DFKI-LT - Derivation Without Lexical Rules

Hans-Ulrich Krieger
Derivation Without Lexical Rules
1 volume 93-27,
DFKI Research Reports, Pages 33, DFKI, Saarbrücken, 6/1993

In Krieger and Nerbonne (1992) we showed how to get rid of LEXICAL RULES for DERIVATION, as they are explicated by Pollard and Sag (1987) in HPSG I, Ch. 8.2. We proposed a treatment of derivation not by means of traditional lexical rules but instead in terms of PRINCIPLES, RULES, and LEXICAL ENTRIES entirely in the spirit of HPSG, together with unification-based inheritance of a very sophisticated kind. One major disadvantage of this approach was the employment of complex functions in certain principles. In this paper I first extend the old approach and then show how to eliminate these functional dependencies in the domain of derivational morphology by going back to simpler ones like cons, first, and rest. But this simplification is only achieved if we assume more complex feature structures than the ones described in Krieger and Nerbonne (e.g., by introducing two different SUBCAT features) and by proposing modified versions of the old Constituent Order Principle and the Subcategorization Principle for morphology. In addition, I postulate a hierarchy of affixes which is cross-classified, for instance, according to the effects these affixes contribute to the subcategorization information of a compound word. The structure of the paper is as follows. We start with a very short introduction about the residence of word-formation rules in modern feature-based theories. After that we present our approach to derivational morphology which is distinguished in that it gives up the notion of lexical rule as a single entity (operator). We describe the structure of affixes and words (e.g., which attributes are appropriate?) and introduce the relevant principles and the rule schema of our approach to derivational morphology. The section shows how to reduce functional dependencies to a minimum at the cost of the size of our feature structures. We also present a technique which allows us to state relational dependencies as they are called by HPSG in a functional manner. In the next section we show how the whole treatment works by applying it to tough phenomena from prefixation and suffixation. The section presents many examples, which might serve as a how to guide to a practitioner. After that we explain the idea which will lead us to the affix hierarchy. We will see that the affix hierarchy is inspired by the work of HPSG on structured lexicons (i.e., by the hierarchy of lexical types). A lot of examples will again be given throughout this section. We finish the paper by summarizing our approach and by saying a few words about the topics which we will tackle next.
Files: BibTeX, Krieger:1993:DLR.pdf, RR-93-27.pdf