| "text": "matter to decide upon when setting up a twolevel description is the following: which inflectional phenomena are to be treated in the lexicon system. anp which ones arc-best described by rules? In order to get a survey of the different types of inflection al phenomena that occur In French, the following classification Is of help: . deviation from regular inflection taking place only for some words with a certain structure The rule component is designed for those Inflectional phenome na that are productive, frequent and do not affect long sequences of signs (6). The stem lexicon together with mini lexicons for affixes account for regular inflection, as well as'for those phe nomena that deviate from regular Inflection, but do not depend on the phonological or morphological context. The stem lexicon can a !sc De used to l^st irregular word ^orrs Category a), regular inflection, should thus be described in the lexicon component, whereas category b), context-dependent deviation from regular inflection, is best described by rules. The lexicon component is best suited to take care of those phe nomena that belong to category c), deviations from regular inflection that are not context-dependent. I have used 15 lexicons of endings to describe the nominal pi i catr 1ce If the number of words with a certain \"variation irreguliere\" is > 1, I have always decided to set up a lexicon insteac of enu merating inflected forms in the stem lexicon. In order to handle category b), \"variation r6gul1ere\", there are six rules. I show them here in a simplified form. tenses and nraods differ in most cases between tHP conjugations. anc no element always signalling for example person or number exists. The ending -rai. for instance, has as its signifi\u00e9 futur, 1st person, singular, and -irent pass\u00e9 sim ple, 3rd person, plural. Thus it is obvious that the number of lexicons for endings becomes large. By keeping the endings as constant as possible and, when necessary, using more than one stem, a good descriptional economy is achieved (7). A regular verb, such as par1er. gets one entry in the stem lexicon, while an irregular verb, vivre for Instance, gets more than one. Excep tions exist, though. So far I have encountered one Irregular verb that needs only one entry, courir: each Stem Is attached a reference to a continuation class etc. , , above). The continuation class then points to the lexicons containing the endings possible to add to the stem. Vl, for example, points to 15 lexicons. There are 62 continuation classes, 57 of which are needed for the description of the Irreg ular verbs. The lexicons I use are: within the two-level framework are word-internal. The two-level model analyses the words as Isolat ed uhlts, and gives all the possible analyses. In this respect It works very well for French, whose Inflectional phenomena are all accounted for in a satisfactory way. with one small excep tion. A handful of nouns, such as arc-en-c1el get their plural ^ after the first nouh; arcs-en-clel. These plurals have to be listed in the stem lexicon. Homographies are very frequent in French, a lahguage that can be characterized neither as analytic nor as synthetic. The homo graphs can be divided into two groups: The homographles of both types are disambiguated by the syntagmatlc relations of the sentence. A good deal of the informa tional contents Is thus to be found in the relations between the words. Since the two-level model only is concerned with the iso lated sequences of signs constituting words, the number of words with more than one analysis gets very high for French. This Is not the case for languages such as Finnish, whose morphological elements carry more information. Often the analysis resulting from the two-level model Is quite satisfactory, but for the purpose of vocabulary studies the defi nition of the word as a sequence of signs separated by blanks is not adequate for a language like French. It becomes necessary to push the analysis further and disambiguate the homographles. For the same purpose, discontinuous signs have to be accounted for. In French they occur for example among the verbs (a1/ be identified to achieve an analysis, and how are these segments identified? These are interesting questions to which I will try to find some answers in my future research. I will also investigate how the two-level model is to be combined with a subsequent syntactic analysis. In orde'' to test the two-level description, ] have a corpus consisting of 102 signed editorials from the daily French newspa per Le Figaro. Together they constitute 50012 running words. At present there are in the stem lexicon about 2030 nouns, proper names, adjectives and abbreviations, and about 1400 verbs." |
