Date post: | 16-Jul-2015 |
Category: |
Technology |
Upload: | yaevents |
View: | 390 times |
Download: | 2 times |
Mark L. Latash with Irina L. Mikaelian,
Elizaveta M. Latash, Jaebum Park, and Yao Sun
Simple information transmissionby different natural languages
The Pennsylvania State University
Language and Movement
An axiom: These two functions have a lot in common inthe organization of neural control.
Reasons: Both are characterized by hierarchical controland abundance of solutions.
Both are biological time processes.
How to study a biological time process?• Average characteristics;• Variability;• Stability.
Very little is known about speech.
Where to Start?
How long does it take a person to create and uttera phrase describing a situation (a meaning, in ageneral sense) quickly and accurately?
Surprisingly, there is little information.
Sternberg:Quick production of words, numerals, and non-words.Tasks were presented in advance.Speech time increased with NW and NSYL non-linearly(speeding-up for longer tasks).It was not necessary to create a phrase.
Fitts’ Law
MT = a + b•log2ID; where ID = 2D/W,D is movement distance,W is target width,a and b are constants.
Speech analogies:D - number of objects to be described,W - inverse of the number of characteristics of each object.
We expect:ST = a + b• log2ID, where ID = (NO•NC)NO- number of objects; NC - number of characteristics
Experiment-1
Subjects: Six English speakers, six non-English speakersfluent in English.
Task: Describe a picture as quickly and accurately aspossible (non-English speakers performed twice).
Pictures: One to six objects, all of the same kind or alldifferent.
Four Series:S1 - only objects; S2 - objects and color (out of 4); S3 -objects, color, and size (out of 2); S4 - objects, color, size,and relative location (over/on, under).
Speech time changed linearly with NO and NC:ST = a + k•NO•NC
Different from the log-relation typical of Fitts’ Law.
Lessons from Experiment #1
Performance in the second language was characterizedby longer RT and ST, while the function remained the same.
Why?1). A specific strategy of description - sequential.
2). Tracking the initial segment of a non-linear function.
Experiment-2
Subjects: Six Indo-European speakers (three English andthree Russian), six Chinese (Mandarin) speakers.
Task: Describe a picture as quickly and accurately aspossible.
Pictures: Always only two objects.
One Series: Presented as a sequence of three mini-series.
0.8
0.9
1
1.1
1.2
1.3
stick fork knife
CHIEALL
RT (s
)
OBJECT
0.8
0.85
0.9
0.95
1
1.05
1.1
1.15
1.2
simple complex
CHIE
RT (s
)
Orientation
Reaction time depended on ID and wasshorter in Chinese speakers
0
2
4
6
8
10
LsOs LsOc LcOs LcOc
ForkKnife
Stick
ST (s
)
CONDITION
A
1
2
3
4
5
6
7
0 5 10 15 20 25 30 35ID
B
1
2
3
4
5
6
7
0 0.5 1 1.5 2 2.5 3 3.5
y = 1.31 + 1.58x R= 0.956
ST(s
)
ln(ID)
C
1
2
3
4
5
6
7
0 0.5 1 1.5 2 2.5 3 3.5
y = 1.27 + 1.56x R= 0.996
ln(ID)
D
Stepsin
SpeechTimeData
Processing
y = 1.27 + 1.56x R= 0.996 y = 1.70 + 1.92x R= 0.996y = 1.51 + 1.17x R= 0.993y = 1.71 + 1.04x R= 0.995 y = 1.33 + 1.58x R= 0.997 y = 0.99 + 1.29x R= 0.982
y = 1.51 + 0.497x R= 0.982y = 1.34 + 0.750x R= 0.990y = 1.33 + 0.785x R= 0.995y = 1.40 + 0.665x R= 0.957y = 1.41 + 0.451x R= 0.990y = 1.96 + 0.625x R= 0.987
IE: y = 1.42 + 1.43x R= 0.999 CH: y = 1.49 + 0.63x R= 0.996
IE CH
IE
CH
}}
0
2
4
6
8
10
0 0.5 1 1.5 2 2.5 3 3.5
ST (s
)
ln(ID)
ST = a + k•log2(ID)Chinese Speakers were Much Faster
Speech time changed logarithmically with ID:Typical of Fitts’ Law.
Lessons from Experiment #2
Chinese speakers were much faster:Tone as a means of information transmission?
Schematic pictures reminding Chinese characters?Quicker analysis of the picture (supported by shorter RT)?
Trading speed for clarity, more errors?
