Nirenburg and Raskin [88] describe the main design issues or choices faced in lexical semantics. These are shown in Figure 2. (As part of this discussion, they provide an in-depth criticism of the generative lexicon approach and compare it with sense-enumeration approaches. See below.) Several of these should be fairly self-evident and uncontroversial. The main ones will be discussed briefly here. The first issue concerns how lexical entries are derived. Extreme viewpoints would be deriving all related senses from a single sense (generative lexicon) versus listing out all the senses (sense enumeration).
The second issue concerns the extent to which syntax can be used to determine semantics, as in [73]. The fourth issue concerns the underlying theory on how word meanings are constructed. In some cases, there is an over-reliance on formal theories without clearly indicating how they are used in practice. The fifth issue concerns when rules are worth pursuing. For instance, the grinding rule [17], which is discussed below, is criticized as being too special purpose. The eighth issue concerns the trade-off between in-depth analysis of selected topics versus achieving broad descriptive coverage of the language, a stance favored by Cruse [31]. Finally, the last issue is a metaphoric description of the opposing camps in the economics of the theory of lexical semantics: those ``supplying'' theories for consumption without concern for consumer systems versus those being attuned to the ``demands'' of the consumer systems by just providing what is needed.
The collection of papers in [104] provides a good coverage of the main issues in computational semantics with respect to lexical semantics. These issues include the organization of lexical entries, the role of lexical functions and relations, and the role of inheritance in the lexicon [105]. One case of interest not covered later concerns the role of lexical functions, as in Mel'cuk's meaning text theory.
Heylen [54] gives an overview of lexical functions in the meaning text theory and shows the relation to deep cases and to the generative lexicon theory of Pustejovsky [96] and others. Mel'cuk's Explanatory Combinatorial Dictionary (ECD) can be viewed as a semantic network where the nodes are lexemes and the links are either lexical functions giving related collocations or argument positions for predicates in the meaning decomposition for a lexeme. Heylen shows that this forms a basis for an explicit lexical knowledge base by showing how the components of the generative lexicon can be derived from the ECD. Another interesting case is the formalization of thematic roles by Bés and Lecomte [11], using the Montague Semantics framework. Optional thematic functions (e.g., cause) are handled by omitting reference to these functions in the internal representation when their arguments haven't been instantiated via associations with positions in the syntactic frame.
In addition, semantic features (e.g., [+animate]) can influence the interpretation via extensions to the type coercion mechanism defined by Pustejovsky [96] for the Generative Lexicon.
