Figurative Language Processing
Advanced course in Language and Computation, taught at ESSLLI 2007
19th European Summer School in Logic, Language and Information, Trinity College, Dublin, Ireland, 6-17 August, 2007
|
Figurative Language ProcessingAdvanced course in Language and Computation, taught at ESSLLI 200719th European Summer School in Logic, Language and Information, Trinity College, Dublin, Ireland, 6-17 August, 2007 |
|
Thanks to the course participants for their feedback! Below, you will find an overview of the course topics as well as copies of the slides for download.
... and Natural Language Processing (NLP)
The literal-figurative scale; figurativeness criteria from Figurative Language (Dobrovol'skij and Piirainen 2005); borderline cases: non-literal, but non-figurative expressions. When applied to lexical units, "literal" and "non-literal" refer to different word senses. Why do people use figurative language? Relevance to NLP, relevance to translation/MT, discussion of word alignment experiment (examples).
Definition (one entity stands for a related one). Regular patterns of metonymy: Existence of patterns universal, but not all individual patterns available in each language. Violation of selectional restrictions led to proposal of type coercion; curious phenomenon of multiple reference led to proposal of predicate coercion. Corpora have been annotated for metonymically used named entities. Metonymy recognition as Word Sense Disambiguation: A supervised learning approach by Nissim and Markert (2003); other machine learning approaches.
Linguistic theories of metaphor; (psycholinguistic processing models and experiments). Metaphor and Computation. Metaphor and selectional restrictions. Metaphor reasoning systems. Metaphor resources (databases, annotated corpora). Metaphor detection.
A class of linguistic expressions with fuzzy boundaries: Most idioms are multi-word expressions; most idioms are "irregular", "special", or "restricted" in at least some respects. Idioms across languages and cultures. Why should NLP care? - Ambiguity, frequency, idiom dictionaries, lexical representation, Machine Translation. Idiom extraction: distinguish between regular multi-word expressions and idioms (Fazly and Stevenson 2006; Degand and Bestgen 2003).
Humor is different from previously discussed phenomena: lexicon vs. text, detection (recognition) vs. generation. Humor theories. Approaches to computational humor generation: types and strategies. Pun generation vs. NLG: templates, communicative goal. Joke generation vs. joke collection and retrieval. An anaphora joke generator: resources and components.