Conceptual combination during sentence comprehension

Abstract:
We examines how words representing simple concepts are combined to form more complex concepts during ongoing language comprehension. Consider the phrase jello knees, as in the sentence: The retired singer began a series of lectures on how to avoid jello knees during public performances. Any standard feature bundle describing the meaning of knees will not have "jello" as a normal property. And in fact, the combination of jello with knees in this case leads to the 'emergent' property of "nervousness." What is the process by which this takes place?

Previous work has suggested that determining the meaning of such a phrase is a two stage, serial process (Smith, Osherson, Rips, & Keane, 1988). In the first stage individual 'standard' word properties are activated, following which the compound (emergent) meaning of the phrase is determined by 'feature combination'. Alternatively, Hampton and Springer (1989) and Springer and Murphy (1992) have provided data which they take to demonstrate that the 'emergent' meaning of a adjective-noun phrase is computed before the 'standard' properties of the noun are made available (the adjective effectively changes the properties accessed for the noun). However, all of the previous work in this area has used off-line methodologies (post-phrase or post-sentence judgments), which, in principle are not capable of making the fine grained temporal distinctions required to differentiate and/or examine these sorts of claims.

The present study employed an on-line, cross-modal priming paradigm in which auditory sentences containing compounds such as jello knees were presented to 80 participants. Activation of the 'standard' meaning of the head noun (here, "knees") and the emergent properties of the Adj-N compound (here, "nervous") was determined by presentation of probe words related to each of these meanings at four different points in processing: 1) a baseline; 2) during presentation of the noun; 3) during presentation of the word following the noun/compound; 4) during presentation of the second word following the noun/compound. For example (probe points are indicated by numbers on key words):

The retired singer held a series of lectures on how ₁ to avoid jello ₂ knees ₃ during ₄ public performances.

The results, as seen in Table 1, demonstrate that the 'standard' noun meaning is accessed earlier than the 'emergent' phrase meaning: (priming difference (related minus control probe) scores are given here for the sake of convenience; actual means for all probes will be provided in the poster).

Priming (related vs. control differences) was significant for the 'emergent' phrase properties only at probe point three (after the compound), F(1,83) = 8.531, p=0.004, and for the noun only at probe point 2 (after processing the noun), F(1,83) = 6.156, p=0.015. For the 'standard' noun meanings, there was a strong trend for an interaction in priming/activation level between probe points 1 and 2 , F(1,160) = 3.650, p=0.058 and for the 'emergent' phrase meanings, there was a significant interaction in priming level between probe points 2 and 3, F(1,83) = 8.531, p=0.004. These interactions demonstrate that the noun meaning has been made active and available to the processor while the noun is heard (at probe point 2) and that the 'emergent' phrase meaning is generated only once the entire phrase has been heard (probe point 3).

These temporally fine-grain results support a serial model in which literal, standard word meanings are accessed first, and are only then composed into phrasal concepts. These results also clearly argue against Springer and Murphy (1992), in which phrasal properties are accessed earlier than the properties of individual words.