the semantics of color terms. a quantitative cross ...gjaeger/slides/slidesleipzig.pdf · colors...
TRANSCRIPT
![Page 1: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/1.jpg)
The semantics of color terms. A quantitativecross-linguistic investigation
Gerhard [email protected]
May 20, 2010
University of Leipzig
1/124
![Page 2: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/2.jpg)
The psychological color space
physical color space has infinite dimensionality — everywavelength within the visible spectrum is one dimensionpsychological color space is only 3-dimensionalthis fact is employed in technical devices like computer screens(additive color space) or color printers (subtractive colorspace)
additive color space subtractive color space
2/124
![Page 3: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/3.jpg)
The psychological color space
psychologically correct color space should not only correctlyrepresent the topology of, but also the distances betweencolors
distance is inverse function of perceived similarity
L*a*b* color space has this property
three axes:
black — whitered — greenblue — yellow
irregularly shaped 3d color solid
3/124
![Page 4: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/4.jpg)
The color solid
4/124
![Page 5: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/5.jpg)
The Munsell chart
for psychological investigations, the Munsell chart is beingused2d-rendering of the surface of the color solid
8 levels of lightness40 hues
plus: black–white axis with 8 shaded of grey in betweenneighboring chips differ in the minimally perceivable way
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
5/124
![Page 6: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/6.jpg)
Berlin and Kay 1969
pilot study how different languages carve up the color spaceinto categories
informants: speakers of 20 typologically distant languages(who happened to be around the Bay area at the time)
questions (using the Munsell chart):
What are the basic color terms of your native language?What is the extension of these terms?What are the prototypical instances of these terms?
results are not random
indicate that there are universal tendencies in color namingsystems
6/124
![Page 7: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/7.jpg)
Berlin and Kay 1969
extensions
Arabic
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
7/124
![Page 8: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/8.jpg)
Berlin and Kay 1969
extensions
Bahasa Indonesia
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
8/124
![Page 9: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/9.jpg)
Berlin and Kay 1969
extensions
Bulgarian
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
9/124
![Page 10: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/10.jpg)
Berlin and Kay 1969
extensions
Cantonese
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
10/124
![Page 11: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/11.jpg)
Berlin and Kay 1969
extensions
Catalan
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
11/124
![Page 12: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/12.jpg)
Berlin and Kay 1969
extensions
English
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
12/124
![Page 13: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/13.jpg)
Berlin and Kay 1969
extensions
Hebrew
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
13/124
![Page 14: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/14.jpg)
Berlin and Kay 1969
extensions
Hungarian
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
14/124
![Page 15: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/15.jpg)
Berlin and Kay 1969
extensions
Ibibo
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
15/124
![Page 16: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/16.jpg)
Berlin and Kay 1969
extensions
Japanese
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
16/124
![Page 17: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/17.jpg)
Berlin and Kay 1969
extensions
Korean
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
17/124
![Page 18: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/18.jpg)
Berlin and Kay 1969
extensions
Mandarin
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
18/124
![Page 19: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/19.jpg)
Berlin and Kay 1969
extensions
Mexican Spanish
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
19/124
![Page 20: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/20.jpg)
Berlin and Kay 1969
extensions
Pomo
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
20/124
![Page 21: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/21.jpg)
Berlin and Kay 1969
extensions
Swahili
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
21/124
![Page 22: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/22.jpg)
Berlin and Kay 1969
extensions
Tagalog
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
22/124
![Page 23: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/23.jpg)
Berlin and Kay 1969
extensions
Thai
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
23/124
![Page 24: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/24.jpg)
Berlin and Kay 1969
extensions
Tzeltal
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
24/124
![Page 25: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/25.jpg)
Berlin and Kay 1969
extensions
Urdu
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
25/124
![Page 26: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/26.jpg)
Berlin and Kay 1969
extensions
Vietnamese
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
26/124
![Page 27: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/27.jpg)
Berlin and Kay 1969
identification of absolute and implicational universals, like
all languages have words for black and whiteif a language has a word for yellow, it has a word for redif a language has a word for pink, it has a word for blue...
