A grammar design accommodating packed argument frame information on verbs

18864

Authors

nan

Abstract

. This paper presents a comparison of two designs for implementing argument frame information on verbs. In the first design, alternating verbs will be represented with one lexical entry pr. possible argument frame. In the other design, each verb form will be associated with only one lexical entry containing a packed representation of the possible ar- gument frames. The first design represents how valence alternations are treated in lexicalist grammars, while the second shows how valence alternations can be handled in a construction- alist grammar. The comparison is done with an implemented “deep” grammar of Norwegian. Keywords: HPSG, grammar engineering, lexical representations, under-specification. 1

Keywords

nan

Full Content

A grammar design accommodating packed argument frame information on verbs Petter Haugereid Division of Linguistics and Multilingal Studies, Nanyang Technological University, 14 Nanyang Drive, Singapore 637332, Singapore petterha@ntu.edu.sg Abstract. This paper presents a comparison of two designs for implementing argument frame information on verbs. In the first design, alternating verbs will be represented with one lexical entry pr. possible argument frame. In the other design, each verb form will be associated with only one lexical entry containing a packed representation of the possible ar- gument frames. The first design represents how valence alternations are treated in lexicalist grammars, while the second shows how valence alternations can be handled in a construction- alist grammar. The comparison is done with an implemented “deep” grammar of Norwegian. Keywords: HPSG, grammar engineering, lexical representations, under-specification. 1 Introduction In most “deep” grammar implementations, information about the argument frame of a verb is spec- ified in the lexicon. This can be observed in grammatical frameworks such as Head-driven Phrase Structure Grammar (HPSG) (Pollard and Sag, 1994), Lexical Functional Grammar (LFG) (Bresnan, 2001), and Combinatory Categorial Grammar (CCG) (Steedman, 2000). Deep grammars need to be as precise as possible since they are not only expected to parse grammatical sentences, but also not to parse ungrammatical sentences. So, in order to avoid overgeneration, the grammar needs, among numerous other things, to contain information about what syntactic frames a verb is expected to appear in. The most natural place to put this information is in the lexicon. In this paper I will discuss one problem associated with the way argument frame information is specified in a lexicalist grammar, namely the use of multiple lexical entries for one verb form in cases where the verb may appear in more than one argument frame. Multiple lexical entries for one form leads to an increased processing effort for the parser. A possible solution to this problem is to pack the information from the different entries into one. I will present a constructionalist grammar design, where each verb form is assigned a single lexical entry with a packed representation of the possible argument frames of the verb, using a type hierarchy of argument frame types to account for argument frame alternations. I will show how the grammar design compares to HPSG, which is the framework mostly used for deep grammar implementations, and compare two versions of the implemented grammar, one with an expanded lexicon and one with a packed lexicon, with regard to competence and performance. 2 Valence in HPSG In HPSG, a transitive verb has the information in Figure 1. The figure illustrates how the valence requirements of a verb is represented by means of lists and also how the linking to the semantics is accounted for. Copyright 2011 by Petter Haugereid 25th Pacific Asia Conference on Language, Information and Computation, pages 31–40 31   PHON � admire � CAT   HEAD verb VAL   SUBJ � NP 1 � COMPS � NP 2 �     …

Content truncated to 500 words

Labeled Text

. This paper presents a comparison of two designs for implementing argument frame information on verbs. In the first design, alternating verbs will be represented with one lexical entry pr. possible argument frame. In the other design, each verb form will be associated with only one lexical entry containing a packed representation of the possible argument frames. The first design represents how valence alternations are treated in lexicalist grammars, while the second shows how valence alternations can be handled in a constructionalist grammar. The comparison is done with an implemented “deep” grammar of Norwegian. Keywords: HPSG, grammar engineering, lexical representations, under-specification. 1

Publication Details

Paper ID:
18864
Year:
2011
Publication:
TIPSTER
Publication Type:
conf
DOI:
nan