MacDonald (1993)

Comments

advocating for a constraint-based aproach where lexical and syntactic ambiguities cannot be separated constrasting the "delay" model in Frazier & Rayner (1987)

reworked findings of F&R:

• reading times in the ambiguous region of a sentence where faster than in the corresponding region of the unambiguous control sentence
• reading times in the end region of an ambiguous sentence where slower than in the unambiguous control sentence

sample material from F&R

• the desert trains
• this desert trains
• these desert trains

objection: deictic determiners introduce other effects, i.e. without context (see p.699)

renewed reading experiments in F&R using supportive and nonsupportive semantic context to strongly bias lexical cross-categoryal ambiguities;

additional material

	supportive bias	unsupportive bias
ambiguous	The union told reportes that the corporation fires many workers each spring without giving them proper notice.	The union told reportes that the warehouse fires many workers each spring without giving them proper notice.
unambiguous	The union told reportes that the corporations fire many workers each spring without giving them proper notice.	The union told reportes that the warehouses fire many workers each spring without givin them proper notice.

See p.701 for a detailed discussion.

Extraction of probablistic measures from WSJ and Brown:

• to constrain head vs modifying noun
• to constrain noun/verb ambiguity
• word coocurrence

More handcrafted constraints grounded on experiments with native english-speakers on plausibility: semantic bias between noun and ambiguous noun/verb

conclusion: different levels of linguistic representations and measures can be unified under one processing architecture: "the nature of ambiguity may differ across different levels of representation, but the processes that resolve the ambiguities at each level may all operate via the partial activation of alternative interpretations in the light of available evidence." (p. 711ff) (sounds sooo familiar); "constraint-based accounts have the weakness in the specification of the kinds of information that can affect ambig. resolution and how this information is weighted" (paraphrased). this clearly contrasts Frazier (1989) where different levels of representation require qualitatively different processing mechanisms.

-- MichaelDaum -- 10 Nov 2004