Empty Versus Null

So I've been having a lot of fun this summer teaching Introduction to Linguistic Analysis. I admit I wish attendance was higher - but it's the summer session, so what can you do? Those students who do come are by and large bright and inquisitive - they ask the right kinds of questions - and really, being thrown for tough questions is a teacher's greatest reward.

When you ask people what they get out of teaching, they usually say sentimental things about "reaching people" or "shaping minds." And I guess those things are nice, yes. But what I like best about it is that teaching something, like programming it, forces you to really look closely at the subject - from fundamentals up.

This is hugely important! It's easy to get complacent about what you know and locked into "established wisdom" in your field. Teaching it to novices can give valuable perspective.

I got a question like this today. I was teaching X-Bar Theory - the idea that all phrases in natural language have the same essential structure. And of course, one of the problems with such a theory is that we postulate lots of positions in the structural representation of a phrase (or sentence) for purely theory-internal reasons - just to keep all our phrases consistent with the template that we want to believe in. In more detail - all phrases are assumed to have a head (a single item which determines the type of phrase), a specifier and a complement. Complements are usually arguments (like the direct object of a verb) - things that are required for grammatical purposes. And specifiers are usually modifiers of some kind. But not all phrases have obvious specifiers and complements, and so attempting to draw syntax trees sometimes leaves us with a lot of "blanks" in the structure that are not being used. The theory assumes they have specifier and complement positions - but not every phrase needs to use these positions all the time, and so we're left with blanks in our structure trees.

Of course, sometimes we want to say that positions are filled even though they're not pronounced - and this notion is a big problem for some people to grasp, counterintuitive as it is. In these situations we sometimes say that there is a "null" item - an item that occupies a place in a structural representation but is not actually physically present (which in the case of language means "pronounced" or "written").

One of my students today wanted to know what the difference is between a "null" item and an "empty" position. Why can't we just say that null items are "empty" too?

I think it's a very good question for an intro student to be asking. Of course - I have an answer. But trying to phrase the answer just right for him got me thinking - and it's this kind of thinking, I believe, that is the real reward for teaching.

The answer is that null items occupy positions that the theory needs filled, whereas "empty" positions are cases where there's really no theory-internal reason to believe that anything needs to go there. To give an example: notice that the article system in English has a "hole:"

	Singular	Plural/NC
Definite	the	the
Indefinite	a(n)	---

This is the kind of situation where we want a null item: because there's reason to believe that NOT pronouncing a determiner in front of a plural (or noncount) noun actually means something. NOT saying "the" or "a" (or "some" or "these") is what indicates indefinite plural. That is, there is a position in front of noun phrases where determiners go - and NOT putting something in that position is just as meaningful and grammatical, in some cases, as pronouncing an existing determiner.

The same is not true, say, for adjectives. There is no sense in which choosing not to modify a noun implies anything. There are are a potentially infinite number of adjectives that could have been used to fill the position ahead of the noun - and the fact of our choosing none of them doesn't narrow it down to the last remaining (but tragically unrepresented) meaning or any such thing.

So in the case of the missing adjective, the position is simply "empty," nothing implied. But in the case of the indefinite plural determiner, it is the fact of not pronouncing anything in the position that gives us the meaning we're after. This is what "null items" are for.

Granted - lots of syntacticians get carried away with such theory-internal invisible placeholdes. One of the dangers of allowing invisible (read: unpronounced) items, of course, is that you can really explain anything this way. The theory is too powerful, and the explanations it gives thus strike some as hokey.

To the extent that Linguists get carried away with the theory's technology (to the point, say, of assuming an ACTUAL null item in the mental lexicon that people bother to "learn" and then "not pronounce") - this impression is certainly valid. But provided we stay aware of the difference between a placeholder in a theory and an actual postulated item "in the world," there's really no way to deny the existence of null items. The important thing is just not to get to carried away with the word "item."

Well, explaining this today in class sort of pushed me further away from HPSG and further toward Minimalism. I've been slowly drifting this way for about a year now - having started out very anti-Minimalist, but gradually coming to see the point of it all. (Nothing, by the way, has done more to improve my impression of Minimalism than this book.)

One of the attractive things about HPSG was always that it wants to only deal with items we can "see." That is, it takes as one of its tennants NOT postulating (as far as it can help it) the existence of purely theory-internal items. It builds syntax up from what can be seen on the page, as it were.

Going over the distinction between null items and empty places today, though, this struck me as a terrible prejudice. It hit me when joking with the student asking the question that "language" had nothing to do with "speech" that in some important sense I really believe that. Language and speech are separable things - and given such a philosophy of language, there's no real reason why I need to be overly attached to "the signal" when building up my theory. Really - it isn't the job of syntax necessarily to account for why people pronounce what they pronounce in the way they pronounce it. What syntax is meant to do is simply lay bare the grammatical structure of what gets pronounced - as a means to developing an algorithm that derives and all and only the grammatical sentences of a language. If we have reason to believe that a certain thing - say determiners - play a crucial role in the structure we're aiming for, and if the complete lack of a determiner in some cases "means" something determiner-like, then there's absolutely nothing wrong in syntax with putting a "null item" in the place where the determiner would have gone. Just because such an item isn't pronounced doesn't mean "it" isn't playing a role in the structure we need.

Theories like HPSG, I think, lose sight of what it is that they're primarily charged with. It's constituency that matters. It's grammaticality that matters. It's the grammatical template that matters. It isn't primarily the words that matter (though of course they are of crucial importance). If we can build a (theoretical) machine that generates the right sentences, and it does so through use of a "null item," then it is enough. And if you're ending up with the right pronounced sentences through use of such a thing - then in some important sense it doesn't matter that it's "there," since it isn't pronounced anyway. Indeed, there isn't a syntactic reason to steer away from null items at all! The only reason is science theoretic: so as not to needlessly multiply entities. But if there's a need?

And indeed, it begs an interesting question. Like any theory of syntax, HPSG needs to deal with things that aren't necessarily always "on the page." That is, it has to assume, at the very least (and it does!) some syntactic properties of words pulled from the lexicon that explain their distribution - i.e. why you can't say things like *Hat my is on table the. And in some important sense these properties are "invisible" too - we simply induce their existence based on distributional facts. So HPSG's attachment to only dealing with things that you can "see on the page" amounts to an assertion that anything abstract you postulate (and all theories of syntax MUST postulate such things) must be tied to items on the page. Which is all well and good, of course, and useful as a constraint on the power of the theory - but it doesn't seem like a thing motivated by principle. Is there any principled reason to draw the line here? I can't see that there is.

Anyway - this is one of my frustrations with the direction Linguistics is taking. There's a big push to make the field more "objective." As though that would be a good thing. Some fields simply aren't primarily physical - and to treat them as though they were, or to conflate objetivity with materialism, is a mistake. Dealing in abstractions, provided they are motivated by data, does not mean a field lacks objectivity. What syntax has to account for isn't a physical thing - and so I do wish that some syntacticians (me included) would get less worried about trying to please the wing of the department that only beieves in what it sees directly.