[Prev][Next][Index]

[XTAG_summary]


Meeting Summary from 9/3/98 (Tonia's summary of Carlos' presentation)
###########################

I. Progressive, perfect, and conditional features
*************************************************
Carlos proposed eliminating the features: progressive, perfect, conditional
These features were used in constraining the adjunction of auxiliary verbs, and
therefore were located on the VP node and passed up to the S node (for inverted
auxiliaries). Carlos proposes that we only need <passive> and <mainv> features
to constrain aux adjoining. 

apparently it had already been decided to eliminate these
features. Subsequently, however, it was discovered that if these features are
eliminated, then sentences like "do you have eaten the apple" would parse. 

This is a potential reason to put <progressive> and <perfect> features back in
the grammar. However, Carlos proposes that <mainv> be used to rule sentences
like this out, rather than <progressive> and <perfect?. That is, the auxiliary
"do" would only adjoin on to a VP that is <mainv:+>.

Apart from this, no other uses were found for <progressive> and
<perfect>. Since the meeting, however, we found that "(*)John was having eaten"
will parse, suggesting that maybe we need the <perfect> feature after
all. ("be" could require that it only adjoin on to VPs that are <perfect:->,
thus ruling out the above sentence.) Carlos is looking into this.

> Does anyone else see any reason why we need <progressive> and <perfect>
features? This may need to be discussed further, given the problem discussed in
the previous paragraph. 

One thing that I wanted to clarify: My understanding is that the proposal is to
delete the features completely from the grammar, not just the links from VP to
S. Is that right, Carlos?

Also, it was suggested that <conditional> should not be deleted because we
might need the feature to implement conditional inversion. I think that this
topic also merits further discussion. 

Other Conclusions: <mainv> and <passive> are needed on VP and must be linked to
S in regular clauses. <mainv>, but not <passive> is needed in small clauses. 

II. confusion about use of <mainv>
**********************************
related to I. If <mainv> is going to be used to constrain auxiliary adjoining,
then it needs to be used consistently. It sounded like it was sometimes used to
refer to a main verb (as opposed to an auxiliary) and other times it was used
to indicate that a verb form or clause could or could not appear as a matrix
verb/clause (as opposed to a subordinated clause). An example is that
subjunctives in the grammar are marked <mainv = ->. 

> Is there a reason we need mainv to mean non-matrix? if so, can we change this
feature to be <matrix> or something similar, rather than mainv?

along the same lines: right now all @PRES verbs are marked mainv = -. again,
this seems wrong. Can Carlos go ahead and change this?

III. null adjoining constraints on anchors?
*******************************************
The question was whether there are anchor nodes which require null adjoining
constraints. 

The consensus at the meeting seemed to be "no" -- we should not have null
adjoining constraints. Even idioms which don't allow some kinds of
modifications ("let's bury the little hatchet"), do allow others ("let's bury
the proverbial hatchet"); and the constraint seems to be semantic, not
syntactic. 


Date: Wed, 9 Sep 1998 10:48:28 -0400 (EDT)
From: "Carlos A. Prolo" <prolo@gradient.cis.upenn.edu>
To: Tonia Bleam <tbleam@linc.cis.upenn.edu>
Subject: Re: summary from last week's meeting

It couldn't be better the summary. Here are some additional comments.

Although I initiallly proposed that small clauses (SC) don't use <passive>
(as it is currently), I raised the point of whether raising verbs in
the past participle can anchor SC trees to form perfect constructions:
	John had seemed tired. Have John seemed happy ?
Currently they can't (at least syn db doesn't have ppart entry for seem and 
appear), but the answer in the meeting  (I think) was "yes" they should be 
allowed. Hence we have to include <passive> in SC to block "was seemed". Right?

As for wiping out the features from the lexicon, yes, that's the idea. But
only when/if we are very certain. It's easier to change the grammar than
the lexicon. As Tonia said it requires care and maybe more discussion.

BTW, what about
	"By this time, tomorrow, John will be having completed half
		a century of existence"
Does this sound good or not ? If "be" can't really take perfective
constructions than we have new data and as Tonia suggested <perfect> would
be used to block, with the proper additions to the lexicon.

One last thing to think about. There's nothing to force us to be minimal 
w.r.t. the number of features. We may conclude that all the five features
are sound and independently motivated, even if not stricly independent in
a more "mathematical" sense (i.e. we MIGHT handle everything without some).
However, if we decide for that, then I think ALL the auxiliary entries in the
lexicon should tell about their status w.r.t. the features (i.e. not only
to guarantee blocking as last recourse). Again, topic for discussion.

Carlos