In Support of Incremental Semantics During Reading: Evidence from Phrasal Verbs

Martin Corley
Martin.Corley@ed.ac.uk
Department of Psychology and Human Communication Research Centre
University of Edinburgh

An abundance of evidence exists that when humans read sentences, they attempt to assign structure incrementally, i.e., as new words are encountered. However, there is less direct support for the notion that meaning is incrementally assessed. In this paper, we address the question of meaning via the "compositional" semantics of (seperable) phrasal verbs, such as 'calm ... down' in (1a).

(1)

a.

The fireman calmed the pretty young woman down after he rescued her.

b.

The fireman chatted the pretty young woman up after he rescued her.

c.

The fireman dropped the pretty young woman off after he rescued her.

According to Frazier (1995; see Frazier, d'Arcais, and Coolen, 1993), phrasal verbs, much like seperable-prefix verbs in languages such as Dutch, have their meanings represented by discrete "Morphological Integration" (MI) nodes (Schreuder, 1990). Encountering the first lexical constituent of a phrasal verb ('calmed' in (1a)) should partially activate the MI node for 'calm down', as well as activating other associated nodes. Once the particle ('down') is encountered, the composite meaning is fully activated. On a strong interpretation of this hypothesis, 'calm down' becomes a partial synonym of 'calm' (albeit one requiring further evidence), which should be (partially) available as soon as 'calm' is encountered. There should be little or no difficulty reading subsequent material as long as it is likely later to be licensed by a particle. Similarly, there should be little difficulty in reading a particle, since the corresponding MI will already be partially activated.

We tested these hypotheses using examples like (1) in a self-paced reading experiment. In (1b), the object NP following the head of the phrasal verb ('chatted') is not licensed by the head in isolation. However, as long as the 'chat up' MI node is partially active, participants should experience little or no difficulty in reading the object ('the pretty young woman'). In (1c), the particle 'off' substantially changes the meaning of the head ('dropped') relative to the control ('calmed ... down'). However, if 'dropped off' and 'calmed down' are interpreted via activation of MIs, there should be no difference in reading patterns at the particles.

The results showed clear evidence of incremental interpretation: subjects read the object NP slower in (1b) than in (1a) or (1c), and the particle in (1c) was read more slowly than that in (1a). We suggest that these findings are incompatible with (some versions of) an MI hypothesis, suggesting instead that the semantic interpretation of a sentence is made incrementally on the basis of the evidence encountered so far.