Past Event

On the semantic underpinnings of complement selection

Paul Melchin
(Carleton University)

In this talk I present work from my thesis, in which I argued that selectional restrictions of predicates need not be arbitrarily specified, but instead are predictable based on certain aspects of the predicate’s semantics. To do this, two things must be shown. First, it must be shown that the properties being selected for are semantic properties of the complement, rather than syntactic features or categories. Second, the selectional restrictions of a given predicate must be shown to be predictable from the meaning of the selecting predicate. I focus on each of these in turn.

I argue first that the elements that get selected for by predicates are semantic properties of the selected  argument, rather than syntactic properties. That is, apparent selection for clausal categories (CP or TP) is in fact selection for propositional entities (questions, assertions, facts, etc.); selection for bare verb phrases (vP) is actually selection for eventualities (events or states); and selection for nominal (DP) is actually selection for objects/things. Each of these semantic categories corresponds to a syntactic phase, and selection is restricted to the highest phase of the complement. This allows for a semantic account of the properties that get selected for, while explaining certain selectional asymmetries that are otherwise problematic (Bruening 2009; Bruening et al 2018).

Next I move to the selectional restrictions themselves, examining two selectional phenomena that have previously evaded principled semantic explanation. First I examine the well-known difference between eat and devour: eat allows omission of its object, while devour does not (John ate/*devoured yesterday). I show that this is linked to the aspectual properties of the verbs: devour denotes an event where the object necessarily undergoes a complete scalar change, which means the complement must be syntactically realized (Rappaport Hovav and Levin 2001; Rappaport Hovav 2008). Eat does not entail a complete change of state, and so the complement is optional. I show that this holds for a wider group of verbs as well. Thus, the c-selectional properties of these verbs need not be specified in the lexicon.

Finally I discuss the phenomenon commonly known as l-selection, in which verbs require a complement headed by a specific lexical item; e.g., rely requires a complement headed by on. I show, contrary to common assumptions, that there are regularities in both the verbs and the prepositions involved in this phenomenon, such that the verb-prepositions are (at least partially) predictable. Furthermore, I show evidence that l-selection is an instance of complex predicate formation, and showing that a certain kind of complex predicate (exemplified by l-selecting predicates like rely on) is possible in satellite-framed languages but not in verb-framed languages. All of this suggests that l-selection has semantic correlates, and need not be stipulated in the lexicon.