LINGUIST List 3.238

Wed 11 Mar 1992

Disc: Parsing, Selectional Restrictions

Editor for this issue: <>

Directory

1. Rick Wojcik, Re: 3.223 Parsing Problems
2. Carl Alphonce, Parsing
3. "Larry G. Hutchinson", Re: 3.229 Selectional Restrictions

Message 1: Re: 3.223 Parsing Problems

Mike Maxwell wrote:
> A few years ago, I developed a grammar of English for a
> parsing program written by Phil Harrison at Boeing
> Computer Services. After reading the above, I thought
> it would be fun to see if "my" grammar parsed the
> sentence ambigously. Phil tried it out, and behold,
> there is also a third parse...
> which can be paraphrased as "the player who was kicked
> the ball that was kicked to him." (Both Ss are reduced
> relatives.)

Right. The Boeing parser (nicknamed the "Sapir Parser" now) produces a total
of 8 parses for the word sequence. Here they are:

 [The player kicked the ball] kicked him.
 ...subject...

 [The player] kicked [the ball kicked him].
 ...object...

 [The player kicked] the ball kicked him *gap*. ('kicked' = passive verb in
 ...topicalized NP... reduced relative)

 [The player kicked] the ball kicked him *gap*. ('kicked' = participial adj
 ...topicalized NP... in reduced relative)

 The player kicked [the ball kicked] him. ('kicked' = passive verb...)
 ...object...

 The player kicked [the ball kicked] him. ('kicked' = participial
 ...object... adj...)

 The player kicked the ball kicked him (Mike's NP mentioned above)

 [The player kicked] the ball kicked him (meaning roughly
 "the player kicked which the ball
 kicked him"
> I should mention that Phil had to add a
> subcategorization for this passive to the lexical entry
> for "kick" before the sentence would parse.

Right. But it would have been easier for him to type in the sentence:

 The man sent the man sent him.

This string has the same number of parses, but all the ambiguities are
semantically felicitous. The problem with the original sentence was that
it is semantically anomalous to have balls doing the kicking. Nevertheless,
you want your syntactic parser to see all the possibilities in case it
encounters "the man sent..." type of situation. Some people may have trouble
getting the distinction between postmodifier 'sent' as an adjective and as
a verb. It makes more sense if you take a participle like 'joined', where the
semantic distinction between the verb and adjective readings is more salient.
I do not claim that these are all the possible parses, only the ones that our
system has been programmed to generate.

				-Rick Wojcik (rwojcikatc.boeing.com)

A number of comments and questions regarding the parsing of

	The player kicked the ball kicked him.

Regarding the five possible parses of greenboas.cogsci.uiuc.edu
(Georgia Green):

	I can only get 3 of the 5 parses you claim are
	valid - can you explain the others? Your parses
	are:

	(1) [ the player [ (who) kicked the ball ] ] kicked him

	At least in my dialect, it is impossible in this case
	(where relativization is from subject position) to
	leave out the "who" or "that" - and I believe this is
	generally the case (please correct me if I am wrong):

	(a) [ the rat [ that the cat bit ] ] died
	(a') [ the rat [ the cat bit ] ] died

	(b) [ the rat [ that bit the cat ] ] died
	(b') * [the rat [ bit the cat ] ] died

	(2) [ the player [ (who was) kicked the ball ] ] kicked him

	Fine.

	(3) the player kicked [ the ball [ (that was) kicked (to) him ] ]

	Fine.

	(4) the player kicked [ the ball [ (that) kicked him ] ]

	Same comments as for (1).

	(5) the player complained that the ball kicked him

	At the risk of making my ignorance public (if it isn't already :-)
	I must admit that I had to look up "kick" in the dictionary to
	understand this meaning of the sentence. In an (older) one I found
	this meaning:

		"to show opposition, resentment, or discontent"
		 (from Webster's Seventh New Collegiate Dictionary, 1971)

	Vicki Fromkin <IYO1VAFMVS.OAC.UCLA.EDU> in 3.212 also suggested
	this as a possible parse:

		The player kicked the ball kicked him.
		The player kicked (complained) (that) the ball kicked him.

	I have never heard "kick" empoyed this way - is it perhaps not
	very widely used these days?

Regarding the meaning of "garden path" from <00HFSTAHLKELEO.BSUVC.BSU.EDU>
(Herb Stahlke):

	It could very well be that the term has "undergone a semantic
	shift". I have always understood it to refer to structures which,
	although perfectly grammatical, have a structure which "fools"
	one into pursuing a straightforward analysis which turns out
	to be incorrect - leading one down a garden path . . .
	I must admit I have never heard of it being applied to island
	violation structures - but then there's a lot I haven't heard
	of :-)

	Can anyone enlighten us?

Regarding the third possible parse from maxwelljaars.sil.org:

	Sounds good to me! I guess I was subconsciously thinking
	that it had to be a sentence rather than just an NP.

Carl Alphonce
(alphoncecs.ubc.ca)

Message 3: Re: 3.229 Selectional Restrictions

The principal goal seems to be to get a generative grammar, of whatever
stripe, to generate only sentences a speaker MIGHT actually use. Many
generativists have rejected this goal, after initial flirtations with it
some years ago.

On the one hand, this rejection seems reasonable, since so much rests on what
beliefs various speakers have about the world, rather than about language.

On the other hand, languages do have ways of encoding these beliefs formally,
and a grammar has to say SOMETHING about this. The trick is to specify the
mechanisms and how they encode nonlinguistic beliefs, without enumerating these
 beliefs. This can be done pretty easily, making
a grammar "portable" from one speaker to another, from one knowledge system
or belief system to another. I first proposed "grammatical implicationals"
back in the mid 1970's to do precisely this, in fact. Disregarding niceties
of formulation, such implications take the form
 "If you believe X, do Y"

Examples: 1) If you believe the head denotes human-like objects, use "who"
 as the relative pronoun.

 The apparent oddity of such NP's as "the train who
 arrived late" comes from the fact that it commits a
 speaker who has this grammatical belief about English
 to the belief that trains are human-like, a belief
 most of us don't appear to share. It would be a hideous
 mistake to assume that the only way to get 1) into a
 grammar also requires marking the word "train" in the
 lexicon as [+Human].

 2) A number of African languages show "notional concord"; anaphors
 and concords are taken from a set of forms indicating
 animacy when the controlling referring expression, whatever its
 grammatical gender, is believed to denote an animate object.
 This is a grammatical fact. But there are mismatches between
 European and African belief-systems, so what actually gets
 SAID will vary. For instance, Temnes believe a certain kind of
 tree is animate (capable of casting spells and the like). This
 is an anthropological fact. Temnes use animate concords in
 this case, which the grammar predicts when coupled with the
 belief-system of a Temne speaker. Europeans speaking Temne use
 regular gender-contolled concords in this case, though, as the
 vary same grammar predicts when coupled with their belief-
 system. The grammars do not differ at all, and it is most
 expressly NOT the case that a grammar of Temne has to try to
 specify for every referring expression (an infinitude, by the
 way) whether its denotation is believed by (some, all, most?)
 speakers possesses the nonlinguistic property of animacy.