Deriving Information Structure from Prosodically Marked Text with Lexicalized Tree Adjoining Grammars Gann Bierner and Anoop Sarkar and Aravind Joshi

This paper proposes a method for integrating intonation and information structure into the Lexicalized Tree Adjoining Grammar (LTAG) formalism. The method works fully within LTAG and requires no changes or additions to the basic formalism. From the existing CCG analysis, we denote boundary tones as lexical items and pitch accents as features of lexical items. We then show how prosodically marked text can be parsed to produce a derivation with the correct semantics and the appropriate information structure for the sentence. Although this paper is concerned with the recognition of prosodically marked text, the method described is also applicable to generation. This system has been implemented and tested using a wide-coverage LTAG grammar. The results in this paper also show how an account of intonational structure can be given in a lexicalized grammar with built-in constituencies in LTAG in contrast to lexical systems with exible constituencies as in Combinatory Categorial Grammar (CCG). Submission Type: Regular Paper Topic Areas: L2. Syntax & parsing L3. Semantics, pragmatics, cognition Author of Record: Gann Bierner Under consideration for other conferences (specify)? none Deriving Information Structure from Prosodically Marked Text with Lexicalized Tree Adjoining Grammars Abstract This paper proposes a method for integrating intonation and information structure into the Lexicalized Tree Adjoining Grammar (LTAG) formalism. The method works fully within LTAG and requires no changes or additions to the basic formalism. From the existing CCG analysis, we denote boundary tones as lexical items and pitch accents as features of lexical items. We then show how prosodically marked text can be parsed to produce a derivation with the correct semantics and the appropriate information structure for the sentence. Although this paper is concerned with the recognition of prosodically marked text, the method described is also applicable to generation. This system has been implemented and tested using a wide-coverage LTAG grammar. The results in this paper also show how an account of intonational structure can be given in a lexicalized grammar with built-in constituencies in LTAG in contrast to lexical systems with exible constituencies as in Combinatory Categorial Grammar (CCG).This paper proposes a method for integrating intonation and information structure into the Lexicalized Tree Adjoining Grammar (LTAG) formalism. The method works fully within LTAG and requires no changes or additions to the basic formalism. From the existing CCG analysis, we denote boundary tones as lexical items and pitch accents as features of lexical items. We then show how prosodically marked text can be parsed to produce a derivation with the correct semantics and the appropriate information structure for the sentence. Although this paper is concerned with the recognition of prosodically marked text, the method described is also applicable to generation. This system has been implemented and tested using a wide-coverage LTAG grammar. The results in this paper also show how an account of intonational structure can be given in a lexicalized grammar with built-in constituencies in LTAG in contrast to lexical systems with exible constituencies as in Combinatory Categorial Grammar (CCG).