Wednesday, August 09, 2006

Confusion creates the possibility of topicalization

I recently received a long but clearly personally addressed email in German. Besides being personally addressed, it was clearly about linguistics, so I was fairly confident that it wasn't from the wife of a dead Nigerian millionaire. Consequently, due to my very embarrassingly complete ignorance of German, I was forced to resort to an online translator to get the gist of the message.

(A person can know that German is SOV with V2 word order in matrix clauses, that it has four cases, three genders and strong and weak determiners, that the verb 'help' in German takes a dative object, and that the position of objects in the middle field is correlated with their definiteness and specificity, and that person still can be completely unable even to order a beer by any means other than pointing and saying, ' "Beer", bitte.' Luckily this works. Sometimes I'll throw in an 'Ein' before the 'beer' to fool people into thinking that I'm saying 'Bier', but it's just a pathetic pretense. Very, very sad.)

Thanks to the translator, I got it, and am extremely grateful to the translation service, since it did so much better than I could have in the equivalent time. Nonetheless, (since I do know about the V2ness), I can't help being especially amused at one of its mistakes.

Here's the relevant German sentence:
Stefan Schierholz und ich möchten Sie hiermit persönlich einladen, ebenfalls an diesem für unser Fach so wichtigen Projekt mitzuarbeiten.
And here's the translation:
You would like to invite Stefan Schierholz and I herewith personally to cooperate likewise in this so important project for our field.
And I didn't even know I was involved in organizing this so important project for our field!

I *guess* the thing must have imagined that the German sentence was topicalized, and that the pronoun Sie is actually the subject of the verb möchten, though that seems odd given that in fact in this case the preverbal subject mirrors exactly the intended English sentence. This translator knows too much! (It's not because of any context, either, because this translator gives the same results for this sentence in isolation as in the original block of text.)

The irony is that actually this is one of the better-translated sentences in the group; many others are much less comprehensible. Here's another pair, the next sentence in the email:
Um sich eine Vorstellung der Artikel zu machen, finden Sie als Überblick über das Gesamtprojekt auf der Webseite meines Lehrstuhls eine Liste aller zu bearbeitenden Schlagwörter, so genannter Lemmata.
And the English:
To do to itself an image of the articles, find as an overview about the whole project on the web page of my chair a list of all catchwords to be worked on, so-called dictionary entries.
Obviously I can understand what it means, but that impersonal reflexive really threw it for a loop.

Actually, comparing this translator,, to the Babel Fish, Babel Fish does better on who does what to whom, but misses the obvious personal name in that conjoined subject. Here's the Babelfish interpretation:
Stefan ski wood and I would like to invite you hereby personally to likewise cooperate in this project so important for our subject.
And, just for completeness, here's how the Babelfish does on the other sentence:
In order to make itself a conception of the articles, you find a list of all as overview of the overall project on the web page of my chair key words which can be worked on, Lemmata so mentioned.
Here the Paralink translator loses in the first clause but wins in the second. The Babelfish is boggled by the reflexive too, but manages 'conception' rather than 'image', which is much closer to what's intended, I think, so that's better. But the Babelfish makes a total hash of what should be a noun phrase, 'a list of all the key words', interleaving the modifier 'as overview of the overall project' and location PP 'on the web page of my chair' in the middle. The other translator uses 'catchwords' instead of 'key words' but otherwise does better with the structure of the noun phrase, and gets the better 'so-called dictionary entries' where Babelfish leaves 'Lemmata so mentioned'. So basically they're both problematic but have different problems. The Paralink translator makes you look at a popup ad before you can see your translation, though, so it's not as nice if you don't have a good popup blocker.

But who am I to complain about the results? I'll have plenty of time to order my Biers and listen to German in the next three weeks while participating in the DGfS/GLOW summer school in Stuttgart, which I'm really looking forward to. Maybe I'll even learn how to say where I'm from and what I do. And if they say, Linguist? But you don't speak German? I'll demonstrate my beer-ordering capabilities, and that ought to mollify them.

Sunday, August 06, 2006

The case of the $2,000,000 comma and the ambiguous adjunct

Check it out — a comma in a contract is costing a Canadian communications company $2,000,000! A Basic Rule of Punctuation.