27/124
![Page 28: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/28.jpg)
The World Color Survey
B&K was criticized for methodological reasons
in response, in 1976 Kay and co-workers launched the worldcolor survey
investigation of 110 non-written languages from around theworld
around 25 informants per languagetwo tasks:
the 330 Munsell chips were presented to each test person oneafter the other in random order; they had to assign each chipto some basic color term from their native languagefor each native basic color term, each informant identified theprototypical instance(s)
data are publicly available underhttp://www.icsi.berkeley.edu/wcs/
28/124
![Page 29: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/29.jpg)
Data digging in the WCS
distribution of focal colors across all informants:
Distribution of focal colors
Munsell chips
# na
med
as
foca
l col
or
2050
200
1000
29/124
![Page 30: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/30.jpg)
Data digging in the WCS
distribution of focal colors across all informants:
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
30/124
![Page 31: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/31.jpg)
Data digging in the WCS
partition of a randomly chosen informant from a randomlychosen language
31/124
![Page 32: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/32.jpg)
Data digging in the WCS
partition of a randomly chosen informant from a randomlychosen language
32/124
![Page 33: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/33.jpg)
Data digging in the WCS
partition of a randomly chosen informant from a randomlychosen language
33/124
![Page 34: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/34.jpg)
Data digging in the WCS
partition of a randomly chosen informant from a randomlychosen language
34/124
![Page 35: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/35.jpg)
Data digging in the WCS
partition of a randomly chosen informant from a randomlychosen language
35/124
![Page 36: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/36.jpg)
Data digging in the WCS
partition of a randomly chosen informant from a randomlychosen language
36/124
![Page 37: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/37.jpg)
Data digging in the WCS
partition of a randomly chosen informant from a randomlychosen language
37/124
![Page 38: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/38.jpg)
Data digging in the WCS
partition of a randomly chosen informant from a randomlychosen language
38/124
![Page 39: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/39.jpg)
Data digging in the WCS
partition of a randomly chosen informant from a randomlychosen language
39/124
![Page 40: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/40.jpg)
Data digging in the WCS
partition of a randomly chosen informant from a randomlychosen language
40/124
![Page 41: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/41.jpg)
What is the extension of categories?
data from individual informants are extremely noisy
averaging over all informants from a language helps, but thereis still noise, plus dialectal variation
desirable: distinction between “genuine” variation and noise
41/124
![Page 42: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/42.jpg)
Statistical feature extraction
first step: representation ofraw data in contingencymatrix
rows: color terms fromvarious languagescolumns: Munsell chipscells: number of testpersons who used therow-term for thecolumn-chip
A0 B0 B1 B2 · · · I38 I39 I40 J0
red 0 0 0 0 · · · 0 0 2 0green 0 0 0 0 · · · 0 0 0 0blue 0 0 0 0 · · · 0 0 0 0black 0 0 0 0 · · · 18 23 21 25white 25 25 22 23 · · · 0 0 0 0
......
......
......
......
......
rot 0 0 0 0 · · · 1 0 0 0grun 0 0 0 0 · · · 0 0 0 0gelb 0 0 0 1 · · · 0 0 0 0
......
......
......
......
......
rouge 0 0 0 0 · · · 0 0 0 0vert 0 0 0 0 · · · 0 0 0 0
......
......
......
......
......
further processing:
divide each row by the number n of test persons using thecorresponding termduplicate each row n times
42/124
![Page 43: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/43.jpg)
Principal Component Analysis
technique to reduce dimensionality of data
input: set of vectors in an n-dimensionalspace
first step:
rotate the coordinatesystem, such that
the new n coordinates areorthogonal to each otherthe variations of the dataalong the new coordinatesare stochasticallyindependent
second step:
choose a suitable m < n
project the data on those mnew coordinates where thedata have the highestvariance
43/124
![Page 44: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/44.jpg)
Principal Component Analysis
alternative formulation:
choose an m-dimensional linear sub-manifold of yourn-dimensional spaceproject your data onto this manifoldwhen doing so, pick your sub-manifold such that the averagesquared distance of the data points from the sub-manifold isminimized
intuition behind this formulation:
data are “actually” generated in an m-dimensional spaceobservations are disturbed by n-dimensional noisePCA is a way to reconstruct the underlying data distribution
applications: picture recognition, latent semantic analysis,statistical data analysis in general, data visualization, ...
