Generating Determiners and Quantifiers in Hebrew

This paper presents a specific part of HUGG, a generation grammar for Hebrew. This part deals with determiners and quantifiers. Our main goal is to determine which set of features must be present in the input to the generation grammar to control the generation of complex determiners and quantifiers. Hebrew determiners are characterized by two properties: (1) definiteness is marked in several places in the NP and it interacts with compound constructs; (2) the order of appearance of determiners and quantifiers within a complex NP is flexible but still restricted. HUGG is developed with the goal to design an input specification language for syntactic realization as close as possible for English and Hebrew, to allow easy development of bilingual generation applications. We show in this paper how the original set of features controlling the generation of determiners in SURGE, a large generation grammar for English, has been enriched to account for the specificities of Hebrew determiners and how, in the process, SURGE has been modified and improved. We present a set of functional features organized around the categories of amount, partitive and identity and show how these features determine a variety of syntactic constructs. 1 I n t r o d u c t i o n A text generation system generally includes two main components: content determination (that decides what to say) and surface realization (that decides how to say it). Recent research has led to the development of reusable surface realization systems which encapsulate knowledge of syntax for specific languages. Most of this research has been conducted for English, and large generation grammars for English are now available [4], [9]. Multi-lingual generation now appears as the "next frontier" in generation research: it is often easier to develop applications to generate text in different languages than it is to translate text. When developing a multi-lingual application, the engineering strategy is to develop a single content determinat ion module and to plug it more or less unchanged into several surface realization modules for different languages each encapsulating knowledge of a different language. The main problem to address in this context is to determine which interface is appropriate between the content determination module and the linguistic modules. We are investigating this problem for the case of bilingual generation English-Hebrew. We start from an existing surface realization system for English, SURGE, and have developed a generation grammar for Hebrew called HUGG with the goal that the input specification language for both grammars be as close as possible. This paper illustrates the general issue of designing a bilingual generation system by focusing on the specific part of HUGG that deals with determiners and quantifiers. Our main goal is to