Fleshing it out: A Supervised Approach to MWE-token and MWE-type Classification

12255

Authors

nan

Abstract

Although some multiword expressions (MWEs) like How do you do? have ex- clusively idiomatic meaning, other MWE- types like the phrase kick the bucket may be idiomatic or literal depending on con- text. The recently developed OpenMWE corpus provides the largest freely avail- able collection of annotated MWE-tokens suitable for supervised classification, but so far its potential has only been super- ficially investigated and only for classifi- cation of MWE-types in the corpus. In- stead, we train and evaluate classifiers for crosstype classification and introduce novel features specialised to this task. Our best crosstype classifiers performed as well on non-trained MWE-types as a ma- jority class baseline which has knowledge of the MWE-type. 1

Keywords

nan

Full Content

Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 911–919, Chiang Mai, Thailand, November 8 – 13, 2011. c⃝2011 AFNLP Fleshing it out: A Supervised Approach to MWE-token and MWE-type Classification Richard Fothergill and Timothy Baldwin Department of Computer Science and Software Engineering The University of Melbourne VIC 3010 Australia r.fothergill@student.unimelb.edu.au, tb@ldwin.net Abstract Although some multiword expressions (MWEs) like How do you do? have ex- clusively idiomatic meaning, other MWE- types like the phrase kick the bucket may be idiomatic or literal depending on con- text. The recently developed OpenMWE corpus provides the largest freely avail- able collection of annotated MWE-tokens suitable for supervised classification, but so far its potential has only been super- ficially investigated and only for classifi- cation of MWE-types in the corpus. In- stead, we train and evaluate classifiers for crosstype classification and introduce novel features specialised to this task. Our best crosstype classifiers performed as well on non-trained MWE-types as a ma- jority class baseline which has knowledge of the MWE-type. 1 Introduction A multiword expression (MWE) is an idiosyn- cratically interpreted linguistic unit which con- sists of more than a single word (or “crosses word boundaries”) (Sag et al., 2002; Baldwin and Kim, 2009). The nature of these idiosyncrasies can vary greatly — from traffic light and street light which are remarkable only in that they are not in- terchangeable — to how do you do?, the meaning of which is non-compositional in modern English. MWEs will typically resist lexico-syntactic variation to some extent (Fazly et al., 2009). For example, the phrase a picture is worth a thousand words does not allow the freedom of lexical substi- tution and modification that its constituent words would usually enjoy, making otherwise equivalent variations — an image is worth fifty score words, a picture is worth approximately a thousand words — sound wrong or at least unnatural to a native speaker. The meaning of a MWE as a whole may not derive literally from the composition of its con- stituent words (Baldwin et al., 2003). For exam- ple, the English expression kick the bucket may be used to refer to death in any number of ways un- related to either kicking or buckets. In these cases we say the MWE has an idiomatic interpretation. When talking about MWEs we make a dis- tinction between an MWE lexeme out of context, which we call an MWE-type, and specific in- stances of the MWE in context, which we call an MWE-token. Some MWE-types have only an idiomatic meaning, such as the English greeting How do you do?, interpretation of which can be perplexing if attempted literally. However for others, the literal uses are still perfectly valid, and individual MWE- tokens may be ambiguous between an idiomatic and literal meaning. For example, kick the bucket may indeed refer to a violent act against a bucket and have nothing to do with death at all. Relation to Word Sense Disambiguation MWE-token disambiguation can be approached as if it …

Content truncated to 500 words

Labeled Text

Although some multiword expressions (MWEs) like How do you do? have exclusively idiomatic meaning, other MWEtypes like the phrase kick the bucket may be idiomatic or literal depending on context. The recently developed OpenMWE corpus provides the largest freely available collection of annotated MWE-tokens suitable for supervised classification, but so far its potential has only been superficially investigated and only for classification of MWE-types in the corpus. Instead, we train and evaluate classifiers for crosstype classification and introduce novel features specialised to this task. Our best crosstype classifiers performed as well on non-trained MWE-types as a majority class baseline which has knowledge of the MWE-type. 1

Publication Details

Paper ID:
12255
Year:
2011
Publication:
TIPSTER
Publication Type:
conf
DOI:
nan