44/124
![Page 45: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/45.jpg)
Statistical feature extraction: PCA
first 15 principalcomponents jointlyexplain 91.6% of thetotal variance
choice of m = 15 isdetermined by using“Kaiser’s stoppingrule”
principal components
prop
ortio
n of
var
ianc
e ex
plai
ned
0.00
0.05
0.10
0.15
0.20
0.25
0.30
45/124
![Page 46: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/46.jpg)
Statistical feature extraction: PCA
after some post-processing (“varimax” algorithm):
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
46/124
![Page 47: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/47.jpg)
Projecting observed data on lower-dimensional-manifold
noise removal: project observed data onto thelower-dimensional submanifold that was obtained via PCA
in our case: noisy binary categories are mapped to smoothedfuzzy categories (= probability distributions over Munsellchips)
some examples:
47/124
![Page 48: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/48.jpg)
Projecting observed data on lower-dimensional-manifold
48/124
![Page 49: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/49.jpg)
Projecting observed data on lower-dimensional-manifold
49/124
![Page 50: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/50.jpg)
Projecting observed data on lower-dimensional-manifold
50/124
![Page 51: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/51.jpg)
Projecting observed data on lower-dimensional-manifold
51/124
![Page 52: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/52.jpg)
Projecting observed data on lower-dimensional-manifold
52/124
![Page 53: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/53.jpg)
Projecting observed data on lower-dimensional-manifold
53/124
![Page 54: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/54.jpg)
Projecting observed data on lower-dimensional-manifold
54/124
![Page 55: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/55.jpg)
Projecting observed data on lower-dimensional-manifold
55/124
![Page 56: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/56.jpg)
Projecting observed data on lower-dimensional-manifold
56/124
![Page 57: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/57.jpg)
Projecting observed data on lower-dimensional-manifold
57/124
![Page 58: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/58.jpg)
Projecting observed data on lower-dimensional-manifold
58/124
![Page 59: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/59.jpg)
Projecting observed data on lower-dimensional-manifold
59/124
![Page 60: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/60.jpg)
Projecting observed data on lower-dimensional-manifold
60/124
![Page 61: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/61.jpg)
Projecting observed data on lower-dimensional-manifold
61/124
![Page 62: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/62.jpg)
Projecting observed data on lower-dimensional-manifold
62/124
![Page 63: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/63.jpg)
Projecting observed data on lower-dimensional-manifold
63/124
![Page 64: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/64.jpg)
Projecting observed data on lower-dimensional-manifold
64/124
![Page 65: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/65.jpg)
Projecting observed data on lower-dimensional-manifold
65/124
![Page 66: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/66.jpg)
Projecting observed data on lower-dimensional-manifold
66/124
![Page 67: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/67.jpg)
Projecting observed data on lower-dimensional-manifold
67/124
![Page 68: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/68.jpg)
Smoothing the partitions
from smoothed extensions we can recover smoothed partitions
each pixel is assigned to category in which it has the highestdegree of membership
68/124
![Page 69: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/69.jpg)
Smoothed partitions of the color space
69/124
![Page 70: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/70.jpg)
Smoothed partitions of the color space
70/124
![Page 71: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/71.jpg)
Smoothed partitions of the color space
71/124
![Page 72: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/72.jpg)
Smoothed partitions of the color space
72/124
![Page 73: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/73.jpg)
Smoothed partitions of the color space
73/124
![Page 74: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/74.jpg)
Smoothed partitions of the color space
74/124
![Page 75: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/75.jpg)
Smoothed partitions of the color space
75/124
![Page 76: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/76.jpg)
Smoothed partitions of the color space
76/124
![Page 77: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/77.jpg)
