Abstract
Distributional hypothesis has been playing a central role in statistical NLP. Recently, however, its limitation in incorporating perceptual and empirical knowledge is noted, eliciting a field of perceptually grounded computational semantics. Typical sources of features in such a research are image datasets, where images are accompanied by linguistic tags and/or descriptions. Mainstream approaches employ machine learning techniques to integrate/combine visual features with linguistic features. In contrast to or supplementing these approaches, this study assesses the effectiveness of social image tags in generating word embeddings, and argues that these generated representations exhibit somewhat different and favorable behaviors from corpus-originated representations. More specifically, we generated word embeddings by using image tags obtained from a large social image dataset YFCC100M, which collects Flickr images and the associated tags. We evaluated the efficacy of generated word embeddings with standard semantic similarity/relatedness tasks, which showed that comparable performances with corpus-originated word embeddings were attained. These results further suggest that the generated embeddings could be effective in discriminating synonyms and antonyms, which has been an issue in distributional hypothesis-based approaches. In summary, social image tags can be utilized as yet another source of visually enforced features, provided the amount of available tags is large enough.
| Original language | English |
|---|---|
| Title of host publication | LREC 2018 - 11th International Conference on Language Resources and Evaluation |
| Editors | Hitoshi Isahara, Bente Maegaard, Stelios Piperidis, Christopher Cieri, Thierry Declerck, Koiti Hasida, Helene Mazo, Khalid Choukri, Sara Goggi, Joseph Mariani, Asuncion Moreno, Nicoletta Calzolari, Jan Odijk, Takenobu Tokunaga |
| Publisher | European Language Resources Association (ELRA) |
| Pages | 969-973 |
| Number of pages | 5 |
| ISBN (Electronic) | 9791095546009 |
| Publication status | Published - 2019 Jan 1 |
| Event | 11th International Conference on Language Resources and Evaluation, LREC 2018 - Miyazaki, Japan Duration: 2018 May 7 → 2018 May 12 |
Other
| Other | 11th International Conference on Language Resources and Evaluation, LREC 2018 |
|---|---|
| Country | Japan |
| City | Miyazaki |
| Period | 18/5/7 → 18/5/12 |
Fingerprint
Keywords
- Antonyms
- Image tags
- Semantic similarity
- Social media
- Synonyms
- Word embeddings
ASJC Scopus subject areas
- Linguistics and Language
- Education
- Library and Information Sciences
- Language and Linguistics
Cite this
Social image tags as a source of word embeddings : A Task-oriented Evaluation. / Hasegawa, Mika; Kobayashi, Tetsunori; Hayashi, Yoshihiko.
LREC 2018 - 11th International Conference on Language Resources and Evaluation. ed. / Hitoshi Isahara; Bente Maegaard; Stelios Piperidis; Christopher Cieri; Thierry Declerck; Koiti Hasida; Helene Mazo; Khalid Choukri; Sara Goggi; Joseph Mariani; Asuncion Moreno; Nicoletta Calzolari; Jan Odijk; Takenobu Tokunaga. European Language Resources Association (ELRA), 2019. p. 969-973.Research output: Chapter in Book/Report/Conference proceeding › Conference contribution
}
TY - GEN
T1 - Social image tags as a source of word embeddings
T2 - A Task-oriented Evaluation
AU - Hasegawa, Mika
AU - Kobayashi, Tetsunori
AU - Hayashi, Yoshihiko
PY - 2019/1/1
Y1 - 2019/1/1
N2 - Distributional hypothesis has been playing a central role in statistical NLP. Recently, however, its limitation in incorporating perceptual and empirical knowledge is noted, eliciting a field of perceptually grounded computational semantics. Typical sources of features in such a research are image datasets, where images are accompanied by linguistic tags and/or descriptions. Mainstream approaches employ machine learning techniques to integrate/combine visual features with linguistic features. In contrast to or supplementing these approaches, this study assesses the effectiveness of social image tags in generating word embeddings, and argues that these generated representations exhibit somewhat different and favorable behaviors from corpus-originated representations. More specifically, we generated word embeddings by using image tags obtained from a large social image dataset YFCC100M, which collects Flickr images and the associated tags. We evaluated the efficacy of generated word embeddings with standard semantic similarity/relatedness tasks, which showed that comparable performances with corpus-originated word embeddings were attained. These results further suggest that the generated embeddings could be effective in discriminating synonyms and antonyms, which has been an issue in distributional hypothesis-based approaches. In summary, social image tags can be utilized as yet another source of visually enforced features, provided the amount of available tags is large enough.
AB - Distributional hypothesis has been playing a central role in statistical NLP. Recently, however, its limitation in incorporating perceptual and empirical knowledge is noted, eliciting a field of perceptually grounded computational semantics. Typical sources of features in such a research are image datasets, where images are accompanied by linguistic tags and/or descriptions. Mainstream approaches employ machine learning techniques to integrate/combine visual features with linguistic features. In contrast to or supplementing these approaches, this study assesses the effectiveness of social image tags in generating word embeddings, and argues that these generated representations exhibit somewhat different and favorable behaviors from corpus-originated representations. More specifically, we generated word embeddings by using image tags obtained from a large social image dataset YFCC100M, which collects Flickr images and the associated tags. We evaluated the efficacy of generated word embeddings with standard semantic similarity/relatedness tasks, which showed that comparable performances with corpus-originated word embeddings were attained. These results further suggest that the generated embeddings could be effective in discriminating synonyms and antonyms, which has been an issue in distributional hypothesis-based approaches. In summary, social image tags can be utilized as yet another source of visually enforced features, provided the amount of available tags is large enough.
KW - Antonyms
KW - Image tags
KW - Semantic similarity
KW - Social media
KW - Synonyms
KW - Word embeddings
UR - http://www.scopus.com/inward/record.url?scp=85059881983&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85059881983&partnerID=8YFLogxK
M3 - Conference contribution
SP - 969
EP - 973
BT - LREC 2018 - 11th International Conference on Language Resources and Evaluation
A2 - Isahara, Hitoshi
A2 - Maegaard, Bente
A2 - Piperidis, Stelios
A2 - Cieri, Christopher
A2 - Declerck, Thierry
A2 - Hasida, Koiti
A2 - Mazo, Helene
A2 - Choukri, Khalid
A2 - Goggi, Sara
A2 - Mariani, Joseph
A2 - Moreno, Asuncion
A2 - Calzolari, Nicoletta
A2 - Odijk, Jan
A2 - Tokunaga, Takenobu
PB - European Language Resources Association (ELRA)
ER -