104;1097-1102)
SURREY EXP. 2
Fig. 2. A larger category effect is observed in the RVF on a color detection task. (a) Munsell codes of
the stimuli; stimuli varied in hue at constant value and chroma. Hue separations were either five steps
(far set) or 2.5 steps (near set). The target was either in the same color category as the background
(e.g., 10BG on 5B, both blue) or in the adjacent category (e.g., 10BG on 5BG, blue on green). (b)
Illustration of a test frame: white circles show possible target locations around the fixation cross, and
the black circle representing the target. (c and d) Blue-green set: The difference in RT between withinand across-category is larger in the RVF (c). Target-background perceptual separation only affects the
RVF (d). (e) Blue-purple set: Again, the difference in RT between within- and across-category is larger
in the RVF. Error bars are 95% confidence limits. (G. V. Drivonikou, P. Kay, T. Regier, R. B. Ivry, A. L.
Gilbert, A. Franklin, and I. R. L. Davies. PNAS 2007;104;1097-1102)
What about split-brain patients? They should also
show CP stronger in RVF.
Fig. 3. Lexical categories influence perception in the RVF of a callostomy patient. *,
P < 0.05, two-tailed t-test, df = 1; ns, nonsignificant. (Gilbert, A., Regier, T. Kay, P., &
Ivry, R. PNAS. 2003; 100, 489-494)
Fig. 6. Visual search task results from callosotomy patient testing. Error bars show 95%
confidence limits. (Gilbert, A., Regier, T., Kay, P., & Ivry, R. Brain and Language, in press)
Is (near) restriction of CP to RVF limited to color?
Answer: No.
Dog and cat stimuli for extension of COLOR CP
outside of color domain.
(Gilbert, A., Regier, T., Kay, P., & Ivry, R. Brain and Language, in press)
Fig. 2. Sample display for the visual search task [as in BERKELEY EXP. 1] with a betweencategories stimulus pair. Participants were required to press one of two response keys,
indicating the side containing the target. (Gilbert, A., Regier, T., Kay, P., & Ivry, R. Brain and
Language, in press)
Standard search task with dog and cat stimuli
RVF CP observed, except with verbal interference
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
(Gilbert, A., Regier, T.,Kay, P., & Ivry, R. Brain and Language (2007), doi:10.1016/j.bandl.2007.06.001)
What about other languages? Two early reports.
Debi Roberson (U. of Essex) has reported that CP for a color
category boundary occurring in Korean but not English is
present in the RVF and not the LVF (for fastest-responding
participants only).
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
(Source: Roberson, D. et al., Categorical perception of colour in the left visual field is
verbally mediated. Cognition (2007), doi:10.1016/j.cognition.2007.09.001.)
Also, Drivonikou, Davies, Franklin & Taylor, (2007, Lateralisation
of colour categorical perception: A cross-cultural study"
Perception 36 ECVP Abstract Supplement)
report, “Greek has two basic blue terms (ble and galazio), and
many African languages have one term that includes blue and
green. CP was shown by faster detection of targets on differentthan same-category backgrounds. However, lateralisation of CP
was only found when category boundaries were marked by the
language.”
These two studies provide evidence that what’s at work is
linguistically encoded categories. Color CP was, as we’ve seen,
previously established to vary with linguistic boundaries;
Roberson’s and Drivonikou et al.’s recent studies indicate that
this holds for lateralized color CP as well.
Moreover…
VF to Brain Hemisphere inference
has been confirmed in an EventRelated Potential (ERP) study using
the same stimuli as in BERKELEY
EXPS. 1 & 2.
+
Standard
200 ms
+
Fixation
[800-1200 ms]
+
Cross-category
LVF deviant
200 ms
+
Fixation
[800-1200 ms]
+
Standard
200 ms
+
Fixation
[800-1200 ms]
o
Target
200 ms
+
Fixation
[800-1200 ms]
+
Standard
200 ms
+
Fixation
[800-1200 ms]
+
Within-category
RVF deviant
200 ms
EEG experiment protocol. Aubrey Gilbert’s Dissertation (Berkeley, 2007).