Smoothed partitions of the color space
77/124
![Page 78: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/78.jpg)
Smoothed partitions of the color space
78/124
![Page 79: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/79.jpg)
Convexity
note: so far, we only used information from the WCS
the location of the 330 Munsell chips in L*a*b* space playedno role so far
still, apparently partition cells always form continuous clustersin L*a*b* space
Hypothesis (Gardenfors): extension of color terms always formconvex regions of L*a*b* space
79/124
![Page 80: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/80.jpg)
Support Vector Machines
supervised learning techniquesmart algorithm to classify data in a high-dimensional spaceby a (for instance) linear boundaryminimizes number of mis-classifications if the training dataare not linearly separable
gree
nre
d
−3 −2 −1 0 1 2 3
−3
−2
−1
0
1
2
3
o
o oo
o
o
oooo o
o
o
o
o
o
o
ooo
o
oo
o
o
o
oo o
o
o
o
o
oo
o
o
o
o
o
o
o
o
o
oo
o
oo
o o
oo
o
o
o
o o
o
o
oo o
o
o
o
o
o
o
o
o
ooo
o
o
oo
o
o
o
o
o
o
o
o
o
oo
o
o
o
o
o
o o
o o
oo
o
o
o
o
o
o
o
ooo
oo
o
o
oo
o
o
o
o
o
o o
o
o
o o
o
o
ooo
o
o
o
o
o
o
o
o
o
o o
oo
o
o
o
o
o
o
oo
o
o
oo
o
oo
o
oo
o
o
o
oo
o
o
o oo
o
o
o
o
o
oo
o
o
o
o
o
o
o
oo
o
o
o
oo
o
ooo
o
o
SVM classification plot
y
x
80/124
![Page 81: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/81.jpg)
Convex partitions
a binary linear classifier divides an n-dimensional space intotwo convex half-spaces
intersection of two convex set is itself convex
hence: intersection of k binary classifications leads to convexsets
procedure: if a language partitions the Munsell space into mcategories, train m(m−1)
2 many binary SVMs, one for each pairof categories in L*a*b* space
leads to m convex sets (which need not split the L*a*b*space exhaustively)
81/124
![Page 82: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/82.jpg)
Convex approximation
82/124
![Page 83: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/83.jpg)
Convex approximation
83/124
![Page 84: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/84.jpg)
Convex approximation
84/124
![Page 85: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/85.jpg)
Convex approximation
85/124
![Page 86: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/86.jpg)
Convex approximation
86/124
![Page 87: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/87.jpg)
Convex approximation
87/124
![Page 88: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/88.jpg)
Convex approximation
88/124
![Page 89: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/89.jpg)
Convex approximation
89/124
![Page 90: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/90.jpg)
Convex approximation
90/124
![Page 91: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/91.jpg)
Convex approximation
91/124
![Page 92: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/92.jpg)
Convex approximation
on average, 93.7% of all Munsell chips are correctly classifiedby convex approximation
●
●
●
●
●
●
●
●
●
●
0.80
0.85
0.90
0.95
prop
ortio
n of
cor
rect
ly c
lass
ified
Mun
sell
chip
s
92/124
![Page 93: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/93.jpg)
Convex approximation
compare to the outcome of the same procedure without PCA,and with PCA but using a random permutation of the Munsellchips
●●●●
●
●●●●
●
●
●
●●●
●●
●
●
●
●
●●●
●
●●
●●●●
●
●●●●●
●
●
●
●
●●●●●●●●●●●
●
●●●
●
●
●●●●●●
●●
●●
●
●●●●●
●
●●●●●●
●
●●
●
●
●
●
●
●●
●
●
●
●
●●●
●
●
●
●
●
●
●
●●
●●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●●
1 2 3
2040
6080
100
degr
ee o
f con
vexi
ty (
%)
93/124
![Page 94: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/94.jpg)
Convex approximation
choice of m = 10 is somewhat arbitraryoutcome does not depend very much on this choice though
●
●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
0 10 20 30 40 50
5060
7080
9010
0
no. of principal components used
mea
n de
gree
of c
onve
xity
(%
)
94/124
![Page 95: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/95.jpg)
Implicative universals
first six features correspond nicely to the six primary colorswhite, black, red, green, blue, yellow
according to Kay et al. (1997) (and many other authors)simple system of implicative universals regarding possiblepartitions of the primary colors
95/124
![Page 96: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/96.jpg)
Implicative universals
I II III IV V
whitered/yellowgreen/blueblack
whiteredyellowgreen/blueblack
[white/red/yellowblack/green/blue
] whitered/yellowblack/green/blue
whitered/yellowgreenblack/blue
whiteredyellowgreenblueblack
whiteredyellowblack/green/blue
whiteredyellowgreenblack/blue
whiteredyellow/green/blueblack
whiteredyellow/greenblueblack
whiteredyellow/greenblack/blue
source: Kay et al. (1997)
96/124
![Page 97: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/97.jpg)
