-4

I would like a generative BNF-style complete description for English grammar. Some of the more subtle stuff leads to awkward questions of grammaticality (a complete answer to this question, and all related questions, is a publication with a complete description of a comprehensible and comprehensive formal grammar which generates exactly the set of grammatical English sentences):

Here is question 1: is grammaticality of verbs ever dependent on external context from other sentences? If so, then those constructions obviously require some semantics.

The two sentences below are the context

  1. John smiled at Lisa at Kinko's.
  2. John asked Lisa for advice at Kinko's.

After either 1 or 2, you say:

  • I did what John did at Kinkos at James
  • I did what John did at Kinkos of my friend

I believe the second form is certainly not grammatical in case 1. How about case 2? I am not sure if the first sentence is grammatical even in case 1. Should they be

  • I did what John did to Lisa at Kinkos to James?

In other words, do the arguments have to match "do" or the verb that is implied by context by the "do"?

More generally, are there any cases where the grammaticality of a sentence requires looking at other sentences for context?

Ron Maimon
  • 1
  • 1
    Some references for you: "English is not a Context-Free Language", "Evidence Against the Context-freeness of Natural Language", and "Natural Languages and Context-free Languages". The general consensus (actually, no, consensus is too strong of a word) is that most natural language utterances are context-free in the technical sense, but that there are certain constructions that are context-dependent. – Mark Beadles Mar 05 '12 at 03:35
  • I don't have enough info to really answer your question, but I will say that the two examples don't sound right to me even /with/ the context of the preceding sentences, let alone without :) Perhaps they're "grammatically correct" on some semantic level, but in practice I don't think anyone would say it that way. – Lynn Mar 05 '12 at 03:56
  • "of my friend" sounds wrong to me – Bidella Mar 05 '12 at 04:07
  • @Lynn: Consider also "John asked Linda what time it was", "I did too, of James". And "John smiled at Linda", "I did too, at James". These might sound more pleasing to the ear. – Ron Maimon Mar 05 '12 at 05:42
  • @Mark Beadles: I agree, but I think that these papers are generally misused. BNF, with some very minor extra stuff, easily describes English. I am not asking for a rigorously context free description, you can add in extra algorithms that do arbitrary stuff to fix up the remaining very little bit. The cross-serial barrier is nonexistent in English, which is very recursion friendly (no cases). I wrote a sketch for a reasonably complete BNF a while ago, I'm trying to check it, by writing it down precisely, and see if it works for every sentence, and see whether it generates wrong sentences. – Ron Maimon Mar 05 '12 at 05:54
  • 1
    This has been addressed at the linguistics SE, at http://linguistics.stackexchange.com/questions/1090/how-to-test-if-a-string-of-words-is-a-grammatical-sentence – prash Mar 05 '12 at 10:04
  • Your questions seem to be using the word "grammaticality" in an utterly meaningless sense. – David Schwartz Mar 05 '12 at 10:58
  • @David Schwartz: What are you talking about? "James smiled at Linda. I did too, at Wanda." Is this grammatical? I am using it in the standard intuitive sense. I am not asking people here if these sentences are formally grammatical, I am asking if they are intuitively grammatical. – Ron Maimon Mar 05 '12 at 13:39
  • @prash: Thanks for the link--- it seems that somebody provided a link to a page http://erg.delph-in.net/logon that will parse your sentence in as many ways as it can be parsed, and for the examples I checked, the intuitively correct reading is always present. The parser identifies the possible meanings of "I did what James did at Kinko's at James." (but with duplicated readings--- there are only three meanings and it gives 5 readings. The duplication can be eliminated by a smarter ruleset, it is related to the unnecessary ambiguity of levels of embedding in the BNF they are using) – Ron Maimon Mar 05 '12 at 13:55
  • @RonMaimon: People tend to use these words in various ways, and all these ways are encoded in ERG. Reducing the number of parses is one of the things that the delph-in gods are working on. There are 22 analyses of that sentence, not 5. The analyses are all different from each other, but that's something only a person who understands HPSG thoroughly would discern. – prash Mar 05 '12 at 15:28
  • @prash: There are not 22 analyses of this sentence, or 5. There are exactly 3. I understand that you can make 22 analyses which seem "all different" from each other in terms of the formal parsing, but I am saying that there are commutativity rules that make them no different at all in reality of language. These commutativity rules are not apparent in the baroque scheme that these folks are using. The differences come from the order of binding of things whose binding order commutes. It's as wrong as saying that 3+4+5+7 has a bazillion parsings: (3+4)+(5+7),((3+4)+5)+7), etc. – Ron Maimon Mar 05 '12 at 19:39
  • @prash: I am also saying that I understand thoroughly why HPSG is failing to get a good natural language model. These guys are making a stupid elementary error of confusing a tree with a real structure. When there is a commutative operation of binding, there are exponentially many tree descriptions of the binding, as the example of an addition list shows. The assumption of binding commutativity at each level of recursion fixes this problem (and simultaneously also fixes the non-context free issue of some natural expressions). This is the main new idea I am using for the formal grammar. – Ron Maimon Mar 05 '12 at 19:46

