Docsity
Docsity

Prepare for your exams
Prepare for your exams

Study with the several resources on Docsity


Earn points to download
Earn points to download

Earn points by helping other students or get them with a premium plan


Guidelines and tips
Guidelines and tips

Limiting Locality in Combinatory Categorial Grammar: Rules for Grammatical Sentences, Study notes of Linguistics

An overview of locality restrictions in combinatory categorial grammar (ccg), a specific type of categorial grammar. It explains how certain words in a sentence 'look for things with holes in them' and the importance of ensuring that holes cannot be carried too far up in the derivation to prevent ungrammatical sentences. The document also discusses the role of type-raising and function composition in ccg and the need to introduce restrictions on type-raising to prevent certain problematic derivations.

Typology: Study notes

Pre 2010

Uploaded on 07/30/2009

koofers-user-q8g
koofers-user-q8g 🇺🇸

5

(1)

10 documents

1 / 3

Toggle sidebar

Related documents


Partial preview of the text

Download Limiting Locality in Combinatory Categorial Grammar: Rules for Grammatical Sentences and more Study notes Linguistics in PDF only on Docsity! Locality Restrictions in Categorial Grammar By looking over this stuff you should come away with a basic understanding of the kinds of ways we can fine-tune this grammar formalism to rule out ungrammatical sentences. It’s not important to remember exactly which restrictions get placed on which rules to account for which ungrammatical sentences, but you should understand the kinds of restrictions we can use in categorial grammar1 to enforce constraints on “how far holes can be carried”. • We’ve seen various cases where certain words “look for things with holes in them”. 1. The first was quantificational sentences like ‘John met every girl’. The phrase ‘every girl’ needs to combine with something of type S/NP, or “an S with an NP-shaped hole in it”. So we did some slightly fancy tricks (type-raising then function composition) to get ‘John met’ into a phrase with this type. John met S/NP every girl S\(S/NP) S < 2. Then we saw relative pronouns, words like ‘who’, ‘whom’ and ‘which’, which also look to combine with an S which has an NP-shaped hole in it. (See Steedman 1996, chapter 3, page 49.) In the following example I’ve put the relative clause ‘whom John met’ together with ‘a’ and ‘girl’ to give a bit of context, but the important part for our purposes here is just the way ‘whom’ combines with ‘John met’. a NP/N girl N whom (N\N)/(S/NP) John met S/NP N\N > N < NP > 3. And questions are another case where something needs to combine with something of type S/NP or S\NP, signifying a property (at least, in English; see the last question of homework 1). • Sometimes the S/NP or S\NP that we need to construct gets very large, meaning that there are a lot of steps which “hold open” the NP slot. In the following example, the hole indicated by the /NP has been carried all the way up from ‘met’. a NP/N girl N whom (N\N)/(S/NP) Bill said Mary thought Fred claimed that John met S/NP N\N > N < NP > 1More precisely, what we’re dealing with here is a specific type of categorial grammar (CG) called Combinatory Categorial Grammar (CCG). All the earlier stuff we’ve covered was generic enough to be in line with any CG, not just CCG, although the books I’ve been using are in fact CCG books. 1 Extraction of subjects • Carrying the hole a long way, however, sometimes seems to create strange problems. The NP in (1) which we just sketched the derivation of above is fine, but if we change it very slightly into (2), it sounds bad2: (1) a girl whom Bill said Mary thought Fred claimed that John met (2) * a girl whom Bill said Mary thought Fred claimed that met John • The basic strategy for ruling out bad sentences like (2) is to ensure that, it won’t be possible to “carry the hole” all the way up from ‘met’ to a point where we have got ‘Bill said Mary thought Fred claimed that met John’ into an S/NP. Happily, the system we already have correctly blocks this quite elegantly, since the derivation gets stuck at this point: that S/S met (S\NP)/NP John NP S\NP > bzzzzt! These two categories, S/S and S\NP, cannot be fit together. In the derivation of the good sentence, (1), this problem does not arise, and we can continue all the way up to get ‘Bill said Mary thought Fred claimed that John met’ into an S/NP: that S/S John S/(S\NP) met (S\NP)/NP S/NP >B S/NP >B • We could, if we wanted (i.e. if we were studying some language where the equivalent of (2) was grammatical) propose a grammar which includes a rule for composing S/S with S\NP. Specifically, the relevant rule would be forwards crossing composition; see Steedman 1996, chapter 3, page 53, and chapter 2, pages 42–44. But since (2) is bad in English, we do not want to include this rule in our grammar of English.3 • The four versions of the composition rule that Steedman gives on page 43 (in chapter 2) are the ones which languages “choose from”. English uses (106a), (106b) and (106d), but not (106c). Other languages will make different choices, resulting in different word orders. Extraction from adjuncts • Other examples where we seem to be limited in where we can carry holes look like (4). The sentence in (3) is here just for comparison, although they don’t make a nice minimal pair like (1) and (2) do. 2Observant readers will have noticed that if we leave out the ‘that’, then both of these sentences are fine. The presence of ‘that’ makes no difference to the story we tell here about these two sentences; in other words, the predictions of the theory as we have it are that you should be able to extract objects but unable to extract subjects, period. If you’re curious about how Steedman goes about allowing extraction of subjects when ‘that’ is not there, see section 3.4.3, pages 57–60. But be warned, it ain’t pretty. 3But recall that we have seen that English does seem to make use of the backwards crossing composition rule, at least sometimes; eg. for composing ‘saw yesterday’ in the last question of homework 1. 2
Docsity logo



Copyright © 2024 Ladybird Srl - Via Leonardo da Vinci 16, 10126, Torino, Italy - VAT 10816460017 - All rights reserved