Speed-difficulty trade-off works in speechsimilarly to how it works in movements.
Experiment-3
Subjects: Eight English speakers, eight Chinese(Mandarin) speakers [plus seven Korean speakers, twoRussian speakers, and four Vietnamese speakers].
Task: Speaker: Describe the picture as quickly andaccurately as possible. Performer: Use the actual objects tocreate the picture described by the Speaker also "asquickly and accurately as possible". Mouse clicks wereused to define the time of action initiation and termination.
Pictures: Always only two objects.
Speaker-Performer & Performer-Only: Each presented asa sequence of four mini-series.
t0(Picture)
tSI(Speech start)
tST(Speech end)
tPI(Action start)
tPT(Action end)
RT = tSI - t0ST = tST - tSIΔSM = tPI - tSIMT = tPT - tPItTT = tPT - t0
RT ΔSM
Error rate was smaller in Chinese subjects (non-sign)
Speaker Performer
Chinese 3.7 4.4English 8.5 4.75Korean 5.4 2.9
Object - 6%Location - 23%Orientation - 71%
0
1
2
3
4
5
6
7
8
ST MT TT
B
Time Index0
0.2
0.4
0.6
0.8
1
1.2
1.4
RT ΔSM
A CHINESEENGLISH
Tim
e (s
)
Time Index
Chinese subjects were faster in all indices
0.6
0.8
1
1.2
1.4
0 0.5 1 1.5 2 2.5 3 3.5
CHINESEENGLISH
y = 0.917 + 0.032x R= 0.730 y = 1.13 + 0.005x R= 0.176
Reac
tion
Tim
e (s
)
ln(ID)
RT showed no scaling with ln(ID) forboth groups
0.8
1
1.2
1.4
1.6
1.8
Chinese English Russian Korean Vietnamese
RT (s
)
Language
RT across the languages
0
2
4
6
8
10
0 0.5 1 1.5 2 2.5 3 3.5
A: Chinese
CH-1CH-2CH-3CH-4CH-5CH-6CH-7CH-8CH-Average
y = 1.33 + 0.79x R= 0.975 y = 1.31 + 0.86x R= 0.913 y = 1.40 + 1.08x R= 0.992y = 1.36 + 0.76x R= 0.964y = 1.71 + 1.06x R= 0.983 y = 1.60 + 1.08x R= 0.998y = 1.75 + 0.80x R= 0.944y = 1.04 + 1.04x R= 0.992
y = 1.44 + 0.93x R= 0.994
Spee
ch T
ime
(s)
ln(ID)
0
2
4
6
8
10
0 0.5 1 1.5 2 2.5 3 3.5
B: English
EN-1EN-2EN-3EN-4EN-5EN-6EN-7EN-8EN-Average
y = 1.84 + 1.35x R= 0.994y = 1.92 + 1.27x R= 0.988y = 1.91 + 1.84x R= 0.990 y = 1.99 + 1.21x R= 0.995 y = 1.39 + 1.60x R= 0.981 y = 1.64 + 1.19x R= 0.980y = 2.19 + 1.21x R= 0.982y = 1.69 + 1.28x R= 0.976 y = 1.95 + 1.17x R= 0.986
ln(ID)
0
2
4
6
8
10
0 0.5 1 1.5 2 2.5 3 3.5
C: Korean
Kor-1Kor-2
y = 2.28 + 1.78x R= 0.991 y = 1.83 + 2.13x R= 0.995
ln(ID)
ST = a + k•log2(ID)Chinese subjects were much faster; Korean subjects were slower
0
2
4
6
8
10
-0.5 0 0.5 1 1.5 2 2.5 3 3.5
y = 1.0795 + 1.5633x R= 0.97551 y = 0.77799 + 1.1513x R= 0.98857 y = 0.76324 + 1.3679x R= 0.97654 y = 0.77267 + 2.3452x R= 0.9726
Spee
ch T
ime
(s)
ln(ID)
Vietnamese subjects wereslower than Chinese
D:Vietnamese
0.8
1
1.2
1.4
1.6
1.8
2
2.2
2.4
Chinese English Russian Korean Vietnamese