The long and the short of it is that Rogers Communications had a deal with another company, Aliant, Inc, who had agreed to string cable across utility poles in Eastern Canada for a fixed rate per pole. Rogers understood their contract (and per-pole price) to be in place for five years, and to be subject to renewal after that. Aliant Inc disagreed. They understood the contract to allow them to cancel at any time, with one year's notice. (After cancelling, they could then renegotiate the rate.)

The court case hinges on whether the the phrase about cancellation with one year's notice takes scope over the entire sentence about the duration of the contract, or (as Rogers thought) only the second conjunct of that sentence. Here's the crucial sentence (as excerpted in the Globe and Mail):
The agreement “shall continue in force for a period of five years from the date it is made, and thereafter for successive five year terms, unless and until terminated by one year prior notice in writing by either party."
So the issue is whether the constituent structure of this coordinated sentence is as indicated in (1) or (2) below. Rogers would have it that it's (1); Aliant is arguing that the placement of the bolded comma indicates unambiguously that it's (2):

(1) [ [ The agreement shall continue in force for a period of five years from the date it is made] and [[ thereafter for successive five year terms] [unless and until terminated by one year prior notice in writing by either party]] ].

(2) [ [[ The agreement shall continue in force for a period of five years from the date it is made] and [ thereafter for successive five year terms]] [unless and until terminated by one year prior notice in writing by either party] ].

Rogers, naturally, argues that the intent of the deal was understood by all at the time of its making:
Rogers was dumbfounded. The company said it never would have signed a contract to use roughly 91,000 utility poles that could be cancelled on such short notice. Its lawyers tried in vain to argue the intent of the deal trumped the significance of a comma. “This is clearly not what the parties intended,” Rogers said in a letter to the CRTC.
But to no avail. The CRTC lawyers, "armed with the rules of grammar and punctuation" decided that a comma setting off an adjunct clause from a conjoined structure necessarily gives it matrix scope, and if Rogers' contract lawyers didn't know that, too bad for them.

Does anyone know *what* prescriptive rule precisely is at issue here? Although the article mentions English textbooks and rules of grammar, no precise formulation of the relevant rule is given. Is it a sensible one? Or a foolish one?

I'm personally inclined to think it might be foolish. The description of the decision implies that the rule states something like this:
When an unless-clause modifies locally, it may not be set off from the clause it modifies by a comma.
That seems to imply that when an unless-clause is used unambiguously, locally modifying a simplex sentence, you are forbidden to use a comma to set it off. That is, the punctuation of the following sentence is incorrect:
Mary will ask John to the dance, unless he asks her first.
Contrast that with the commaless punctuation:
Mary will ask John to the dance unless he asks her first.
It actually seems to me that both are fine, AND that the comma makes a difference—the two sentences would be spoken with quite different intonation patterns (the first with nuclear stress on John and an intonation break before unless, the second with nuclear stress on first and no intonation break). Hence the commaless and commaful variants likely do have different structures at some level—but both are perfectly felicitous, and in both cases, the unless-clause is happily modifying the most local clause. Because the presence or absence of the comma is linguistically signiicant, though, no prescriptive rule should forbid it from preceding a locally-modifying unless-clause.

Then the real question is, can a conjoined sentence with an unless-clause spoken with the intonation break indicated by the comma allow modfication of the second conjunct only? Try for yourself:
Jane will ask Bill to the dance, and John will ask Sue, unless Phil asks Sue first.
In this situation, obviously the unless-clause is intended to modify only the second conjunct (since it would be pragmatically odd if Phil asking Sue to the dance had any bearing on whether Jane asks Bill). In my judgment, speaking the sentence with nuclear stress on the first occurrence of Sue, and with an intonation break before unless does not force the unless-clause to take matrix scope, thereby rendering the sentence pragmatically odd. Rather, the sentence is pragmatically natural, and the intended scope is perfectly accessible, with that intonation.

In my professional opinion then, Rogers has been unjustly done out of its $2.13 million; I think their contract language does permit the reading they thought they'd agreed to. On the other hand, there is no question that the matrix reading is also available (though, contra the court decision, it's not the only reading available). Further, it is the presence of the comma that makes the matrix reading available—the comma introduces ambiguity. Eliminating the comma would have eliminated the matrix reading, since the nuclear stress associated with the second conjunct would then have to fall in the unless-clause. So maybe the Rogers contract lawyers are really in the wrong after all, for letting such an ambiguity remain in their contract language. That's what they're getting the big bucks for, after all. They should have asked a linguist.

