Ana Arregui
University of Massachusetts at Amherst
I argue that quantificational NPs (QNPs) trigger semantic predictions (distinct from syntactic predictions) that help guide the construction of a sentence's structure. I report a self-paced reading study supporting (1):
(1) Quantifier Prediction Principle: When the processor comes across a QNP, it makes a prediction that some constituent denoting a set of individuals will function as the argument of the QNP.
Referential NPs (type e) denote individuals. QNPs (type <<e,t>,t>) denote functions from sets of individuals to truth-values. QNPs can take arguments, and, according to (1), the processor predicts such arguments. This is illustrated in (2a) for "Every child smiled". Q* stands in for the predicted argument. The prediction is cashed out when the argument is processed, as in (2b).
(2a) S: Ux [child(x)-->Q*(x)] (2b) S: Ux[child(x)-->smiled(x)]
/ / \
NP: every child NP :every child VP:smiled
/\ /\ /\
LQ Ux [child(x)--> Q(x)] LQ Ux[child(x)-->Q(x)] Lx [smiled(x)]
Code: L = lambda (the functor corresponding to lambda abstraction)
U = Universal Quantifier
To test (1), semantic predictions must be differentiated from syntactic VP-predictions. [1]/[2] observed that subjects do not detect missing VPs in multiple center embedding constructions [MCECs] ((3b) is judged better than (3a)). The VP prediction corresponding to the second NP seems missing. I compared the effects of QNPs in different positions in MCECs with missing VPs. Forty-eight participants read sentences like (4) using a phrase-by-phrase self-paced moving window technique. Since there is no syntactic VP prediction corresponding to the second NP, (1) predicts that a missing VP will be noted when the second NP is a QNP. The semantic prediction triggered by the QNP will remain unsatisfied. This anomaly will slow down reading times. No such prediction is made if the second NP is a referential NP. The experimental results confirmed this hypothesis and support (1). Mean reading times for regions [d] and [e] were faster for (4c) than for (4b) [1935.479 vs. 2147.542 and 1501.265 vs. 1732.563]. The differences were almost significant by subjects (p = .066 and .052), and fully significant by items (p = .017 and .022). Reading times for (4a) were as slow as those for (4b), perhaps due to similarity-based interference in working memory ([3]). To conclude, I will discuss further implications of the hypothesis that there are semantic predictions, and I will compare my analysis with [4], who analyze the effects of QNPs in terms of a metric for syntactic complexity.
| (3) | a. | The patient who the nurse who the clinic hired admitted met the surgeon. |
| b. | The patient who the nurse who the clinic hired met the surgeon. | |
| (4) | a. | [a] The critic | [b] who the artist | [c] who the gallery was promoting | [d] made unpleasant remarks during the opening, | [e] and several people complained. |
| b. | [a] The critic | [b] who every artist | [c] who the gallery was promoting | [d] made unpleasant remarks during the opening, | [e] and several people complained. | |
| c. | [a] Every critic | [b] who the artist | [c] who the gallery was promoting | [d] made unpleasant remarks during the opening, | [e] and several people complained. |
References
[1] Frazier, Lyn, 1985. Syntactic Complexity, in Natural Language Parsing, 129-189.
[2] Gibson, E., & J. Thomas, 1999. Memory Limitations and Structural Forgetting: LCP 14(3):225-48.
[3] Lewis, R, 2000. Specifying Architectures for Language Processing, in AMLAP, 56-89
[4] Warren, T., & E. Gibson, 2001, Ms. The influence of referential processing on sentence complexity