Partition of the primary colors
each speaker/term pair can be projected to a 15-dimensionalvector
primary colors correspond to first 6 entries
each primary color is assigned to the term for which it has thehighest value
defines for each speaker a partition over the primary colors
97/124
![Page 98: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/98.jpg)
Partition of the primary colors
for instance: sample speaker(from Piraha):
extracted partition:white/yellowredgreen/blueblack
supposedly impossible, butoccurs 61 times in thedatabase
J
I
H
G
F
E
D
C
B
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
98/124
![Page 99: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/99.jpg)
Partition of primary colors
most frequent partition types:
1 {white}, {red}, {yellow}, {green, blue}, {black} (41.9%)2 {white}, {red}, {yellow}, {green}, {blue}, {black} (25.2%)3 {white}, {red, yellow}, {green, blue, black} (6.3%)4 {white}, {red}, {yellow}, {green}, {black, blue} (4.2%)5 {white, yellow}, {red}, {green, blue}, {black} (3.4%)6 {white}, {red}, {yellow}, {green, blue, black} (3.2%)7 {white}, {red, yellow}, {green, blue}, {black} (2.6%)8 {white, yellow}, {red}, {green, blue, black} (2.0%)9 {white}, {red}, {yellow}, {green, blue, black} (1.6%)10 {white}, {red}, {green, yellow}, {blue, black} (1.2%)
99/124
![Page 100: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/100.jpg)
Partition of primay colors
87.1% of all speaker partitions obey Kay et al.’s universals
the ten partitions that confirm to the universals occupy ranks1, 2, 3, 4, 6, 7, 9, 10, 16, 18
decision what counts as an exception seems somewhatarbitrary on the basis of these counts
100/124
![Page 101: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/101.jpg)
Partition of primary colors
more fundamental problem:
partition frequencies aredistributed according to powerlaw
frequency ∼ rank−1.99
no natural cutoff point todistinguish regular from exceptionalpartitions
●
●
●
●
● ●
●
●
●
●
● ● ●
●●●
●
●●
●
●●●●●
●●●●●●●
●●●●●●●●●●●●●●●●●●●●●●●
1 2 5 10 20 50
12
510
2050
100
200
500
rank
freq
uenc
y
101/124
![Page 102: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/102.jpg)
Partition of seven most important colors
frequency ∼ rank−1.64
●
●
●
●
●
●● ●
● ● ● ●
●●●
●●
●●
●●●●●●●●
●●●●
●●●●
●●●●●●
●●●●●●●●●●●●●●●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
1 2 5 10 20 50 100
12
510
2050
100
200
500
rank
freq
uenc
y
102/124
![Page 103: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/103.jpg)
Partition of eight most important colors
frequency ∼ rank−1.46
●
●● ●
●●
● ●●
●●●●
●●●●
●●
●●
●●●
●●●●●●
●●●
●●●●●
●●●●
●●
●●●●●●●
●●●●●●●●●●●●●●●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
1 2 5 10 20 50 100 200
12
510
2050
100
200
rank
freq
uenc
y
103/124
![Page 104: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/104.jpg)
Power laws
104/124
![Page 105: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/105.jpg)
Power laws
105/124
![Page 106: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/106.jpg)
Power laws
from Newman 2006
106/124
![Page 107: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/107.jpg)
Other linguistic power law distributions
number of vowels
vowel systems and their frequency of occurrence
3
14
4
14 5 4 2
5
97 3
6
26 12 12
7
23 6 5 4 3
8
6 3 3 2
9
7 7 3
(from Schwartz et al. 1997,
based on the UCLA Phonetic Segment Inventory Database)
107/124
![Page 108: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/108.jpg)
Other linguistic power law distributions
frequency ∼ rank−1.06
●
●
●
● ●
● ●
● ●
● ●
● ●
● ●
● ● ●●●
●●
1 2 5 10 202
510
2050
100
rank
freq
uenc
y
108/124
![Page 109: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/109.jpg)
Other linguistic power law distributions
size of language families
source: Ethnologue
frequency ∼ rank−1.32
●
●
● ● ●●
●● ●
●
●
●●●
●●●●●●
●●●●●●
●●●●●●●●●●●●●●●●●
●●●
●●●●●●●●●●●●●●●●●●●
●●●●●●●●
●●●●●●●●●●●●●
●●●●●●●●
●●●●●●●●●●●●●●●●●●●●●
●●●●●●●●
1 2 5 10 20 50 1001
510
5010
050
0
rank
freq
uenc
y
109/124
![Page 110: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/110.jpg)
Other linguistic power law distributions
number of speakers perlanguage
source: Ethnologue
frequency ∼ rank−1.01
●
● ●
●
● ● ●
●
●
●●
●●●●●●●●●
●●●●●●●●●●●●●●●
●●●●●●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
1 2 5 10 20 50 100 2005
1020
5010
020
050
010
00
rank
freq
uenc
y (in
mill
ion)
110/124
![Page 111: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/111.jpg)
The World Atlas of Language Structures
large scale typological database, conducted mainly by the MPIEVA, Leipzig
2,650 languages in total are used
142 features, with between 120 and 1,370 languages perfeature
available online
111/124
![Page 112: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/112.jpg)
The World Atlas of Language Structures
question: are frequency of feature values powerlawdistributed?
