- ([info]entangledbank) wrote,
@ 2004-05-07 22:13:00
Previous Entry  Add to memories!  Tell a Friend  Next Entry
Current mood:puzzled

More than I have
Well, Geoff Pullum and Mark Liberman have just both written about it, and there's only one of me, so I suppose that means that more people have written about this than I have.

Except that, as they both point out, that last bit is complete nonsense. But it's not ungrammatical. It seems to make perfect sense. Until you try to work out what it means. More people have written about this than I have. More people have been to Russia than I have. It's completely uninterpretable, taken exactly as written; yet out parsers don't blink at it, not even when you point out it's nonsense. So something very powerful is letting it through: it's a parsing error, but more like a Müller-Lyer illusion than a garden path sentence, in that you can't successfully hear it as ungrammatical even when you know it ought to be.*

So what's going on? Reconstruction turns it into 'More people have written about this than I have written about this'. (Why? The unstressed 'have' at the end indicates stripping: it's an IP with the VP missing, and there's only one prior VP in the sentence, and it fortunately has 'have' as its I. So no choice there.)

Next the scope of the 'more' has to be worked out: the canonical use is 'More people did X than did Y', so we expect something akin to VP coordination, 'Subj VP1 and VP2'; so it's logically something like 'More(Subj) (VP1 than VP2)'. An interpretable example would be 'More people have written about this than have read about this'.

First problem. The interpretation of phonetic 'NPi VP1 and VP2' as 'NPi VP1 and NPi VP2' with reconstructed coindexed subjects overrides another interpretation where VP2 has a PRO subject, perhaps arbitrary, or a contextually supplied anaphor. Why should phonetic 'VP1 and VP2' be so strongly read as a conjoined VP constituent? Why not parts of two IPs of which the second needs its subject pragmatically supplied?

Apply this objection to the 'than' conjunction. In the nonsense sentences, 'VP1 than VP2' is failing to be read as a VP-level conjunction, governed by a single wide-scope 'more'. It's allowing a different subject in the second clause. Okay, I was too hasty about the canonical use of 'more'. What about 'More Americans have been to Moscow than Russians have been to Washington'? That is, instead of 'Q(x) (F(x) than G(x))' we have a more distributed quantification, and I'm not sure how to notate this, 'Q: x.F(x) than y.G(y)'. This allows different subjects; and it allows the predicates to be the same: 'More Americans have been to London than Russians have'. Or even: 'More Americans have written about this than Britons have'. And that's what we feel the original sentence almost means.

So disregard the previous three paragraphs: I was on a wild-goose chase, and the question is why does it sound okay when what sounds okay involves quantifying into 'I': 'More Americans have written about this than More(I) have written about this'? The 'than I have' must be picking up an alternative reconstruction from earlier in the sentence. Well, there's 'Americans have done it more than I have'. It doesn't say that. Or does it? The word 'more' there is quantifying something that isn't in the sentence, not overtly: a frequentative aspect in the inflection, perhaps: more times, more often. How about if at Logical Form that 'more' is up the front along with nominal quantifiers? So we have an LF that begins with 'More()', and the parser has to reconstruct what it's quantifying.

The first thing it tries is the subject of VP1 interpreted as subject of both (e.g. 'More NP write than read'). But VP2 turns out to have a distinct subject, so that interpretation is discarded. Then the quantifier slides down IP1 and finds it can quantify an empty frequency element...

No, it's no use, I still can't explain why it links up with an apparently valid interpretation of a quantified IP2, and I could be here all night doing this, and I want to go to the pub to make myself sick with beer on top of this red wine so I can study the usual amount tomorrow. Hooroo...


* They seem to have settled on a name for them: Escher sentence.


(5 comments) - (Post a new comment)


[info]bicoherent
2004-05-07 09:00 pm UTC (link)
"There are more Ps than Qs" translates to #{x : P(x)} > #{x : Q(x)}, where #A denotes the cardinality of A. "More Americans have been to Moscow than Russians to Washington" will roughly be #{x : American(x) ^ Travelled(x, Moscow)} > #{x : Russian(x) ^ Travelled(x, Washington)}. It can't be expressed in first-order logic, I think.

The "There are more Ps than me" business makes semi-sense if one stashes the "Me" into a singleton set, i.e. #{x : P(x)} > #{Me} = 1.

And remember Lewis Carroll's "Have some more tea"...?

(Reply to this)

more more
[info]entangledbank
2004-05-08 02:14 am UTC (link)
mutters...

There's another structure that's almost what it seems to mean: 'More people than me have done it'. But this means 'I'm not the only one who's done it'. If just one other person has, then more people than me have; but that to me isn't the reconstruction that my parser is trying to tell me works. What I get a sense of is that, say, three or four people have done it, so that's more than me. What about 'More people have done it than Jenny and Susan have'? Same problem. Take out the final 'have' and it makes sense: perhaps Marina has also done it, so it's not just the two of them who have.

But with the 'have' back in, it seems to be saying that some number greater than 2 have done it in the first clause, so at least 5 altogether have...

It doesn't depend on whether the VP is the same, either: 'More people have written it than Jenny and Susan have read it'. That's equally wrong but I don't even know what symbol (* # ?) to put in front of it. Okay, *'More people have written it than Jenny has read it' is pretty clearly ungrammatical, while ?#'More people have written it than Jenny has read' is marginally interpretable -- perhaps five authors have written it and she hasn't read them, but 'more than she has read' really doesn't make much sense, not with reading. It does with other verbs: if Susan buys four apples and Jenny buys three then Susan owns more than Jenny has bought. But reading? Okay, that's therefore probably irrelevant.

Now take off the verb and leave the auxiliary and we're back to nonsense: 'More people have written it than Jenny has'.

puts daisies in hair, sings, skips...

(Reply to this)

Escher sentence
(Anonymous)
2004-05-16 12:38 pm UTC (link)
My parser decoded it as "More (and better qualified) people than I have written about this..." with the rhetorical impact of modesty, perhaps false. That is, it would either be followed by

(1) "...so my contribution might be superfluous"
or
(2) "...so I must be right"

Now, that's garbled, but not meaningless, anymore than "Throw me out the window my sweater" is really confusing (no, I'm not claiming the constructions are parallel, merely a weak analogy). You simply need a bit more context to see if it makes sense: "More people have been to Russia than I have, and not all of them speak Russian." You start saying "More people than I have been to Russia, but you get to the verb phrase too soon ("More people have been to Russia") so when you get to the "than I" it looks naked, which you cover with an extra verb. Alternatively, you start to say "More-qualified people than I" but that doesn't sound right, so you hold off on the "qualified" piece, and then find no room for it later in the sentence.

Are there two things going on - the process by which the sentence is put together, and the Escher process by which it is heard as correct? I'm not sure that the second is as impressive when the generating process is as unremarkable as speed-induced error. Is it?

(Reply to this)

Escher sentence
(Anonymous)
2004-05-17 11:57 pm UTC (link)
I'm confused, too, because the sentence makes perfect sense to me. Even after reading Pullum and Liberman I see nothing wrong with it. Now if fewer people had written about this than I have....

(Reply to this)


[info]perm_den
2004-06-27 06:38 pm UTC (link)
thanks!

(Reply to this)

(5 comments) - (Post a new comment)

Create an Account
Forgot your login or password?
Login w/ OpenID
English • Español • Deutsch • Русский…