Identifying Topic and Focus by an Automatic Procedure

An algorithm for automatic identification of topic and focus of the sentence is presented, based on dependency syntax and using written input, which is much more ambiguous than spoken utterance. 1. The dichotomy of topic and focus, based, in the Praguean Functional Generative Description, on the scale of communicative dynamism (underlying word order), is relevant not only for a possible placement of the sentence in a context, but also for its semantic interpretation. The underlying word order differs from • the surface one especially in that the verb stands m o r e t o the right than all its complementations belonging to the topic of the sentence (or to the local topic of the clause headed by the verb), and more to the left than those belonging to the focus. Using a dependency grammar (or, more or less equivalently, a flat structure in a constituency based grammar), we can illustrate this by the following example, where (1') is a simplified underlying representation of (1) on a reading answering e.g. the question Where has Charles found my pen ?: (1) Charles has found your pen in a box lying on the table. (1') (Charles)Act ((you)App,a pen)Obj find.Pelf Ceox.Indef ((Rel)Act lie (table)L~.o,)c~o, )L~.~, In (1') every pair of parentheses encompasses a dependent item (i.e. corresponds to an edge of the linearized dependency tree), the indices of parentheses denote kinds of dependency (valency slots, or theta roles and adjuncts): Act stands for Actor (underlying Subject), Appurt for Appurtenance (Possessivity in a broader sense), Obj for Objective (underlying ~Object), Loc for Locative, Gener for the General Relationship (of an adjunct to its head); the other indices denote values of morphological categories (Perfect , Indefiniteness) and of adverbial prepositions (in, on), Rel denotes a relative pronoun (here