A multilingual natural-language interface to regular expressions

  • Authors:
  • Aarne Ranta

  • Affiliations:
  • Xerox Research Centre Europe (Grenoble Laboratory) and Academy of Finland, Meylan, France

  • Venue:
  • FSMNLP '09 Proceedings of the International Workshop on Finite State Methods in Natural Language Processing
  • Year:
  • 1998

Quantified Score

Hi-index 0.00

Visualization

Abstract

This report explains a natural-language interface to the formalism of XFST (Xerox Finite State Tool), which is a rich language used for specifying finite state automata and transducers. By using the interface, it is possible to give input to XFST in English and French, as well as to translate formal XFST code into these languages. It is also possible to edit XFST source files and their natural-language equivalents interactively, in parallel. The interface is based on an abstract syntax of the regular expression language and of a corresponding fragment of natural language. The relations between the different components are defined by compositional interpretation and generation functions, and by corresponding combinatory parsers. This design has been inspired by the logical grammar of Montague. The grammar-driven design makes it easy to extend and to modify the interface, and also to link it with other functionalities such as compiling and semantic reasoning. It is also easy to add new languages to the interface. Both the grammatical theory and the interface facilities based on it have been implemented in the functional programming language Haskell, which supports a declarative and modular style of programming. Some of the modules developed for the interface have other uses as well: there is a type system of regular expressions, preventing some compiler errors, a denotational semantics in terms of lazy lists, and an extension of the XFST script language by definitions of functions.