Parsing Deficiencies of the Pc-kimmo System

In this paper, we discuss the possibilities and limitations of the PC-KIMMO system as a recognition device of compound formations in a language like Modern Greek, where compounding interacts with derivation, inflection and lexical phonology. We deal with the computational processing of nominal and verbal compounds and try to show certain limitations of the PCKIMMO software with respect to the principles of compound formation. Compounds are parsed into their structural constituents that are morphemes (i.e. stems and affixes) or words, depending on the case. Stress is also taken into consideration since compounds display peculiar stress properties which are different from other word-stress properties. In particular, we show that stress and syllabification that are crucial for the analysis of such constructions cannot be dealt with in a satisfactory way. 1 Morpho-phonological parsing with PC-KIMMO v.2 PC-KIMMO is a morphological parser based on the model of two-level morphology ([10], [11]). The model distinguishes between the word’s morphotactics that specify its morpheme constituents in the particular order into which they occur, and the word’s morphophonemics which account for the different orthographic forms of the morphemes. In its original conception ([10], [11]), the two-level model segments the word in its constituent parts, and accounts for word-internal phonology and orthography by means of declarative two-level rules expressing correspondences that hold between a lexical and a surface form. These two-level rules apply in parallel, and do not allow any intermediate levels of representation. Because of their relational character (i.e., they represent correspondences between surface and lexical forms) they are bi-directional. Two-level rules are implemented as finite state transducers. A finite state transducer (FST) functions like a finite state automaton but it operates on two input strings. The label on the arc of an FST consists of a valid correspondence pair of symbols of the two input strings1. The lexicon incorporating the morphotactics consists of a list of morphemes. Each lexical entry is characterized by its grammatical category, its morpho-syntactic features, a gloss (additional information), and an alternation index specifying the list of alternative morphemes that may be combined with it. Lexical entries are generally grouped into sublexica, depending on their grammatical category. (1) lists the sublexica used for Greek: (1) N (noun), V (verb), ADJ (adjective), DET (determiner), P (preposition), PR (pronoun), ADV (non inflected adverb), CONJ (conjunction), IJ (interjection), PART (particle), CLITIC, ADI (inflected adverb), PRI (inflected pronoun), DAF (derivational suffix), PREFIX, SUFFIX, INFL (inflectional ending). An example of a lexical entry of the sublexicon of nouns is given in (2): (2) άνθρωπ[anθrop] «man»