WEDNESDAY: Parsing Flexible Word Order Languages

A parser for "flexible" word order languages must be substantially data driven. In our view syntax has two distinct roles in this connection: (i) to give impulses for assembling cognitive representations, (ii) to structure the space of search for fillers. WEDNESDAY is an interpreter for a language describing the lexicon and operating on natural language sentences. The system operates from left to right, interpreting the various words comprising the sentence one at a time. The basic ideas of the approach are the following: a) to introduce into the lexicon linguistic knowledge that in other systems is in a centralized module. The lexicon therefore carries not only morphological data and semantic descriptions. Also syntactic knowledge is distributed throughout it, partly of a procedural kind. b) to build progressively a cognitive representation of the sentence in the form of a semantic network, in a global space, accessible from all levels of the analysis. c) to introduce procedures invoked by the words themselves for syntactic memory management. Simply stated, these procedures decide on the opening, closing, and mantaining of search spaces; they use detailed constraints and take into account the active expectations. WEDNESDAY is implemented in MAGMA-LISP and with a stress on the non-deterministic mechanism. I. Parsing typologically diverse languages emphasizes aspects that are absent or of little importance in English. By taking these problems into account, some light may be shed on: a) insufficiently treated psycholinguistic aspects b) a design which is less language-dependent c) extraand non-grammatical aspects to be taken into consideration in designing a friendly English The work reported here has largely involved problems with parsing Italian. One of the typical features of Italian is a lower degree of word order rigidity in sentences. For instance, "Paolo ama Maria" (Paolo loves Maria) may be rewritten without any significant difference in meaning (leaving aside questions of context and pragmatics) in any the six possible permutations: Paolo ama Maria, Paolo Maria ama, Maria ama Paolo, Maria Paolo ama, ama Paolo Maria, ama Maria Paolo. Although Subject-Verb-Object is a statistically prevalent construction, all variations in word order can occur inside a component, and they may depend on the particular words which are used. 2. In ATNSYS (Cappelli, Ferrari, Moretti, Prodanof and Stock, 1978), a previously constructed ATN based system (Woods, 1970), a special dynamic reordering mechanism was introduced in order to get sooner to a correct syntactic analysis, when parsing sentences of a coherent text (Ferrari and Stock, 1980). Besides psycholinguistic motivations, the main reason for the introduction such heuristics lay in the large number of alternative arcs that has to be introduced in networks for parsing Italian sentences. As a matter of fact, ATN's were not originally conceived for flexible word order languages. (In the extreme free word order case, an ATN would have one single node and a large number of looping arcs, losing its meaningfulness). Work has been done on ATN parsers for the parsing of non-grammatical or extragrammatical sentences in English, a problem related to our one. For instance Weischedel and Black (1981) have proposed a system of information passing in the case of parsing failure. Kwasny and Sondheimer (1981) have suggested the relaxation of constraints on the arcs under certain circumstances. Nevertheless, these problems, together with that of treating idiosyncratic phenomena related to words and flexible idioms, are not easy to solve within the ATN approach. At least two other parsers should be mentioned here.