k-ST
[s/ln
(ID)]
Language
k-values [ST = a + k•log2(ID)] across the languages
0
0.5
1
1.5
2
0 0.5 1 1.5 2 2.5 3 3.5
CHINESEENGLISH
Tim
e (s
)
ln(ID)
ΔSM
Speech-Action Delay was shorter for theChinese subjects; no scaling with ln(ID)
2
4
6
8
10
0 0.5 1 1.5 2 2.5 3 3.5
A: Chinese
CH-1CH-2CH-3CH-4CH-5CH-6CH-7CH-8CH-Average
y = 2.87 + 1.06x R= 0.96 y = 3.14 + 1.04x R= 0.954y = 1.89 + 0.942x R= 0.978y = 2.76 + 1.03x R= 0.984y = 2.49 + 1.16x R= 0.988y = 2.70 + 1.08x R= 0.971y = 2.75 + 1.11x R= 0.981y = 2.75 + 0.88x R= 0.986
y = 2.67 + 1.04x R= 0.990
Mov
emen
t Tim
e (s
)
ln(ID)
2
4
6
8
10
0 0.5 1 1.5 2 2.5 3 3.5
C: Korean
Kor-1Kor-2
y = 2.81 + 2.13x R= 0.998y = 3.10 + 1.98x R= 0.980
ln(ID)
2
4
6
8
10
0 0.5 1 1.5 2 2.5 3 3.5
B: English
EN-1EN-2EN-3EN-4EN-5EN-6EN-7EN-8EN-Average
y = 2.71 + 1.63x R= 0.989 y = 2.34 + 1.27x R= 0.979y = 2.72 + 1.49x R= 0.983y = 2.97 + 1.22x R= 0.977y = 2.11 + 2.06x R= 0.990y = 2.66 + 1.27x R= 0.987y = 1.99 + 1.15x R= 0.983y = 2.12 + 1.59x R= 0.977
y = 2.45 + 1.46x R= 0.995
ln(ID)
MT = a + k•log2(ID)Chinese subjects were much faster; Korean subjects were slower
Vietnamese subjects wereslower than Chinese
D:Vietnamese
2
4
6
8
10
12
-0.5 0 0.5 1 1.5 2 2.5 3 3.5
y = 3.5701 + 2.0791x R= 0.98227 y = 3.4696 + 1.9056x R= 0.99352 y = 3.3562 + 1.5716x R= 0.98669 y = 3.0501 + 2.3029x R= 0.96278
Mov
emen
t Tim
e (s
)
ln(ID)
2
4
6
8
10
12
0 0.5 1 1.5 2 2.5 3 3.5
A: Chinese
CH-1CH-2CH-3CH-4CH-5CH-6CH-7CH-8CH-Average
y = 4.52 + 1.06x R= 0.945y = 4.16 + 1.11x R= 0.969y = 3.10 + 1.07x R= 0.971y = 4.18 + 0.976x R= 0.990y = 3.98 + 1.20x R= 0.991y = 4.06 + 1.15x R= 0.961y = 4.09 + 1.15x R= 0.985y = 4.52 + 1.08x R= 0.982
y = 4.08 + 1.10x R= 0.993
Tota
l Per
form
ance
Tim
e (s
)
ln(ID)
2
4
6
8
10
12
0 0.5 1 1.5 2 2.5 3 3.5
B: English
EN-1EN-2EN-3EN-4EN-5EN-6EN-7EN-8EN-Average
y = 4.70 + 1.71x R= 0.985y = 5.74 + 1.11x R= 0.932y = 4.54 + 1.54x R= 0.981y = 4.87 + 1.22x R= 0.965y = 3.67 + 2.07x R= 0.989y = 4.20 + 1.23x R= 0.962y = 4.45 + 1.31x R= 0.987y = 4.26 + 1.55x R= 0.978
y = 4.55 + 1.47x R= 0.993
ln(ID)
2
4
6
8
10
12
0 0.5 1 1.5 2 2.5 3 3.5
C: Korean
Kor-1Kor-2
y = 4.46 + 2.22x R= 0.998y = 4.76+ 1.92x R= 0.992
ln(ID)
TT = a + k•log2(ID)Chinese subjects were faster; Korean subjects were slower
D:Vietnamese
Vietnamese subjects wereslower than Chinese
4
6
8
10
12
14
-0.5 0 0.5 1 1.5 2 2.5 3 3.5
y = 5.4688 + 2.2244x R= 0.98348 y = 5.4777 + 1.8694x R= 0.99002 y = 5.3316 + 1.4588x R= 0.96922 y = 4.5178 + 2.5043x R= 0.96256
Tota
l Tim
e (s
)
ln(ID)
1.5
2
2.5
3
3.5
0 0.5 1 1.5 2 2.5 3 3.5
A: ChineseCH-1CH-2CH-3CH-4CH-5CH-6CH-7CH-8CH-Average
