Sentiment calculations?

[Darren Cook]

I wondered if SentiWordNet's values (a measure of positive, negative,
objectivity for each Wordnet synset) can be used directly in a Japanese
(or other language) translation of WordNet.

I felt the close tie between language and culture of the Japanese
language clouds the question. For instance:
He is a professor.
He is an entrepreneur.

A Japanese person might judge professor more positive than an American,
and might judge entrepreneur as more negative. But is recording that
kind of judgement the job of a sentiment dictionary?

If anyone is working on sentiment dictionaries in Japanese, or other
Asian languages, please get in touch or introduce your research.

(I was shocked that accepted practice for sentiment calculations in
non-English languages appears to be to machine-translate each sentence
to English, then do the calculations using English tools.)

Darren

[Francis Bond]

G'day,

On Mon, Nov 11, 2013 at 3:34 PM, Darren Cook <dar...@dcook.org> wrote:
> I wondered if SentiWordNet's values (a measure of positive, negative,
> objectivity for each Wordnet synset) can be used directly in a Japanese
> (or other language) translation of WordNet.

Yes. They will not be perfect (and Sentiwordnet is far from perfect
even for English: values are automatically propagated from some
seeds). Still, I think it is a great starting place.

> I felt the close tie between language and culture of the Japanese
> language clouds the question. For instance:
> He is a professor.
> He is an entrepreneur.
>
> A Japanese person might judge professor more positive than an American,
> and might judge entrepreneur as more negative. But is recording that
> kind of judgement the job of a sentiment dictionary?

I think so. In general, as well as these kinds of social differences
between cultures I think there are many

> If anyone is working on sentiment dictionaries in Japanese, or other
> Asian languages, please get in touch or introduce your research.

I know Fuji-Xerox (and many others) are working on this in Japanese.
There was a nice paper on Thai at SNLP 2013
http://saki.siit.tu.ac.th/snlp2013/uploads_final/80__0c1d71ef1609123d7bd721e9eb4a663e/A%20Context-induced%20Bootstrapping%20Approach%20for%20Constructing%20Contex-tual-Dependent%20Thai%20Sentiment%20Lexicon_SHORT_tun2.pdf

And there is some work on Chinese here at NTU (and all over the place).

> (I was shocked that accepted practice for sentiment calculations in
> non-English languages appears to be to machine-translate each sentence
> to English, then do the calculations using English tools.)

--
Francis Bond <http://www3.ntu.edu.sg/home/fcbond/>
Division of Linguistics and Multilingual Studies
Nanyang Technological University