Lexical signs

What are signs, in linguistic terms? Do signs consist of other signs, in the way that sentences like Let's listen to Charlie Byrd! have constituents, or compound words like mousetrap repair shop owner are made up of other words? Or is the quality of being a sign rather a holistic one which only attaches to utterances or even dialogues in context, from `Hi' to the entire proceedings of a business meeting? The present approach to the theory of word formation (the ILEX approach) encompasses the following assumptions about signs:

1. All signs are pairs of some observable form and a meaning.
2. All signs are compositional in principle, down to their smallest phonological constituents.
3. Every language user is familiar with an inventory of more-or-less fixed signs, a lexicon, as well as with non-lexical, freely constructed signs.
4. Lexical signs are assigned to a scale of well-defined ranks, corresponding to linguistic levels of description from phoneme-size through morphemes, simple, derived and compound words, phrases and proverbs to ritualised exchanges, in an idiomaticity hierarchy; the word is a basic rank.
5. At each rank, linguistically significant generalisations are formulated in terms of inheritance relations for sets of inventorised lexical items at this rank: phonology (better, prosody) is the set of generalisations about speech sounds, morphology the set of generalisations about form and meaning of words, syntax the set of generalisations about form and meaning of phrasal idioms, and so on.
6. At any given rank, a sign has, in principle, four properties: its surface (physical appearance, e.g. the forms represented by the transcription , or the spelling rattlesnake), its meaning (its relation to the situation of use, including objects it refers to, speaker and addressee), its category (its co-occurrence with other signs in linguistic structures), and its parts (its internal structure or `child' constituents, which are in general weighted in terms of head and modifier constituents).
7. The surface and the meaning of a sign are its interpretative properties, and the category and parts are its compositional properties.

There are interesting special cases. For example, the traditional phoneme is an inventorised item with no parts, no semantic interpretation and purely structural `meaning'; the morph cran in cranberry has no parts at the same rank (morphology), and no semantic interpretation (except in Norfolk, where it means `a basket of the type freshly caught herring are kept in'). Leprechaun items such as `zero morphemes' and `traces', for those who believe in them, have no parts and no phonetic interpretation, but a category and a semantic interpretation.

Recent work in syntax, notably within the paradigm of Head-driven Phrase Structure Grammar, HPSG (cf. [Pollard & Sag 1987], [Pollard & Sag 1994]), has revived a similar structuralist notion of sign to that outlined here, and formalised it as an attribute-value matrix (AVM). In this approach, a taxonomy (type hierarchy) of sign types is defined, from the most general type sign to the most specific types, individual words; each sign type is characterised by a set of appropriate attributes and appropriate values, and generalisations over more specific sign types are expressed by inheriting the properties of more general sign types along the branches of the sign taxonomy.

It is not yet clear how to integrate lexical problem areas into the word and sentence oriented (albeit lexicalistic) HPSG approach. The HPSG model contains three relevant kinds of entity: a base inventory of words, lexical rules of inflection, word-formation, subcategorisation and semantic selection which define an extended inventory of words, and principles of composition linking the `head-daughter' (head part) and the `complement-daughters' (modifier parts) of a phrase by concatenation and unification or other appropriate operation. Problem areas for this model currently still include the following:

1. Idioms, which are clearly lexical signs, but not elementary ones.
2. Sentence prosody and word prosody, which are involved in compositionality, but by complex varieties of prosodic association and not just by concatenation.
3. Compositional principles for the morphology of inflection, derivation and compounding, including compositionality in morphophonology and morphographemics.
4. Degrees of irregularity in the lexicon.
5. Degrees of compositionality in syntax and morphotactics.
6. Markedness relations based on neutralisation or familiarity.
7. Compositional lexical semantics (hard, if lexical items have no parts).

The present study addresses these problems and proposes an integrated, sign-based solution to lexical explanations. In the following sections, an HPSG-related theoretical framework and an operational DATR model for this theory are used to describe English compounds: linguistic concepts closely related to HPSG are described and implemented with representation techniques from DATR. After a summary of the main directions in Inheritance Lexicon Theory, modelling conventions for the inheritance lexicon are characterised, a summary of lexical properties of the main types of English noun, in particular noun compounds, is given, followed by an account of the DATR lexical knowledge representation formalism. An operational DATR model for English nouns is discussed, and a sample analysis is presented. The main results and conclusions are outlined in the final section.