[Corpora-List] Fwd: Re: Looking for Text Corpora for emotion analysis(list of emotion words with classification)

Michal Ptaszynski ptaszynski at media.eng.hokudai.ac.jp
Mon May 31 04:52:25 CEST 2010

Dear Bei Yu

Unfortunately, the description of the dictionary has appeared mostly in my publications and in a small number of others.

The dictionary has been gathered for about 20 years by a researcher in Japanese lexicography, Akira Nakamura, and contains about 2000 expressions divided into 10 categories. A small number of the expressions appear in more than one category. Nakamura has gathered the expressions manually

from Japanese literature. It was done subjectively, although I have never met any other collection of emotion words performed in such a proficiency. Nakamura was also changing the number of emotion categories during his work, trying different options and mappings of popular emotion categories onto the Japanese language. The final (present) set of categories consists of emotion labels said to be the most appropriate for the Japanese language.

As for the positive and negative words, the dictionary does not differentiate which emotion category is positive, and which is negative. This is, I think, one of the profoundities of this dictionary. As you can easily notice, some categories could be used to express both types of attitudes (e.g. "surprise" could be both positive or negative). However, in my papers I have used a, partially superficial, mapping of those categories on a 2-dimensional affect model by James Russell (valence and polarity). The combination of Nakamura and Russell goes well in most cases, but I had to agree with the fact that the emotion type equivalents are not straightforward. For example, for some positive emotions their negative equivalent would be more than one and vice versa (e.g. equivalent for emotion type "dislike" would be both "foundness" and "joy"). This way of looking at emotions also goes along well with Bob Solomon's claims (see e.g. "Against valence" in "Not Passion's Slave" http://www.psyclab.com/stone/Against%20Valence.doc ) and I would bet that eventually this kind of fuzzy emotion labeling would be widely accepted for English as well.

The only minus of the dictionary by Nakamura is that it covers only about 10-20% of modern Japanese language (I confirmed that several times experimentally). This could mean two things: 1. the dictionary is out of date; or, more probably, 2. the rest of emotive expressions in Japanese is either phrase or sentence based.

There have been some attempts to generate an emotion corpora automatically, but honestly cannot recommend any, as most of them are too simplistic.

At present I am working on developing a method to expand Nakamura's dictionary or generate a new phrasal dictionary.

Best regards and good luck with the research.

PS. I had an interesting conversation about emotions in language and sentiment analysis with Liz Liddy, whom you probably know, some time ago. She came to Hokkaido University for an invited lecture. -- Michal PTASZYNSKI Language Media Laboratory, Graduate School of Information Science and Technology Hokkaido University, Address: Kita-ku, Kita 14 Nishi 9, 060-0814 Sapporo, Japan ptaszynski at media.eng.hokudai.ac.jp, ptaszynski at ieee.org http://arakilab.media.eng.hokudai.ac.jp/~ptaszynski/

------- Treść oryginalnej wiadomości ------- Od: "Bei Yu" <byu at syr.edu> Do: "Michal Ptaszynski" <ptaszynski at media.eng.hokudai.ac.jp> Kopia dla: Temat: Re: [Corpora-List] Looking for Text Corpora for emotion analysis(list of emotion words with classification) Data: Mon, 31 May 2010 00:16:04 +0900

Hi Michael:

Too bad I don't understand Japanese. Do you have any English material about your dictionary? More specially I am wondering how the dictionary was created and the numbers of positive and negative words in the dictionary.

Thanks, Bei

-- Bei Yu Assistant Professor School of Information Studies Syracuse University

On Sat, May 29, 2010 at 10:14 AM, Michal Ptaszynski <ptaszynski at media.eng.hokudai.ac.jp> wrote:
> For what language? I have developed an affect analysis system, for
> Japanese,
> which uses a set of such "emotion words" available as a dictionary.
> http://www.amazon.co.jp/gp/switch-language/product/4490103395/ref=dp_change_lang?ie=UTF8&language=en_JP
> This dictionary uses a 10-type classification of emotions.
> However, if you wish to perform affect analysis for English, you'd be
> probably interested in an emotion classification done especially for this
> language.
> --
> ptaszynski at ieee.org
> ------------------
> Od: Mohammad Fazleh Elahi <mohammadfazlehelahi at gmail.com>
> Do: corpora at uib.no
> Data: Fri, 28 May 2010 00:06:21 -1200
> Temat: [Corpora-List] Looking for Text Corpora for emotion analysis(list
> of
> emotion words with classification)
> I m looking for Text Corpus for emotion analysis.More specificaly list of
> emotion words.
> If there is classifciation of strenght like strong emotion word,Medium
> emotion word,light emotion word then
> that is better for my work.
> I have checked the following sources:
> 1.SentWordNet
> http://sentiwordnet.isti.cnr.it/
> I found Senti word with polarity..positive..negative etc.But there is no
> classification
> of list of emotion words.
> 2.SemEval 2007
> http://www.cse.unt.edu/~rada/affectivetext/#datasets
> There are list of words in seperate file of 6 basic emotions like
> anger,fear,disgust etc.
> This corpora is ok but it covers only very basic emtotions.
> 3.WordNet Domains ..HLT group in FBK
> http://wndomains.fbk.eu/wnaffect.html
> the emotion are classified by positve.negative.neutral.
> I already requested for the corpora still waiting for respeonse.
> 4.Google 'list of emotion words'
> I got couple of them but they are not reliable.
> It will be nice if anyone suggest me that.
> Mohammad Fazleh Elahi
> Master in Language and Communication Technology
> _______________________________________________
> Corpora mailing list
> Corpora at uib.no
> http://mailman.uib.no/listinfo/corpora

More information about the Corpora mailing list