Dependency based logical form transformations

Dependency Based Logical Form Transformations Stephen Anthony and Jon Patrick School of Information Technologies The University of Sydney Sydney, Australia 2006 {stephen,jonpat}@it.usyd.edu.au Abstract This paper describes a system developed for the transformation of English sentences into a first order logical form representation. The method- ology is centered on the use of a dependency grammar based parser. We demonstrate the suit- ability of applying a dependency parser based solution to the given task and in turn explain some of the limitations and challenges involved when using such an approach. The efficiencies and deficiencies of our approach are discussed as well as considerations for further enhanc e- ments. 1 Introduction In addition to the well-known all words and lexi- cal sample tasks deployed in previous Senseval workshops a number of new tasks have been in- cluded in this sense evaluation. These new tasks in- clude identification of semantic roles as in FrameNet (Gildea and Jurafsky 2002), disambiguation of WordNet glosses (Miller 1990; Fellbaum 1998; Harabagiu, Miller et al. 1999), automatic acquisition of subcategorisation frames (Korhonen 2002; Preiss and Korhonen 2002), and Logical Form Identific a- tion (LFI) (Rus 2002; Rus and Moldovan 2002). This paper discusses a solution developed for the LFI task. The approach used here employs a func- tional dependency parser (Järvinen and Tapanainen 1997; Tapanainen and Järvinen 1997) and uses a limited number of additional resources. This contri- bution is intended to demonstrate the suitability of a dependency parser to the given task and also explain some of the limitations and challenges involved when using such an approach. 1.1 Motivation Part of the initial step towards the interpretation of a sentence as postulated by Hobbs et al. (1993) in- volves the proof of the logical form of a sentence. This statement entails the transformation of a sen- tence into a logical form as a fundamental building block towards sentence interpretation. Advantages specifically related to the utilisation of logical forms in language processing include a simplified interface between syntax and semantics, a natural and easily exploitable representation of syn- tactic arguments, and the potential for formation of conceptual predicates (Rus 2002) if predicates are disambiguated with respect to a general ontology such as WordNet. 1.2 Task Description The Logical Form (LF) employed in this task is a flat, scope-free first order logic representation that embeds lexical and syntactic information. A predi- cate is generated for every nominal, verbal, adjecti- val, and adverbial content word. The name of the predicate is a concatenation of the lemmatised word form and part-of-speech category. The sentence be- low is followed by its corresponding target logical form representation. Some students like to study in the mornings. student:n_ (x3) like:v_ (e4, x3, e6) to (e5, e6) study:v_ (e6, x3, x9) in (e7, x9) morning:n_ (x9). Relationships between predicates are shared through their arguments. The two types of argu- ments used are events (e) and entities (x). Using the transformation shown above as an example, the event predicate ‘like’ is labeled as e4 and has subject argument …

Content truncated to 500 words