Machine Translation Based On Logically Isomorphic Montague Grammars

Usually two approaches to machine translation are distinguished: the interlingual approach and the transfer approach (cf. Hutchins [i]). In the interlingual approach translation is a two-stage process: from source language to interlingua and from interlingua to target language. In the transfer approach there are three stages: source language analysiS, transfer and target language generation. The approach advanced in this paper is a variant of the interlingual one. It requires that 'logically isomorphic grammars' are written for the languagesunder consideration. The syntactic rules of these grammars must correspond with logical operations, in accordance withthe compositionality principle of Mentague grammar. Moreover, the grammars must be attuned to each other as follows: if one grammar contains a rule corresponding with a particular logical operation, the other grammars must contain rules corresponding with the same operation. Syntactically, these rules may differ considerably. If the grammars are attuned to each other in this way, 'logical derivation trees', representations of both the syntactical and the logical structure of sentences, can be used as intermediate expressions. The paper is organized as follows. In section 2 the relevant concepts of Montague grammar and the notion 'logically isomorphic grammars' are introduced. In section 3 a version of Montague grammar is described, called M-grammar, which is more suitable for computational use than Montague's original proposals. The property of logical isomorphy is then defined for M-grammars. In section 4 the design of the Rosetta translation system, based on this approach, is outlined, followed by a brief discussion in section 5.