Title: Individual differences in the resolution of scope ambiguity in supportive contexts

Authors: Ted Gibson (MIT), Andrea Gualmini (Utrecht), Ev Fedorenko (MIT)

Abstract:

Research in the sentence processing literature has established that contextual information (among other information sources) can be effectively used to resolve temporary and global ambiguity (e.g.,Tanenhaus et al., 1995).  In this project, we investigate the processing of scope ambiguities, such as in (1):

(1) Every tester didn't win the game.

(1) has a surface-scope interpretation (none of the testers won the game) and an inverse-scope interpretation (not every tester won the game).  Some researchers in the language acquisition literature have suggested that surface-scope interpretations are preferred by children and perhaps by adults too (Musolino & Lidz, 2006; Frazier, 1998, 2000; but cf. Gualmini, 2004).  We therefore investigated adults' interpretations of sentences like (1) in contexts that supported either the surface-scope interpretation (where none of the testers were expected to win the game) or the inverse-scope interpretation (where some of the testers were expected to win the game).

Surprisingly, across a set of five experiments, we found that there is a sizable population of people that effectively ignores the context in interpreting this ambiguity.   Approximately 20-30% of people generally obtain the surface-scope interpretation of an ambiguous sentence like (1), and a similar proportion of people generally obtain the inverse-scope interpretation.  There are several important conclusions that follow from this work so far:

(a) Context does not always guide all individuals' interpretation, as has been implicitly assumed in the sentence processing literature recently.

(b) There are substantial individual differences in how people resolve ambiguity.

Our current work is investigating:

(a) understanding the use of context in language processing: why some contexts disambiguate interpretations for all individuals, and other contexts do not; 

(b) what heuristics people are using when the context does not disambiguate the interpretation for them; 

(c) what cognitive features might correlate with participants' sensitivity to context, and the heuristics that they use to resolve ambiguity.

Finally, much of this work has been possible due to the existence of Amazon.com's Mechanical Turk, a great new resource for gathering linguistic data rapidly.  It is our belief that Mechanical Turk will revolutionize the way that much linguistics research is done.