Finite-state abstractions on Arabic morphology

摘要

There has been much interest recently in two-level and associative models for handling morphologically rich inflectional languages. Such models are claimed to have advantages over generative, rule-based approaches in terms of not just conceptual appropriateness but also computational efficiency. The claim with regard to the former is that, whilst generative approaches to morphology may well be useful for inflectionally simple natural languages such as English (where most of the processing is carried out at the sentence level, with dictionaries and lexicons being accessed to identify secondary inflectional information once primitive words are found), this approach is not at all suitable for inflectionally rich languages where grammatical information is carried not by the combination or pattern of distinct and separate words which make up the sentence but by the combination or pattern of inflections within a ‘word’ where, for instance, there are no clear boundaries between morphological constituents. The claim with regard to the latter is that many generative approaches to natural language are inefficient and, in some cases, computationally intractable, because of the heavy memory and processing demand placed on implementing actual models based on these approaches for anything more than a constrained fragment of a language. This paper describes an application of finite-state automata for Arabic noun inflections which leads to abstractions based on network topology as well as the form and content of network arcs. The idea of specific automata for specific inflection types inheriting some or all of the nodes, arc form and arc content of abstract automata representing more abstract classes of inflection is also introduced. This can lead to novel linguistic generalities and applications, as well as advantages in terms of procedural efficiency and representation.