Bound-Movement Grammar for Natural Language Parsing
Jan 1996
Natural language sentence parsing requires explicit description of restricted (bounded) movement of sentential constituents in interrogative sentences, relative clauses, topicalization, clitics, clefting, etc. This description may be principle-based as in Chomskian approaches, or may have a rather intense programming flavour as in LFG, HPSG and many others theoretical approaches to parsing. In this paper we will be concerned with the computational description of this kind of movement and the usual parsing problems associated with it. So, we will introduce the formalism named BMG (Bound-movement grammar) that takes into account Chomsky's theoretical approaches to Government and Binding and Barriers. This formalism extends earlier logic based approaches to grammar description, namely Definite Clause Grammars and Extraposition Grammars. This is achieved by proliferating the number of data structures - stores (for handling the input data to be parsed and the parsed material that contextually is identified as having been a-moved). Each of these structures is used for keeping track of just one kind of contextually identified movement (interrogation, topicalization, cliticization, clefting, ...). Additionally, syntactic categories (in the left hand side of grammar rules) are explicitly prefixed by named barriers. Each of these barriers act selectively on one store. In this paper we will just present the grammar formalism and will describe a top-down parser using a toy BMG for Portuguese in order to describe topicalization, relative clauses and wh-question.