problem: number of feature values usually too small forstatistic evaluation
solution:
cross-classification of two (randomly chosen) featuresonly such feature pairs are considered that lead to at least 30non-empty feature value combinations
pilot study with 10 such feature pairs
112/124
![Page 113: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/113.jpg)
The World Atlas of Language Structures
Feature 1:Consonant-VowelRatio
Feature 2: Subtypesof AsymmetricStandard Negation
Kolmogorov-Smirnovtest: positive
100
101
102
10−2
10−1
100
Pr(
X ≥
x)
x
113/124
![Page 114: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/114.jpg)
The World Atlas of Language Structures
Feature 1: WeightFactors inWeight-SensitiveStress Systems
Feature 2: OrdinalNumerals
Kolmogorov-Smirnovtest: positive
100
101
102
10−2
10−1
100
Pr(
X ≥
x)
x
114/124
![Page 115: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/115.jpg)
The World Atlas of Language Structures
Feature 1: ThirdPerson Zero of VerbalPerson Marking
Feature 2: Subtypesof AsymmetricStandard Negation
Kolmogorov-Smirnovtest: positive
100
101
102
10−2
10−1
100
Pr(
X ≥
x)
x
115/124
![Page 116: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/116.jpg)
The World Atlas of Language Structures
Feature 1:Relationship betweenthe Order of Objectand Verb and theOrder of Adjectiveand Noun
Feature 2: Expressionof PronominalSubjects
Kolmogorov-Smirnovtest: positive
100
101
102
103
10−2
10−1
100
Pr(
X ≥
x)
x
116/124
![Page 117: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/117.jpg)
The World Atlas of Language Structures
Feature 1: Plurality inIndependent PersonalPronouns
Feature 2:AsymmetricalCase-Marking
Kolmogorov-Smirnovtest: positive
100
101
102
10−2
10−1
100
Pr(
X ≥
x)
x
117/124
![Page 118: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/118.jpg)
The World Atlas of Language Structures
Feature 1: Locus ofMarking:Whole-languageTypology
Feature 2: Number ofCases
Kolmogorov-Smirnovtest: positive
100
101
102
10−2
10−1
100
Pr(
X ≥
x)
x
118/124
![Page 119: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/119.jpg)
The World Atlas of Language Structures
Feature 1: Prefixingvs. Suffixing inInflectionalMorphology
Feature 2: Coding ofNominal Plurality
Kolmogorov-Smirnovtest: positive
100
101
102
103
10−2
10−1
100
Pr(
X ≥
x)
x
119/124
![Page 120: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/120.jpg)
The World Atlas of Language Structures
Feature 1: Prefixingvs. Suffixing inInflectionalMorphology
Feature 2: OrdinalNumerals
Kolmogorov-Smirnovtest: positive
100
101
102
10−2
10−1
100
Pr(
X ≥
x)
x
120/124
![Page 121: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/121.jpg)
The World Atlas of Language Structures
Feature 1: Coding ofNominal Plurality
Feature 2:AsymmetricalCase-Marking
Kolmogorov-Smirnovtest: positive
100
101
102
10−2
10−1
100
Pr(
X ≥
x)
x
121/124
![Page 122: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/122.jpg)
The World Atlas of Language Structures
Feature 1: Position ofCase Affixes
Feature 2: OrdinalNumerals
Kolmogorov-Smirnovtest: negative
100
101
102
10−2
10−1
100
Pr(
X ≥
x)
x
122/124
![Page 123: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/123.jpg)
Why power laws?
critical states
self-organized criticality
preferential attachment
random walks
...
Preferential attachment
items are stochasticallyadded to bins
probability to end up in binn is linear in number ofitems that are already in binn
123/124
![Page 124: The semantics of color terms. A quantitative cross ...gjaeger/slides/slidesLeipzig.pdf · colors distance is inverse function of perceived similarity L*a*b* color space has this property](https://reader035.vdocuments.us/reader035/viewer/2022063014/5fd0fc3ea663457cdb5230c1/html5/thumbnails/124.jpg)
(Wide) Open questions
Preferential attachment explains power law distribution ifthere are no a priori biases for particular types
first simulations suggest that preferential attachment + biasedtype assignment does not lead to power law
negative message: uneven typological frequency distributiondoes not prove that frequent types are inherently preferredlinguistically/cognitively/socially
unsettling questions:
Are there linguistic/cognitive/social biases in favor of certaintypes?If yes, can statistical typology supply information about this?If power law distributions are the norm, is their any content tothe notion of statistical universal in a Greenbergian sense?
124/124