LINGUIST List 3.175

Sat 22 Feb 1992

Disc: Parsing challenges

Editor for this issue: <>

Directory

1. William J. Rapaport, Re: 3.154 Parsing Challenges
2. "NAME " William Marslen-Wilson "", RE: 3.154 Parsing Challenges
3. "Bruce E. Nevin", parsers vs. humans
4. Will Fitzgerald, Re: 3.154 Parsing Challenges

Message 1: Re: 3.154 Parsing Challenges

Well, there's my Buffalo favorite:

Buffalo buffalo buffalo buffalo buffalo.

Here's the story. First, when I was a grad student in philosophy at Indiana
a long time ago, John Tienson gave us the example of:

Dogs dogs dog dog dogs

whose syntax is the same as:

Mice cats chase eat cheese.

We found the -s morpheme unaesthetic, so we came up with

Buffalo buffalo buffalo buffalo buffalo

where to buffalo = to bully around, to do a snow job on. (Poetic license
if you think that buffalo can't buffalo other buffalo.) Of course, there's
also the optional plural of buffaloes as mere buffalo, with no plural marker.

Then, of course, you can make it more interesting by considering the buffalo
in the Buffalo zoo, the Buffalo buffalo. And their peculiar way of
buffaloing other Buffalo buffalo, so peculiar that, like the Tennessee waltz
which you do by Tennessee waltzing, they Buffalo buffalo those other
Buffalo buffalo:

Buffalo buffalo Buffalo buffalo Buffalo buffalo Buffalo buffalo Buffalo buffalo.

			William J. Rapaport
			Associate Professor of Computer Science
			Center for Cognitive Science

Dept. of Computer Science||internet: rapaportcs.buffalo.edu
SUNY Buffalo		 ||bitnet: rapaportsunybcs.bitnet
Buffalo, NY 14260	 ||uucp: {rutgers,uunet}!cs.buffalo.edu!rapaport
(716) 636-3193, 3180 ||fax: (716) 636-3464

Message 2: RE: 3.154 Parsing Challenges

I wonder if the garden path sentence cited by John Limber ("The player
kicked the ball kicked him") is as mystifying when first encountered if
it is spoken (with appropriate intonation) rather than written.

In some recent research we find that the appropriate intonation pattern
disambiguates potentially ambiguous sequences like "The workers considered
the offer from the management..." just as effectively, in an on-line
probe task, as the presence of an overt lexical cue ("The workers considered
that the offer from the management...).

This raises the possibility that many of the popular garden-path phenomena
reflect the inadequacies of English orthography in conveying prosodic
structure, rather than telling us about the normal operations of the
psychological parser.

William Marslen-Wilson

>J_LIMBERUNHH.UNH.EDU

J. Limber offers the following example as being easy for parsing systems
but difficult for humans:

(1) The player kicked the ball kicked him.

One reason (1) is difficult for humans is that (2) is not all that
likely or acceptable:

(2) A player was kicked the ball.

Perhaps there are dialects in which (2) is ordinary. I predict that in
such dialects the relative clause in (1) would also pose much less
difficulty than for speakers of dialects like mine.

The homophony of past tense and past participle is the evident source of
difficulty. There is no problem with either of the following:

(3) A player was thrown the ball.
(4) The player thrown the ball kicked it back.

The distinction blurrs a bit with other verbs, e.g. "send":

(5) A contestant was sent a letter.
(6) The contestent sent a letter sent it back.

Here, it seems to me that (5) is much more acceptable and normal than
(6). This may be because "send" is closer to the core agent/patient
semantics. And indeed, in Harrisian operator grammar, I believe that
the simply transitive "X kick Y" is basic, with the distransitive "X
kick Y to Z" being derived from a source using a verb of the "send" set,
such as "X sent Y to Z by kicking it." (See the discussion of derived
verbs in _A Grammar of English on Mathematical Principles_.)
Nonetheless, (6) is still a garden-path sentence because the past-tense
reading of "sent" is so much more expectable (and probably more
statistically likely) than the passive in a relative clause with
relative pronoun zeroed.

Computational parsing algorithms seldom take into account the relative
likelihood of alternative parses, other than for imposing a search
order. Certainly, they do not model that which in humans gives rise to
the differences of likelihood as a byproduct.

Perhaps this is covered in the cited
>discussion
>in:Limber, J. (1976). Syntax and sentence interpretation.
>In R. Wales (Ed.), Walker, E. C. T. (pp.). Amsterdam: North Holland.

	Bruce Nevin
	bnbbn.com

Message 4: Re: 3.154 Parsing Challenges

I'm sorry, but I can't resist. The 'had had' sentence (John, where Mary
had had "had" had had "had.") takes advantage of the use/mention
distinction. Of course, you can mention other examples of mentioning, so
you can create infinitely long sentences. Mark, where I had had "had had
'had' had had 'had'" had had " had 'had' had had 'had.'"; extend this for
as long as you like. Points off for incorrect punctuation.

/-------------------------------------------------------------------------\
| Will Fitzgerald | In February it will be my snowman's anniverary |
| Institute for the | with cake for him and soup for me! |
| Learning Sciences | Happy once, Happy twice |
| Northwestern Univ.| Happy chicken soup with rice. - M. Sendak |
\-------------------------------------------------------------------------/