November 21, 2003

Phineas Gage gets an iron bar right through the PP

On September 14, 1848, the Free Soil Union in Ludlow, Vermont, carried a news item that began:

As Phineas P. Gage, a foreman on the railroad in Cavendish, was yesterday engaged in tamping for a blast, the powder exploded, carrying an iron instrument through his head an inch and a fourth in circumference, and three feet and eight inches in length, which he was using at the time.    [from a scan on Malcom Macmillan's Phineas Gage information page]

I happened to read this item a couple of days ago while preparing a lecture on emotion for Cognitive Science 001. It reminded me of something that I left out of my earlier post on crossing dependencies in discourse structures: within-sentence syntactic relationships also often tangle.

To understand the phrase "carrying an iron instrument through his head an inch and a fourth in circumference" as the writer intended us to, we have to recognize that the inch-and-a-quarter measurement modifies the iron, and not Phineas' head -- which is in the way in this sentence, just as it was on that September day in 1848.

It's fair to consider this an unhappy stylistic choice. On the other hand, folks sometimes write this way, and they talk this way even more often (and often the results are not so likely to be mentally red-penciled by the audience). In some languages, and some registers of English, syntactic tangling like this is normal. In fact, the only thing that's really troublesome in the Gage example is that 'which' struggling to swim upstream to 'instrument' ...

Tangling of surface syntactic relations is certainly not a new discovery. Among recent treebanks, the German TIGER corpus project's "syntax graphs" permit crossing edges, and so does the analytical level of the Prague Dependency Treebank (where crossing relations are called "non-projectivity").

Of course, different frameworks of syntactic description, and different theories about how to explain them, offer different stories about what such apparently crossing relations really are, how they arise, how to think about them. This is the source of many of the non-terminological differences among approaches to syntax. Are the issues in tangling discourse-level relations the same, or partly the same, or entirely different?

Posted by Mark Liberman at November 21, 2003 07:48 AM