The Hourglass

I was browsing some blogs recently when I happened upon this post by Tigerhawk. He noticed the following AP headline and lede.

House Democrats to unveil Iraq war plan

In a direct challenge to President Bush, House Democrats are advancing legislation requiring the withdrawal of U.S. combat troops from Iraq by the fall of next year.

He objects to the title because the democrats' proposal is not a war plan but rather a retreat plan. One commenter, Shochu John, takes the view that the headline sounds fine: "It means plan regarding the Iraq war, no?" This is an interesting difference of opinion. Whose interpretation is correct? Well, linguistically, both of them.

The difference arises from the structural ambiguity of the compound noun phrase Iraq War Plan. There are two different ways to parse this phrase, as follows.

1. [Iraq [war plan]]
2. [[Iraq war] plan]

Tigerhawk seems to be interpreting the phrase with the parse shown in (1), with the meaning of a war plan concerning Iraq, while Shochu interprets it with the second parse with the meaning of a plan for the Iraq war.

This kind of ambiguity is quite common with compounds as the following examples show.

3. big book bag
4. American football players

Sometimes, though, the ambiguity is less obvious because one of the interpretations is just entirely implausible as in the following cases.

5. red shopping list
   [red [shopping list]]
   ?[[red shopping] list]
6. red vegetable increase
   ?[red [vegetable increase]]
   [[red vegetable] increase]

The only natural interpretation for (5) is that the list is red—I have no idea what red shopping might be. Similarly, the only natural interpretation for (6) is that the vegetables are red—a red increase means nothing to me. Of course, red shopping and red increase might be able to take on some sort of idiomatic meanings, but the point here is that they can't be used meaningfully on the fly: No consistent meaning can be derived from them compositionally.

So both Tigerhawk and Shochu John are right. However, that raises a couple of different but related questions. First, is there any natural reason to prefer one alternative over the other, and second, why did Tigerhawk and Shochu not see each other's interpretations?

For the first question, there are some studies to show that people do have some underlying preferences for how to parse ambiguous structures. One central idea here that might give us some answer to this question is the late closure principle (usually attributed to Lyn Frazier in her Ph.D. dissertation). According to this, people prefer to attach incoming words and phrases to the existing structure within the most recently analyzed constituent. For instance, there is a preference to interpret (7) below as meaning that Jim met Barbara yesterday rather than that Sally said so yesterday. In this sense, the processor prefers to keep the current constituent 'open' as long as possible—hence, the name 'late closure'.

7. Sally said that Jim met Barbara yesterday.

The late closure principle has mainly been used to look at attachment of adverbials to the most recent embedded clause or a higher independent clause. I have not yet found a study which seeks to apply it to the context of noun compound ambiguity. Assuming that late closure does extend to noun compound parsing, then what would we expect? Let's use (3) above, big book bag, to illustrate how the processor will parse the phrase. The parse would be built as shown in (8)-(10).

8. [_A big]
9. [_NP [_A big] [_N book]]
10. [_NP [_A big] [_N [_N book] [_N bag]]]

When the processor sees big it will parse it as an adjective node. Then when it sees book it will parse it as a noun node and then combine it with big to form a noun phrase (NP) node. However, when the processor sees bag then it will parse it as another noun node and then combine it with the most recent constituent, book to form a noun-noun compound. Thus, the preferred interpretation of the phrase would be a book bag which is big.

This fits my intuition when I read the phrase with neutral prosody—flat intonation (stress on different words in the phrase varies my judgment, as non-neutral prosody is known to do). I admit, though, that I'd like to know if this judgment is broadly observed. As I said above, I've not yet found a study that actually investigates the processing of ambiguous compounds, so this might be a nice area for some fresh research.

The remaining question is why Tigerhawk and Shochu John differ in their interpretations of Iraq war plan. Well, Tigerhawk appears to follow the late closure principle and therefore assumes the default interpretation. More speculatively, his interpretation might be reinforced because of the position he wishes to take that the Democrats' proposal is not a war plan. But how about Shochu John? how does he arrive at the alternative interpretation? Well, here my answer is entirely speculative. It is clear from his several comments on the page that Shochu intends to take a contrary position to Tigerhawk's point, and the alternate interpretation of the phrase supports that. So the question is whether Shochu is overlooking the default interpretation in order to arrive at the alternate interpretation. But for this question, there is no direct evidence one way or the other.

One final question is to what degree the writer of the AP headline was aware of the potential ambiguity of the headline as written. Here, again, we can only speculate, but it is somewhat telling that, according to Tigerhawk, the headline was later changed to “Democrats want Iraq pullout by fall 2008″. Tigerhawk remarks that this "is both illuminating and more precise." Perhaps so, but crucially here, it's no longer ambiguous.

[Update (1 Sep 2007): Heidi Harley describes a similar problem with ambiguity in noun-phrase formation and drives the point home nicely with graphical depictions of the competing structures as well as audio files showing how prosody disambiguates the two possibilities.]

This entry was posted on Sunday, June 24th, 2007 at 11:58 and is filed under Linguistics, Syntax. You can follow any responses to this entry through the RSS 2.0 feed. Responses are currently closed, but you can trackback from your own site.