ŅIn a paradigm similar to ones used in previous visual
mismatch negativity (V-MMN) studies, event-related
potentials were recorded in response to standard and
deviant color stimuli presented lateral to a centralized
task. Deviant stimuli were either from the same
(within-category) or from a different (cross-category)
lexical category than the standard stimuli. Two
deviance-related negativities (DRNs) were observed:
an early (150-300 ms), left hemisphere-lateralized
component at occipital and extrastriate sites, and a
late (400-700 ms) bilateral, frontal component that
was larger in the left hemisphere. These DRNs were
evoked only when the cross-category deviant was
presented in the right visual field. Evoked potentials
for all other deviant conditions did not differ
significantly from those for the standard stimuli.Ó-Aubrey Gilbert (Berkeley Dissertation. 2007)
Figure 3. Grand-averaged ERPs to standard (in black) and each of the deviant stimuli.
The only deviant to evoke ERPs with significant differences from those evoked by
standard stimuli was the cross-category deviant (in red) and these diff erences only
occurred when this deviant was presented in the RVF. The signifi cant differences of note
are an earli er (~150-300 ms) increased negativity at occipital and extrastriate sites that is
laterali zed mostly to t he LH, and a later (~400-700 ms) increased negativity at frontal
sites that is observed bil aterally .
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
(Tan, L-H, Chang, A H D, Kay P, Khong, P-L, Yip, L K C, & Luke, K-K. PNAS, 105, 4004-4009 (2008)).
EXPERIMENT
Calloso tomy 1
Calloso tomy 2
BERKELE Y 1
BERKELE Y 2
SURRE Y 2
Roberson (fas test Ss)
SURRE Y 1
Roberson (slowes t Ss)
RT (mean, approx.)
(400)
(700)
425
440
540
740
1375
1420
LVF CP?
NO (transfer impossible)
NO (transfer impossible)
NO
NO
NO/YES*
NO
YES (transfer probable)
YES (transfer probable)
Table 1: Approximate mean res pons e times for th e eigh t exp eriments that sh ow lateralized Who rf effects, with
ind ication of s ign ificant LV F C P. Only the two studies wit h RTs well ov er 10 00 m s s how LVF C P, s ugg es ting
tran s-callos al tran sfer an d scan ning as pos sible caus es . *In this experimen t one co lor bou ndary sh owed LVF CP an d
one did no t
PATIENTS
CONTROL S
Figure 13. Results of 15 aphasic patients and 12 controls tested on the no-interference visual search task of Gilbert,
et al. (2006). Source: Paluy, Y ., Gilbert , A.L., Baldo , J.V ., & Ivry, R.B. (2007, August) . Is Whorf Right? (or Left?).
Poster presented at the 29th Annual Cognitive Science Society. Nas hville, TN
SUMMARY SO FAR:
1. Lateralized RVF color CP has been found in a variety of tasks
on normal adult s.
2. Lateralized RVF color CP has been found in callosotomy
patients
3. Lateralized RVF color CP is absent in aphasics with left
hemisphere lesions; thes e patients show LVF color CP.
4. Lateralized RVF color CP has been found at lexical
boundaries not present in English for speakers of languages
containing those boundarie s (Korean and Greek) and not for
English speakers.
5. Lateralized RVF CP has been found outside the color domain
(dog and cat silhouettes).
6. Lateralized RVF CP in normal adults is consistently
suppressed by verbal, but not by non-verbal, interference.
7. Colo r CP has been found in LVF in some experiments. It is
always weaker than RVF CP and the longer response times in
these experiments sugges t trans-callosal transfer and/or
scanning. Either or both of the latter factors could render
apparent LVF CP illusory.
8. The inference from RVF CP to left hemisphere activity has
been confirmed in an EEG study.
9. The inference from RVF CP to left hemisphere activity has
been confirmed in an fMRI study
So far everythin g is consistent with left hemisphere language
causing color CP in norma l adults.
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
Figure 14: Prelinguistic infants show color CP lateralized to LVF. Source: (A. Franklin,
G.V. Drivonikou, L. Bevis, I.R.L. Davies, P. Kay, & T. Regier. PNAS 105, 3221–3225.(2008).)
Assuming:
1. LVF CP in normals is illusory, reflecting trans-callosal transfer
and/or scanning,
2. LVF CP in aphasics with LH lesions reflects transfer of function,
Then the apparent conflict between RVF CP in normal adults and
LVF CP in infants can be explained as language acquisition in the
LH taking over the categorization function from the RH as the child matures.
Take home lesson:
Normal adults, under normal viewing conditions, get
two pictures of the word simultaneously: one filtered
through linguistic categories and one not so filtered.
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
“Me worry?”
Download