Resource-logical Event Semantics for LFG (Draft)

In this paper I implement the core of an event semantics for LFG using the resource-logic semantic framework of Dalrymple et al. (1997). The interaction of event variables with tense, aspect, adverbs, and quantified NPs is explored. The particular analysis given here is motivated by certain problems and results from categorial grammar treatments. 1 Event semantics in categorial grammar Following Davidson (1967), a number of ‘neo-Davidsonian’ semanticists (e.g. Parsons 1990) have concluded that the interpretation of the sentence Brutus stabbed Caesar violently should be something like (1), which asserts a violent stabbing event e and specifies e’s thematic agent and patient. (1) ∃e[stabbing(e) ∧ violent(e) ∧ agent(e,Brutus) ∧ patient(e,Caesar)] Actually implementing an event semantic system to compositionally derive logical forms like (1) poses certain challenges, especially within the categorial (or ‘type-logical’) grammatical tradition of Montague. First, the traditional semantic types of verbs (and modifiers) must be altered in order to accommodate the introduction of event variables. By way of example, Table 1 lists four representative approaches to incorporating an event variable into the interpretation of the transitive verb stab from example (1). Then, once the types of verbs are altered, these changes must be propagated onto other categories which depend on verbs, such as the scope of generalized quantifiers. The second challenge, which we might call the “random access” problem (following Francez and Pratt (1997)), arises because the sequence of β-reductions on a complex lambda term is strictly fixed. Event modifiers like violently and arguments like Caesar must be applied to the verb in a specific order, which may or may not match the phrasal structure of the sentence. For example, violently cannot directly reach into a predicate like λxλyλe.φ to modify the event variable e inside φ. As a result, categorial treatments are forced to adopt strategies like flexible categories, typeshifting, or special semantic combinator functions to get LFs like (1) to come out correctly. In order to clarify the issues raised above, I will spend the remainder of this section on a brief review each of the four implementations represented in Table 1.1 The conclusions drawn from the examination of these systems will be used to motivate the particular LFG analysis developed later in this paper. Note that these four analyses are taken out of context and are not themselves the main point of their respective papers. Therefore this section should be taken as an explication of certain general principles, not as a criticism of these particular papers or approaches.