(36) World-knowledge and frequency in resolving number ambiguities

Poster session 1
Monday, September 5, 17:30
Markus Bader & Jana Häussler
University of Konstanz
markusdotbaderatuni-konstanzdotde
In German, number-ambiguous NPs like ''Peters Lehrer'' ('Peter's teacher/teachers') are quite common. A prior experiment using speeded-grammaticality judgments with sentences like (1) has shown that the preference for either assigning singular or plural correlates substantially with corpus frequencies. The more often a noun occurs with number x, the more often it is analyzed as having number x during first-pass parsing, leading to difficulties when encountering a verb specified for the opposite number. This paper will discuss two questions: (i) Are the first-pass preferences due to frequency, or is there an alternative explanation in terms of world-knowledge? (ii) Can the results be replicated with a more on-line method like selfpaced reading?

To adress the first question, we asked participants for each ambiguous noun to imagine a typical situation, and to write down the number of referents involved. The resulting numbers were coded as either singular (number of referents = 1) or plural (> 1). The resulting world-knowledge data and the original frequency data correlate strongly (r = .78). The original speeded-grammaticality judgment data correlate well with both measures (mean r = .43).

We then conducted two word-by-word non-cumulative selfpaced-reading experiments with the disambiguating finite verb either in sentence-final position (as in the SGJ-study, cf. (1)), or followed by three words (cf. (2)). For sentences like (1), reading times on the auxiliary correlate substantially with the frequency and world-knowledge data (mean r = .46). For sentences like (2), we found only marginal correlations in the disambiguating region, both for word-wise and region-wise reading times.

Our findings allow two main conclusions. First, one has to be cautious with conclusions concerning frequency and first-pass preferences, as there might be additional factors underlying both frequency and preferences. Second, for processes triggered by the clause-final verb, speeded-grammaticality judgments and self-paced reading give comparable results. Furthermore, despite measuring at the sentence wrap-up position, having the disambiguating verb in clause-final position leads to sharper results than having it some words before the end of the clause. We hypothesize that this is due to verb-related processes being an inherent part of sentence-wrap-up in verb-final constructions.