Update: Actually, now I think my example is not the best one to make this point. For a conjoined sentence to be true, both its conjuncts must be true. So truth-conditionally, it makes no difference whether the unless-clause in my example is attached locally to the second conjunct or to the matrix clause, taking scope over the first conjunct as well; [A and [B unless C]] has the same truth conditions as [[A and B] unless C]. Doesn't it? Though maybe the comma after A indicates that it is to be evauated as an independent proposition. Hmmm.

Update update: No, no, not so. [[A and B] unless C] is true if B and C come to pass but A doesn't. But [A and [B unless C]] is false in that situation. Phew. I think it's all ok & my original remarks stand.

Update update update And see the comments for a much better example with the temporal properties of the original from koldito. Also, Mark at the Language Log has weighed in on the matter, also suggesting the lawyers should have watched out for the ambiguity. But I still think that if the intended reading is available, and if common sense tells you that it WAS the intended reading, then the judge can't claim that the comma forces the matrix reading.

Update update update updateMark Liberman has now posted the text of (one version of) the actual rule in question, thanks to an alert reader, and it does say that a comma should be taken to indicate scope over the entire clause, and that otherwise one should follow the Law of the Last Antecedent. Looks like Late Closure was discovered in 1891 -- very sensibly surrounded with acknowledgements that the overall sense of the text should be taken into consideration. Not an adequate basis for a $2,000,000 decision, I would think!

Friday, August 04, 2006

Cognitive jetsam

(1) On yesterday's Daily Show, guest Vali Nasr summed up the situation in Lebanon with the following lovely example of a nonce causative manner-of-motion verb, one my fave features of English grammar:
You can't shock-and-awe Hezbollah out of Beirut.

B: With enough computational space, couldn't you implement a natural-selection approach to any programming problem? Just give it a few clues so the stuff will compile, give it a few little algorithms that it can embed or modify, and let the thing string together lines of code, looking for sequences that will do something useful towards a user-defined goal. Take the best ten results from every generation as the starting point for the next, and may the best code win. Maybe people are already doing this? (Thanks to a commenter for pointing me to this link about the progress being made in this area -- it's only a matter of time before humans never have to think again... :) )

III. Maybe pagoda-style roofs are good in rainy climates? Rain wouldn't run along to collect on the corners and erode foundation-threatening holes at the corners of the building, but rather drip with less impact off the length of the eaves?

Δ - Trying to decide on the meaning of a completive aspectual verb, ya'ate, in Hiaki (Yaqui, Yoeme) today, we were looking at the following sentence:

Haivu yukya'atek.
Haivu yuk-ya'ate-k
already rain-ya'ate-perf
"It's already stopped raining."

We had some reason to think that ya'ate means something more like 'finish' than 'stop', but the main verb in the example here, and the translation, of course, made me wonder if it might really be more like 'stop' than 'finish'. I said as much, and someone asked me why -- what's the difference between stop raining and finish raining? Well, I said, let me put it this way: in Newfoundland, where I grew up, it does occasionally stop raining. But it never finishes raining. Har.
[Here's the thing: 'Rain' is usually a pretty good example of an activity verb, rather than an accomplishment like, say 'evaporate'—activities just come to a halt whenever, rather than culminate at a necessary endpoint. The meaning of 'stop' goes with pretty much any verb with a duration, while 'finish' is a bit pickier, preferring to go with a verb that has a natural endpoint built in. With accomplishment predicates you can really see the difference:

The water stopped evaporating ---> still some water left
The water finished evaporating ---> water all gone

He stopped building the house ---> complete house doesn't exist
He finished building the house ---> complete house exists, move on in.

So if ya'ate means something like 'finish', then it might be a little unusual to see it in combination with a non-culminating verb like 'rain', in an apparently episodic context like this. But maybe not; it's not like 'It finished raining' is uninterpretable or even unnatural. Heck, rain here in the southwest really does 'finish' -- the monsoon rains we're having right now will finish in a couple of weeks, and we won't see any more for at least three or four months. We're going to have to get some examples with good accomplishment predicates, and check the entailments.

Let me tell you, figuring this stuff out is fun! I am one lucky linguist.]