y = 1.90 + 0.22x R= 0.843 y = 2.26 + 0.21x R= 0.851 y = 1.69 + 0.16x R= 0.833 y = 1.64 + 0.23x R= 0.878y = 1.60 + 0.17x R= 0.857y = 2.03 + 0.14x R= 0.824y = 1.95 + 0.25x R= 0.775y = 1.54 + 0.13x R= 0.925
y = 1.82 + 0.19x R= 0.914
ln(ID)
Tota
l Per
form
ance
Tim
e (s
)
1.5
2
2.5
3
3.5
0 0.5 1 1.5 2 2.5 3 3.5
B: EnglishEN-1EN-2EN-3EN-4EN-5EN-6EN-7EN-8EN-Average
y = 2.37 + 0.05x R= 0.470 y = 1.83 + 0.07x R= 0.897y = 1.44 + 0.22x R= 0.866y = 1.69 + 0.14x R= 0.906y = 1.91 + 0.16x R= 0.912y = 1.64 + 0.14x R= 0.851y = 1.54 + 0.11x R= 0.943y = 1.83 + 0.13x R= 0.911
y = 1.78 + 0.13x R= 0.948
ln(ID)
Performer-only
Performer-Only: PT = a + k•log2(ID)Chinese subjects were significantly SLOWER
1
1.5
2
2.5
3
3.5
0 0.5 1 1.5 2 2.5 3 3.5
KOREAN
y = 1.81 + 0.196x R= 0.9840y = 1.84 + 0.153x R= 0.991y = 2.1 + 0.199x R= 0.8234y = 1.64 + 0.168x R= 0.875y = 1.84 + 0.236x R= 0.936y = 1.87 + 0.27x R= 0.981y = 1.81 + 0.101x R= 0.912y = 2.1 + 0.186x R= 0.938
y = 1.87 + 0.189x R= 0.981
MT
- Per
form
er -
Onl
y (s
)
ln(ID)
5
10
15
20
25
30
35
0 0.5 1 1.5 2 2.5 3 3.5
A: Chinese
CH-1CH-2CH-3CH-4CH-5CH-6CH-7CH-8CH-Average
y = 6.38 + 1.79x R= 0.832y = 6.42 + 2.56x R= 0.965y = 7.80 + 3.41x R= 0.973y = 8.07 + 2.04x R= 0.782y = 11.0 + 3.40x R= 0.977y = 11.6 + 2.84x R= 0.993y = 7.78 + 2.75x R= 0.977y = 7.03 + 3.36x R= 0.964
y = 8.38 + 2.77x R= 0.982
N SYL
5
10
15
20
25
30
35
0 0.5 1 1.5 2 2.5 3 3.5
B: English
EN-1EN-2EN-3EN-4EN-5EN-6EN-7EN-8EN-Average
y = 10.9 + 1.84x R= 0.924y = 9.83 + 2.76x R= 0.96y = 11.2 + 5.31x R= 0.989y = 11.4 + 4.72x R= 0.898y = 13.7 + 6.14x R= 0.949y = 11.3 + 3.65x R= 0.982y = 9.24 + 4.60x R= 0.982y = 12.2 + 3.69x R= 0.968
y = 11.2 + 4.09x R= 0.987
Number of Syllables: NSYL = a + k*ln(ID)Chinese subjects used fewer syllables per ID unit
2
3
4
5
6
7
8
9
10
0 0.5 1 1.5 2 2.5 3 3.5
A: ChineseCH-1CH-2CH-3CH-4CH-5CH-6CH-7CH-8CH-Average
RSP
(NSY
L/s)
ln(ID)
2
3
4
5
6
7
8
9
10
0 0.5 1 1.5 2 2.5 3 3.5
B: EnglishEN-1EN-2EN-3EN-4EN-5EN-6EN-7EN-8EN-Average
ln(ID)
Speech Rate
Speech rate did not differ between the groups.There was a significant drop from ln(ID)=0 and no further changes.
Speech time changed logarithmically with ID:Typical of Fitts’ Law (NB: Performer-Only!)
Lessons from Experiment #3
Chinese speakers are faster in every timing index.
Speed-difficulty trade-off works across languages
The advantage is NOT in:Chinese characters
TonesHigher error rate
Topics for Discussion:
Not all languages are created equal.
Why is Chinese so much more efficient???
RT = RTTRUE + STLATENT
Do we have a nice tool to study languages and populations?
TOEFL should be replaced with TOCFL.