1 Answers1

2

Upon further reflection, and Ron's response to my comment, I did want to expand my comment to an answer.

In other words, do the arguments have to match "do" or the verb that is implied by context by the "do"?

"Do" is a special verb that usually does not stand alone. Consider these exchanges:

"Did you meet Patty?" "I did, at the mall."

"Did you go with Patty?" "I did, to the mall."

The responses are really just ellipsis. The full thoughts are:

"I did meet Patty at the mall."

"I did go with Patty to the mall."

Despite the omission of the helping verb, the preposition still needs to match it in the ellipse or it will sound wrong.

Now something closer to your example:

I did what Patty did, at the mall. (Patty put up some posters for the musical)

I did what Patty did, to the mall. (Patty vandalized the theater)

Depending on what Patty did, either could be correct.

So I guess the short answer is yes: context matters.

However, in all of these situations, the context you need comes from the speaker's intent not from the surrounding sentences. When I say "what Patty did" I could be talking about something that happened months or years ago. When I make an ellipsis in thought, the context necessary to decipher that ellipsis may or may not be present in the preceding sentences - it may hearken back to something earlier in the conversation.

Community
  • 1
Lynn
  • 17,801
  • +1: yes, I agree. Unfortunately, this sort of thing dooms an attempt to say whether sentences are grammatical in a stand-alone way, since it requires you to read the speakers mind. I believe this type of construction only occurs in informal speech, and is not allowed in newspapers (but I might be wrong--- "Congress took a vacation, despite the heavy workload. Obama did too, in August.") – Ron Maimon Mar 05 '12 at 13:51
  • 1
    It is not limited to informal speech, though I would agree it's more likely to be found there: From The Economist "Rio’s shareholders opposed the sale but many reckon that the Australian government did, too." (Of course, that one has a "but" in it but didn't have to.) – Lynn Mar 05 '12 at 16:38
  • Thanks, but I this doesn't settle things, because "... the Australian government did, too" is easy to parse, there are no arguments. If they said "... the Australian government did too, of the purchase", then you would have to look at what verb "did" refers to to figure out if it is grammatical. This type of argument shifting is what I was hoping doesn't occur. My Obama Congress example doesn't do it either, I now realize. – Ron Maimon Mar 05 '12 at 18:04
  • @RonMaimon: I see what you mean. I was finding a real instance of your Obama example. I'm sure if you looked enough you would find an example in written form of the real situation you're looking for. It's an uncommon construct, period, but it is not limited to the spoken word. In fact, I would speculate that it's actually more common in the written word because it sounds really awkward to talk that way. – Lynn Mar 05 '12 at 22:49