Lexical relatedness and the lexical entry: A formal unification

Based on the notion of a lexicon with default inheritance, I address the problem of how to provide atemplate for lexical representations that allows us to capture the relatedness between inflectedword forms and canonically derived lexemes within a broadly realizational-inferential model ofmorphology. To achieve this we need to be able to represent a whole host of intermediate types oflexical relatedness that are much less frequently discussed in the literature. These includetranspositions such as deverbal participles, in which a word's morphosyntactic class changes(e.g. verb ⇒ adjective) but no semantic predicate is added to the semantic representation and thederived word remains, in an important sense, a "form" of the base lexeme (e.g. the 'presentparticiple form of the verb'). I propose a model in which morphological properties are inherited bydefault from syntactic properties and syntactic properties are inherited from semantic properties,such as ontological category (the Default Cascade). Relatedness is defined in terms of a GeneralizedParadigm Function (perhaps in reality a relation), a generalization of the Paradigm Function ofParadigm Function Morphology (Stump 2001). The GPF has four components which deliver respectivelyspecifications of a morphological form, syntactic properties, semantic representation and a lexemicindex (LI) unique to each individuated lexeme in the lexicon. In principle, therefore, the samefunction delivers derived lexemes as inflected forms. In order to ensure that a newly derived lexemeof a distinct word class can be inflected I assume two additional principles. First, I assume anInflectional Specifiability Principle, which states that the form component of the GPF (whichdefines inflected word forms of a lexeme) is dependent on the specification of the lexeme'smorpholexical signature, a declaration of the properties that the lexeme is obliged to inflect for(defined by default on the basis of morpholexical class). I then propose a Category ErasurePrinciple, which states that 'lower' attributes are erased when the GPF introduces a non-trivialchange to a 'higher' attribute (e.g. a change to the semantic representation entails erasure ofsyntactic and morphological information). The required information is then provided by the DefaultCascade, unless overridden by specific declarations in the GPF. I show how this model can accountfor a variety of intermediate types of relatedness which cannot easily be treated as eitherinflection or derivation, and conclude with a detailed illustration of how the system applies to aparticularly interesting type of transposition in the Samoyedic language Sel'kup, in which a noun istransposed to a similitudinal adjective whose form is in paradigmatic opposition to case-marked nounforms, and which is therefore a kind of inflection.