After a brief hiatus, Ling Tea is back this week!
Who: Jeesun Nam (Hankuk University of Foreign Studies)
When: Wednesday, March 26, 3:00-4:00 pm, Rm. 117
Title: Linguistic Resource-based Approach to Automatic Annotation of Polarity-Shifted Expressions
Abstract:
Among a vast amount of work devoted to the analysis of subjective expressions that contain opinions, evaluations or sentiments, comparatively little work has been conducted in examining polarity shifting devices (PSDs) such as negation markers (Polanyi & Zaenen 2004, Kennedy & Inkpen 2006 and Li et al. 2010). PSDs make inappropriate the assumption that the sentimental orientation of the whole text depends on the simple sum of the prior polarities of content words. For example, in the sentence I was hardly satisfied that is comprised of a positive opinion word satisfied, the polarity of whole sentence is reversed because of a PSD hardly. PSDs should be taken into consideration to properly calculate the polarity of opinion sentences.
This study presents ongoing work on a linguistic resource-based approach to automatic annotation of polarity-shifted expressions. In the literature, given that lexicon- or rule-based approaches have shown serious shortcomings such as ‘performed-on-word-level’ problems or ‘poor-recall’ problems, statistical approaches have dominated the research in opinion classification and achieved the state-of-the-art performance. However, the latter approaches rely on the availability of a large amount of human-tagged training data, and the performance is hard to improve unless more reliable linguistic information is provided.
The linguistic resources I propose in this study essentially include two types: Korean electronic dictionary DECO (Nam 2010) conceptually corresponding to the French electronic dictionary DELA constructed in LADL at Paris 7 University, and local syntactic information represented by finite-state local graphs (i.e. Local Grammar Graphs (LGGs) (Gross 1997, 1999)). The lexicon DECO provides the information of inflectional classes, POS types, and morpho-semantic properties including polarity-orientation of opinion words. The LGGs graphically represent PSDs such as negation markers (e.g. ani ‘not’), polarity-reversing predicates (e.g. silphayha- ‘(to) fail’) or concessive connectors (e.g. –ciman ‘although’) occurring in online review texts. The lexicon DECO and LGGs are applied to the detection and automatic annotation of the polarity-shifted expressions through the multi-lingual text processing platform UNITEX, compatible with the above linguistic resources (Paumier 2003, University Paris-Est-Marne-la-Vallée: http://www-igm.univ-mlv.fr/~unitex).
In this talk, I will briefly introduce the organization of the electronic dictionary DECO as well as those of the DELA-French and DELA-English, and the LGG formalism by illustrating some examples of the LGGs on polarity-reversed expressions in Korean and in English. Finally, I will discuss how to recognize and annotate these expressions by applying the dictionaries and LGGs to online review corpora through a freeware platform UNITEX. If time permits, I will demonstrate how to process non-European languages such as Thai or Arabic by UNITEX.