July 16, 2006

Linguist thought able to read isn't

A grim story, but what a great garden path headline!

Doctor Suspected in Town House Collapse Dies

I was led down multiple garden paths with this one. First, I thought that the doctor suspected something ...

Then I came across the "in", and tried again, this time thinking that it's the beginning of a relative clause, and the doctor is suspected (by someone) to be in the town house ...

Then came "collapse", and it took me the longest time to figure out that it wasn't a typo ...

But I finally figured out that it was the town house that had collapsed, that the doctor had been a suspect in the collapse, and that the doctor had died (as it turns out, as a result of the collapse) ...

Phew! Who'd have thought it'd be so hard to parse a simple headline?

Other mentions of garden path phenomena on Language Log:

Garden paths at the Guardian (9/21/2004)
Blinded by content (6/4/2005)
Burnt offerings (9/30/2005)
Surprising crocodile kin (1/26/2006)

[ Comments? ]

[ Update, 7/17 --

Maryellen MacDonald, a renowned authority on garden path sentences (and psycholinguistics more generally), offers some useful clarification to my post:

The headline Eric cites contains two syntactic ambiguities that I and many other psycholinguists have studied. The first one is the most widely-investigated in sentence processing research, typically called the main verb/reduced relative ambiguity, in this case whether "thought" is the main verb of the current clause or is the start of a reduced relative clause, where "reduced" here refers to the omission of the optional complementizer (who, that, which) and auxiliary verb, as in "Doctor who is suspected..." The second ambiguity concerns the structure of NPs, specifically whether a particular N is the head of the NP or a pre-nominal modifier of some upcoming head N. In the headline, it turns out that the third N (collapse) was the head, but Eric initially thought that "house" was the head and "collapse" was a verb. This ambiguity is less extensively studied, but it's very useful for illustrating how pervasive the task of syntactic ambiguity resolution is--every noun a comprehender encounters admits this head/modifier ambiguity, and unless they're hermits, people encounter thousands of nouns a day.

Psycholinguists who study syntactic ambiguity resolution have made a lot of progress in the last few decades understanding why people, constantly bombarded with ambiguity, typically get to the intended interpretation of an utterance with no conscious awareness that they had any ambiguity to resolve, and they only rarely encounter a garden path. One view most closely identified with UMass linguist Lyn Frazier (who gave the field the term "garden path" as well as many other insights) is that people initially adopt the syntactically simplest interpretation (e.g. as in Eric's first tree) and are garden pathed (yes, we make the term into a verb) if a more complex structure is needed. An alternative view often called "constraint-based ambiguity resolution" holds that people rapidly and unconsciously converge on the correct interpretation by weighing lots of information, in part concerning how the words and word combinations have been used in past experience. A garden path often occurs when the usage in the ambiguous sentence is unexpected given past usages of words and phrases.

For the "doctor suspected" ambiguity, some relevant information includes how often "suspected" is used in the active vs. passive voice and how plausible it is for a doctor to suspect vs. be suspected (and many other constraints). One source of the garden pathing here is that doctors are people who do a lot of suspecting (e.g., about the cause of someone's illness), and unlike, say, mobsters, they are not highly frequent suspects themselves. For the head/modfier ambiguity, some of the relevant information includes the frequency with which a noun has previously served as a head vs. a modifier, the frequency of a collocation (e.g. town house), and the frequency with which a word is used as a noun vs. a verb. Part of the difficulty in the headline comes from the fact that "collapse" is highly frequent as a verb, which favors the interpretation that "house" is the head of the NP.

Psycholinguists have observed before that the truncated style of headlines makes them a good source of garden paths. Perfetti and colleagues, in a 1987 Journal of Memory and Language article [see citation below--EB], have a trove of amusing ones, including PENTAGON PLANS SWELL DEFICIT; TEACHER STRIKES IDLE KIDS, and TORONTO LAW TO PROTECT SQUIRRELS HIT BY MAYOR.

Perfetti, C.A., S. Beverly, L. Bell, K. Rodgers, and R. Faux. 1987. "Comprehending newspaper headlines." Journal of Memory and Language 26, pp. 692-713.

-- end update. ]

Posted by Eric Bakovic at July 16, 2006 03:15 PM