Noncompositional

A few days ago I happened to catch a show on the Game Show Network on some of the worst moments in game show history. One of the segments was of an episode of Family Feud. Dawson was host, and it was the Fast Money portion of the game (where two members of the winning family are chosen to each answer the same five questions for a chance at the grand cash prize). The two players were teenagers, and both of them couldn’t find a response to this prompt:

Name something to which you might lose your keys.

The first player heard the prompt several times before Dawson gave a longer explanation (”you know, you have things, and you have keys that you need to use them: what is one of those things?” or something like that). The second player gave a false start to the question, and then said, “What? What does that mean?” Dawson gave another explanation, to which he got a response, and a “Oh, so that’s what they want” from the kid.

The next day I gave the prompt to two people, one a linguist and the other not. Interestingly, neither of them could understand the question the way the writers on the game show intended it. In case you still don’t get it, a clearer phrasing would be name something the keys to which you might lose (or, even clunkierly, name something such that it needs a key to be used, and you might (be likely to) lose that key). Both of the people I told the prompt to interpreted it as a sort of indirect object construction on a par with name someone to whom you might give a present. It’s not often the case that we lose keys to other people, and even rarer that we lose them to things — you can lose your fortune to the stock market, but it’s not a salient use of lose. So interpreting the question like that would lead to confusion. So would, I imagine, not being used to pied-piping with relative pronouns.

But for whatever reason, I had no problem interpreting the prompt. The key to understanding it is being able to recognize extraction of certain arguments out of an NP. William Davies and Stanley Dubinsky have an article (in vol. 21 of NLLT) in which they argue that extraction out of NPs is conditioned on lexical semantic properties of the head noun. Drawing on the (previously-mentioned) categorization of nominals into complex event, result, and concrete nominals, D and D show that nouns “with participants” allow extraction out of the NPs they head. Participants? I’ll get to it.

First D and D show that complex event nominals, which they assume, following Grimshaw, have argument structure, allow extraction (what (play) did you see [the production of _]). Result nominals also allow extraction, but show “definiteness effects,” which means that extraction out of a definite DP is impossible (who do you hope for [a victory over _], but *who did they hope for [the victory over _]). Concrete nominals do not allow extraction: *who did you play with [a dog of _], *what features are you looking for [a camera with _]). They then take on the case of two-faced (er…polysemous) nouns that have both concrete readings and “informational” readings. These are the so-called representational nouns, or “picture” nouns, like picture, book, and film.

Representational nouns (or rather the lexemes that correspond to them) have two readings: concrete and informational. You can read, download, and enjoy a book, but you can also burn, bind, and enjoy (in a different way) a book. D and D show that in the informational reading, these nouns behave like result nominals (allowing extraction out of non-definite NPs), and in the concrete readings, they behave like other concrete nouns like dog. Thus: which presidents do children often read books about, but not *which presidents do children often shelve books about.

The differences between complex nominals and result/informational nominals is seen in their lexical conceptual structure. Complex nominals have argument structures identical to that of verbs, plus an event variable. In Grimshaw/DD style, this looks like (Ev, (x (y))), with the external argument being outermost. In contrast, result nominals look like (for victory and essay) N, (R = x) such that x results from y winning over z or N, (R = x) such that x results from y writing about z, where the nominal is bound to some Role (R). Concrete nouns are bound to a Role, but have no associated prose with participant variables. And that’s where the shit hits the fan. Or so one might think.

First, these are supposed to be somehow semantically determined. D and D say that only the “necessary participants” are part of the LCS of the informational nouns. They say

If we understand result nominals to denote an entity or state that arises as the outcome of some event, then we can restrict participants of result nominals to the necessary participants of the related event.

So for a victory the participants are the two sides. For an (abstract, information-denoting) essay there is the writer and the topic. Some problems: (1) what is the ontological category of “participant?” — human, sentient, animate, physical, entity, what? (2) what is “necessary?” (3) is it really true that only nouns with participants allow extraction? (4) why should these informational/representational nouns be special in having participants (i.e., being profilable by their creator and “aboutness”), as opposed to, say, furniture, which has a (human) creator and a particular purpose, or kinship terms, which are inherently relational and thus should be associated with some sort of “argument” structure? Let’s go one by one.

First, the ontological category of “participant.” Well, with just victory and essay we have winners, losers, writers and topics. The first three are generally human, though metaphorical uses allow diseases and such to win and be won over. But topics: they can be anything: times, places, people, events, linguistic theories, and so on. So it may be hard to restrict one’s self ontologically. Why couldn’t one say that the LCS of essay was “x that results from y writing about z during time-interval t at place(s) p[1...n]“? I suppose you could, since events generally occur at a time and place, so they are “necessary.” Unfortunately, extraction of place/time phrases out of NPs is disallowed (*what location were they hoping for victory on _). Here D and D could have used something like FrameNet’s concept of core and peripheral arguments, or any sort of participant-profiling system where some “participants” are just more salient than others: and for an essay the time of creation just isn’t as inherently important as the topic. But they would run afoul of some other examples, and this brings me to (2): what is “necessary?” First some abstract nouns:

Who do you think [a/the risk to _] is great? (risk to the elderly) What (action) do you think [a/the risk of _] would be unwise? (risk of standing up to one’s superiors) Who were you hoping for [a resemblance to _] when you sculpted this thing? (a resemblance to one’s husband) What (aspect) are you examining this ancient pot for [a similarity to Tupperware in _ ]? (similarity in design) What would you condone [vengeance on an enemy for _]? (vengeance for the killing of one’s brother)

Some of these seem okay, like risk (to) and resemblance, and possibly vengeance. The other ones, I don’t know. I mean, they’re all sort of weird. But I’d argue that all of the extracted portions are “necessary participants:” a risk is a chance of a bad think happening to someone, or (in another sense) of some particular event resulting in a bad thing; resemblance is to something; similarity is in some aspect, and vengeance is for some past negative action that deserves retribution. If grammaticality differs between these examples, then there must be some other factor involved that D and D didn’t consider.

And then points (3) and (4), the supposed “specialness” of informational nouns. This brings me to the original example: what the heck is a key, anyway? Surely one reading has it as a physical artifact. But each key is a key to something. Sure enough, extraction is possible out of an NP[key] in simpler environments than the Family Feud prompt: what is that a key to (cf *who is that a dog of). But can it be argued that there is some “informational” content of a key just like book and photograph? I don’t know if I really know how to go about answering that question. It seems like the wrong question to ask, in fact: there shouldn’t be any need to resort to “informational” meanings: just necessary semantic roles evoked by the nouns in question. Of course, you could come back and say that you find the Family Feud prompt ungrammatical: in which case, I would ask: “Why, if keys are keys to things, should there not be a participant structure licencing the extraction?”

In the end, consider this: a linguist might look at a contrast like the one just below and nod, thinking, “yes this is an important contrast which must be accounted for.”

Who have you most recently read a book about? *Who have you most recently burned a book about?

Meanwhile, a lay person (cough you know who you are cough), confronted with the same two sentences, might have no problem interpreting and answering both of those questions. What do we make of that?