Information Technology Reference
In-Depth Information
However, besides few markings on the other adjectives, the general tendency comes
clear in the word test: the gesture is described as expressing the speaker's uncertainty
and lack of knowledge. Further more, most subjects interpret the gesture as signalling
the speaker is indecisive about what to say (hesitant about the topic), but 16%
interpret the speaker's disinclination to speak as the speaker being sure and certain in
the situation (no need or willingness to continue).
Typically the subjects selected 4-5 adjectives, except for the Greek and Uzbek,
who had picked the total of 13 and 10 adjectives, respectively. Based on the selected
adjectives, and a few background features of the subjects such as age, nationality,
experience abroad, we then formed vectors that characterized each subject concerning
their interpretation of the shrug, and then used these vectors to compare the
interpretations automatically. Comparison was done with the Support Vector Machine
technique in the Weka software [11], as a simple binary classification of the vectors
into those that represented Western (European) vs. Eastern (non-European) culture
groups. The classification reached 75% accuracy over 10-fold cross-validation, i.e.
3/4 of the subjects were correctly classified into either group on the basis of the
features encoded in the vectors.
Considering then how similar the vector representations are, and if they form any
natural clusters on the basis of the adjectives, we used the Expectation-Maximization
clustering of Weka. The evaluation of the clusters with respect to the two culture
groups produced 61% accuracy, i.e. almost 2/3 of the subjects had similar
characterizing vectors (had selected similar adjectives) as those who belong to the
same culture group.
Both accuracies can be considered fair results, especially as the dataset is small and
the vectors are sparse. However, as already mentioned, our study was not intended as
a statistical analysis of the interpretation of the shoulder shrug but as a preliminary
exploration of intercultural communication. It is obvious that a larger group of
subjects with a balanced sampling of various affecting characteristics and parameters
would be necessary for a more detailed statistical study, but it is interesting that this
kind of preliminary classification and clustering experiments already provide support
for the cultural differences and the tendency to describe the shoulder shrug differently
in the Eastern and Western cultures is visible in the answers drawn from a small set of
42 individuals only.
Finally, the word selection test also showed that a more detailed analysis of the
semantics of hesitation related words is necessary. In the above experiments, using
Weka's automatic feature analysis techniques, we found that the adjectives
certain
and
unconvinced
as well as
unhesitant
and
inconfident
were selected as the most
predictive adjectives with respect to the two classes (culture groups). The selection
was based solely on the statistical properties of the vectors by measuring the
information gain with respect to the class, and considering the individual predictive
ability of each adjective together with the degree of redundancy between them (i.e.
correlation of the adjectives with the classes). The result is in accordance with what
has been discussed above qualitatively: interpretation of the shoulder shrug can be
described in terms of Western (European) view of the speaker who is
unconvinced
and
inconfident
or with the Eastern (non-European) view of a
certain
and
unhesitant
speaker (which further implies that the gesture is not much used as too arrogant
behaviour is socially inacceptable). However, it would be useful to conduct a
