Effect of referential information on the resolution of multiple NP head attachment ambiguities

Yuki Kamide, Gerry T.M. Altmann, and Emma Stirling
y.kamide@psych.york.ac.uk, g.altmann@psych.york.ac.uk
Department of Psychology, University of York

The present study explores the question of whether referential information plays a significant role in determining attachment preferences for relative clause attachment ambiguities with multiple NP heads (e.g., "the servant of the actress who was on the balcony"; Cuetos & Mitchell, 1988)

Amongst numerous previous research, Gilboy et al.'s (1995) questionnaire study compared attachment biases for the ambiguity with different patterns of determiners for the NP heads, for instance, "the servant of the actress who" (the-the), "a servant of the actress who" (a-the), and "the servant of an actress who" (the-a), to investigate the effects of definiteness on the resolution of NP attachment ambiguities within the framework of referentially-based theories (e.g., Crain & Steedman, 1985; Spivey-Knowlton & Sedivy, 1995; Frazier & Clifton, 1996). Their overall result with English suggested that the definiteness of the potential attachment hosts does not have a statistically reliable effect on actual attachment preferences (NP2 preference: the-the, 55%; a-the, 59%; the-a, 54%; planned comparisons not reported). Our first (questionnaire) experiment revisited the issue of definiteness, adding an extra condition (e.g., "a servant of an actress who"; a-a). The following NP2 preferences were obtained in each condition: the-the, 65%; a-the, 70%; the-a, 58%; a-a, 65% (overall NP2 preference: 64.6%). An ANOVA showed that the main effects of Definiteness of NP1 and Definiteness of NP2 were highly significant. There was no interaction between the two. Planned comparisons further confirmed statistical reliability of all the above numerical differences across the conditions. Thus, the overall finding suggests that the definiteness of potential attachment sites has an effect on attachment preference (cf. Gilboy et al., 1995), and that the definiteness of both preferred (NP2) and dispreferred (NP1) sites is taken into account during ambiguity resolution, indicating that the eventually dispreferred attachment site does initially compete with the preferred site.

A second questionnaire study was conducted to investigate this same issue, using relative clause attachment ambiguities involving 3 NP sites (e.g., "the lamp near the painting of the house that was damaged in the flood"; Gibson et al., 1996). Five conditions were tested (the-the-the, the-a-a, a-the-a, a-a-the, a-a-a). Overall, NP3 attachment was preferred significantly more often than the other options (overall preference: NP1: 17.6%; NP2: 18.6%; NP3: 64.4%). Further planned comparisons revealed that the definiteness of both NP1 and NP3 modulates NP3 preferences (the-a-a (59%) vs. a-a-a (65%); a-a-the (69%) vs. a-a-a (65%)), whist that of NP2 does not (a-the-a (65%) vs. a-a-a (65%)). Therefore, this pattern of the data seems to indicate that the definite determiner can have an effect on the most preferred attachment choice only when its head is supported by other underlining factors (relativized relevance for NP1, recency for NP3; Gibson et al., 1996), and also that such hosts (NP1 & NP3) can enter the competition whereas other hosts without external support (NP2) may not.


AMLaP Conference, Saarbrücken, September 2001