MASIVE: Open-Ended Affective State Identification in English and Spanish
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2024
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866912115141378048 |
|---|---|
| author | Deas, Nicholas Turcan, Elsbeth Mejía, Iván Pérez McKeown, Kathleen |
| author_facet | Deas, Nicholas Turcan, Elsbeth Mejía, Iván Pérez McKeown, Kathleen |
| contents | In the field of emotion analysis, much NLP research focuses on identifying a limited number of discrete emotion categories, often applied across languages. These basic sets, however, are rarely designed with textual data in mind, and culture, language, and dialect can influence how particular emotions are interpreted. In this work, we broaden our scope to a practically unbounded set of \textit{affective states}, which includes any terms that humans use to describe their experiences of feeling. We collect and publish MASIVE, a dataset of Reddit posts in English and Spanish containing over 1,000 unique affective states each. We then define the new problem of \textit{affective state identification} for language generation models framed as a masked span prediction task. On this task, we find that smaller finetuned multilingual models outperform much larger LLMs, even on region-specific Spanish affective states. Additionally, we show that pretraining on MASIVE improves model performance on existing emotion benchmarks. Finally, through machine translation experiments, we find that native speaker-written data is vital to good performance on this task. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2407_12196 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | MASIVE: Open-Ended Affective State Identification in English and Spanish Deas, Nicholas Turcan, Elsbeth Mejía, Iván Pérez McKeown, Kathleen Computation and Language In the field of emotion analysis, much NLP research focuses on identifying a limited number of discrete emotion categories, often applied across languages. These basic sets, however, are rarely designed with textual data in mind, and culture, language, and dialect can influence how particular emotions are interpreted. In this work, we broaden our scope to a practically unbounded set of \textit{affective states}, which includes any terms that humans use to describe their experiences of feeling. We collect and publish MASIVE, a dataset of Reddit posts in English and Spanish containing over 1,000 unique affective states each. We then define the new problem of \textit{affective state identification} for language generation models framed as a masked span prediction task. On this task, we find that smaller finetuned multilingual models outperform much larger LLMs, even on region-specific Spanish affective states. Additionally, we show that pretraining on MASIVE improves model performance on existing emotion benchmarks. Finally, through machine translation experiments, we find that native speaker-written data is vital to good performance on this task. |
| title | MASIVE: Open-Ended Affective State Identification in English and Spanish |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2407.12196 |