February 10, 2005

Can There Be a Purely Semantic Account of NPI Licensing?

Let's start with a whole mess of disclaimers. NPI licensing has been thoroughly studied, and not by me. So these examples may well have been gone over, or I may have missed something obvious. Also, my intuitions may be slightly odd because of my exposure to different dialects of English (as I explain below). But...

I've mentioned that positive "anymore" is used in Pittsburghese. So

(1) Gas is really expensive anymore

is acceptable in Pittsburghese, but not in the dialect I speak most of the time and perhaps not in the dialect you speak. [I'm going to call that dialect "SAE," though like Suzette I'm not sure whether there is any such thing or if it's what I speak. But if I don't have a three-letter abbreviation for a dialect that's not Pittsburghese, this post will never get finished.]

(1) means the same as "Gas is really expensive nowadays," which is OK in either Pittsburghese or SAE. The following sentence is also acceptable in SAE as well as Pittsburghese:

(2) Gas isn't really expensive anymore

which means, or at least implicates, that gas used to be expensive and now is not expensive. That (2) is acceptable in SAE but (1) isn't indicates that "anymore" is a negative polarity item (NPI) in SAE. It's controversial how to characterize the environments that license NPIs. Here's a paper by Kai von Fintel on them (I haven't completely absorbed it, so maybe it contains a solution to the problems I'm going to raise).

OK, now consider these two sentences:

(3) I can't look at a pretzel anymore.
(4) Anymore I can't look at a pretzel.

My feeling is that (3) is acceptable in SAE and Pittsburghese, but (4) is acceptable only in Pittsburghese. I slip between dialects a bit, so maybe these intuitions aren't the most reliable, but that's my intuition. That would indicate that in SAE "X I can't look at a pretzel" isn't an NPI-licensing environment for X, whereas "I can't look at a pretzel X" is. And that seems to be hold for "ever," which is an NPI in Pittsburghese as well as SAE:

(5) *Ever I can't look at a pretzel.
(6) I can't look at a pretzel ever.

(Actually, I don't think "ever" is always an NPI--it's not in "It was ever thus"--but I think when it isn't an NPI it always sounds somewhat archaic. Put another disclaimer in there.)

Now, as far as I can tell the most common accounts of NPI licensing have to do with the semantics of the environment in which the NPI appears. NPIs, it is thought, are licensed within the scope of an operator that is downward entailing, or Strawson downward entailing, or not upward entailing, or some variation thereof. (See von Fintel and Daniel Rothschild for explanations of these terms, and some theories.)

But it seems to me that the difference between (3) and (4) can't be explained in terms of their semantics, because in Pittsburghese, where both are acceptable, they mean exactly the same thing. Indeed, it seems to me that for many temporal modifiers X ("anymore," "ever," "lately," "on Friday"), "X I can't look at a pretzel" and "I can't look at a pretzel X" mean the same thing if they're both acceptable. If that's true, then there's no hope for a purely semantic account of NPI licensing.

There are some cases in which the order makes a difference:

(7) Every day I can't look at a pretzel.
(8) I can't look at a pretzel every day.

In (7) "every day" scopes "I can't look," in (8) it's the other way around. So perhaps in (3) "anymore" is within the scope of the negation, and in (4) it isn't. Nevertheless, as far as the semantics go it seems as though they have exactly the same readings. Does this mean that here there's a difference in scope that determines NPI licensing even when it doesn't determine truth conditions? Maybe.

On the other hand there's

(9) On Fridays I don't teach.

(10) I don't teach on Fridays.

In both (9) and (10), on the most natural readings, "on Fridays" scopes the negation. So this doesn't settle the question of scope. If anyone can settle the question of scope for me, and explain whether this restores the possibility of a semantic explanation for NPI licensing, I'd appreciate it.

Posted by Matt Weiner at February 10, 2005 11:17 AM

Some quickies. More some other time.

1. You conclude "in SAE 'X I can't look at a pretzel' isn't an NPI-licensing environment for X, whereas 'I can't look at a pretzel X' is." I think that's right.

2. I would say that the reason is that in the one form (3) negation has scope over the adverb "anymore", while in the other (4) the adverb is outside the scope of negation. Of course, your objection is that in Pittsburgh the two sentences mean the same. But I would say that's because in Pittsburgh there are two "anymore"s, one of which is the SAE NPI, the other which means not much other than "now". It just so happens that the two meanings come out to be the same in these structures.

3. Even if the semantic scope aren't clear, English is known to have something like an additional syntactic requirement on NPI-licensing, namely that the licensor (here negation) has to c-command the NPI, in essence it needs to be introduced higher in the tree structure. This is used to explain why even though in English subjects can have semantic scope under negation in sentences like "All that glitters is not gold", we don't find sentences like "Anybody is not here" with the expected meaning that nobody is here.

4. So, in some sense you might be right that there can't be a purely semantic theory of NPI-licensing. But I'm not sure that your particular data make the point because they are amenable to a scope-based explanation.

5. The equivalence of (9) "On Fridays I don't teach" and (10) "I don't teach on Fridays" is indeed puzzling. It's a fact that "generic" phrases like "on Fridays" seem not behave as if scope with respect to negation is meaningful. This is a problem for an analysis that would make them something like universal quantifiers. In my paper on Bare Plurals, Bare Conditionals, and "only", I rehearse some more facts like that.

Posted by: Kai von Fintel at February 11, 2005 06:19 PM

I wish I could take back my comment from last night. This needs some further thought. I'm currently suspecting that what's going on in Pittsburgh is very similar to what goes on in Standard English with respect to "even". "Even" has been argued to be ambiguous between an NPI "even" and an item "even" that appears in non-negative contexts. Other analyses involve a scope ambiguity instead of a lexical ambiguity. The analysis of Pittsburgh "anymore" could profit from looking at the relevant work on "even". There's a chapter in Mats Rooth's dissertation, and papers by Wilkinson (in "Natural Language Semantics") and Rullmann (on his webpage), and much more.

Posted by: Kai von Fintel at February 12, 2005 10:16 AM

If you really like, I can delete the comment, but it's interesting enough that I'd like to leave it up--of course with the retraction intact. Point 3 is the well-known thing that I was missing about NPI licensing (because well-known about NPI-licensing doesn't mean I know it).

I should say that I'm not really a reliable informant about Pittsburgh English--I've been around it a fair amount but am not really a native speaker.

Positive "anymore" is slightly odd in that I don't know if "Anymore p" entails that p once was false or only implicates it. It will be interesting to compare it to "even."

Posted by: Matt Weiner at February 12, 2005 11:23 AM