Promoting interest in HPSG
Ivan A. Sag
sag at csli.stanford.edu
Fri Jun 25 16:00:05 UTC 2004
Hi everyone,
Just to throw in my 2 cents worth... I believe people working in HPSG should
submit their work to mainstream linguistic conferences and journals. I enclose
a message on this topic (with updated weblinks) that I sent to a number of
mailing lists in 1998. If a new conference is created, it definitely should
not have anything like `nonderivational' or `nontransformational' in the
title. Rather it should simply lead by example, showcasing excellent work in
nonderivational frameworks...
(People disagree with me about this, but) I personally think that the annual
HPSG meeting should not be seen as the final publication destination for much
of anything and hence that it should be organized less like a conference and
more like a workshop, with time allocated for discussion, more topical
sessions, invited presentations, maybe sessions dedicated to nurturing student
presentations, etc..
Best,
Ivan
PS: The deadline for the January, 2005 LSA meeting in San Francisco is
September 1.
-------------------------------------------------------------------------
Ivan A. Sag
Professor of Linguistics and Symbolic Systems
Stanford University. Stanford, CA 94305
Email: sag at csli.stanford.edu
WWW: http://lingo.stanford.edu/sag/
-------------------------------------------------------------------------
-----------------------------------------------------------------------
from 1998:
-----------------------------------------------------------------------
All,
The deadline for receipt of abstracts for the LSA meeting in January
is September 1. In a recent study of syntax papers presented at the
last two LSA meetings (by a colleague who will remain nameless),the
distribution of frameworks employed in the papers presented looked
something like this.
---------------------------------------------
1996-1997 LSA Meetings aggregated: # %
Total theory-specific syntax 61 100%
Chomskyan (GB/Minimalism) 50 82%
Non-Chomskyan combined: 11 18%
---------------------------------------------
People I talk to frequently express the opinion that it is pointless
to submit non-GB/Minimalist linguistic papers to linguistics
conferences (or journals!) because the reviewing is biased. While it
would be surprising if there were no bias wrt article or abstract
reviewing in a field with a predominant framework, nonetheless, I
believe the primary reason there are so few LFG, HPSG, Categorial
Grammar, or TAG papers at linguistics conferences is simple: there are
very few submissions.
This impression is confirmed by conversations I have had over the last
few years with members of various program committees. Moreover, it
appears that many people who work in these constraint-based,
lexicalist frameworks, perhaps because they are trained in a different
field, research tradition, or research culture, do not submit the kind
of carefully argued abstract that stands a chance of getting accepted
in a highly competitive review process. One colleague serving on a
program committee complained to me that (s)he was tired of reading
abstracts of the form:
1. Here is some data
2. Here is how X is analyzed in GB.
3. Here is how X is analyzed in HPSG.
[end of abstract]
I'd like to suggest that we should all submit more abstracts to
conferences like LSA, WCCFL, NELS, GLOW, ESCOL, WECOL, CLS, BLS, etc.
It may surprise some of you to know that some of these conferences are
so concerned with having some kind of theoretical balance that they
are tempted to look more sympathetically at the few non-GB abstracts
they receive. You may also be surprised at how many people working in
GB/P&P/Minimalism feel that conference review processes are biased
against them.
Anyway, here are a few suggestions for writing the kind of abstract
that can be competetive. Doubtless, some of you reading this have
other suggestions to make. I also enclose (with the permission of the
authors, who will also remain anonymous) three abstracts that were
accepted by program committees at recent general linguistics
conferences. (The residual latex commands are my fault. Sorry - I
wanted to get this out quickly.)
Reactions welcome.
Ivan
-------------------------------------------------------------------------
Ivan A. Sag
Stanford University
Email: sag at csli.stanford.edu
Office: 415-725-2323 (Cordura 228)
WWW: http://hpsg.stanford.edu/hpsg/sag.html
Fax: 415-725-2166
-------------------------------------------------------------------------
Suggestions for writing syntax/semantics abstracts:
1. There are guidelines for abstract writing available from the LSA.
They are useful. The guidelines and abstract submission forms can
be found on-line at [updated June 25, 2004]:
http://www.lsadc.org/annmeet/abstractguide.html
See also the discussion of model abstracts at:
http://www.lsadc.org/dec02bulletin/model.html
2. Deadlines are strictly enforced. Also, you should follow all
specifications to the letter.
3. An abstract should succinctly state an empirical problem and the
nature of the proposed solution. Don't waste any words. You will
almost always want to use as many words as you are allowed. Learning
how to state a problem clearly in a few sentences is essential.
4. An abstract should make reference to relevant previous work. The
trick is to know how to summarize very succinctly what the
essence of a previous insight is or what the defect in some previous
analysis is. It is in your interest to cite relevant current work.
5. The kiss of death for an abstract is to say something like: a
solution will be presented to problem X. The feel of the abstract
has to be more like: `I show that X is really Y (pace
Sapir 32). Five arguments support this claim: (1) X participate
in nasal assimilation like Ys (`yabayaba';`yadayada'); (2) the common
assumption that X is Z (ref) leads to a contradiction about agreement
in relative clauses (in forms such as `xxx'); (3) Only Ys are subject to
W (ref), yet W also applies to X (cf. `foobar')...'
6. Keep the scope of the abstract modest. State a smaller problem and
focus on your solution to that. Abstracts are often rejected on the
grounds that what the author claims (s)he will present in 15 or 25
minutes would obviously take 2 hours to present adequately.
7. Keep your readers in mind. Even if a generalist cannot evaluate
all the details of your argument, (s)he should appreciate some of
your points and above all the tightness of your reasoning.
----------------------------------------------------------------------------
West Greenlandic noun incorporation
as a mixed category construction
Noun incorporation (NI) in West Greenlandic exhibits a challenging mix of
syntactic and morphological properties. To account for NI and related
phenomena, Sadock (1985) proposes that morphology and syntax be treated as
autonomous and orthogonal modules of grammar, so that "every expression
will have two distinct representations, one morphological and one
syntactic" (383). And, he argues, West Greenlandic NI can best be
accounted for as a mismatch between syntax and morphology. However,
syntactic and morphological structures are typically homomorphic and even
in the most extreme cases they diverge in certain highly constrained ways.
So, Sadock (1991) offers a number of principles governing the kinds of
structural mismatch allowed.
In this paper, I offer a strictly lexicalist analysis of West Greenlandic
NI within the framework of Head-driven Phrase Structure Grammar (Pollard
and Sag 1994) which does not require positing divergent morphological and
syntactic structures. Furthermore, I show how Sadock's homomorphism
constraints follow directly from the architecture of the lexicon.
West Greenlandic NI is a category changing morphological operation that
converts a noun into a verb by the addition of one of a set of bound
verbalizing suffixes (Sadock 1991:94):
(1) Marlunnik ammassattorpunga
marluk-nik ammassak-tor-punga
two-INST/PL sardine-eat-INDIC/1SG
`I ate two sardines.'
(2) Ammassannik marlunnik nerivunga
ammassak-nik marluk-nik neri-vunga
sardine-INST/PL two-INST/PL eat-INDIC/1SG
`I ate two sardines.'
The resulting denominal verb (DV) has the full distribution of a verb in
West Greenlandic. Unlike a verb, though, a DV can also occur with an
ergative possessor and a nominal complement which are associated with the
incorporated nominal. On this basis, Sadock argues that the incorporated
nominal must have an independent syntactic existence. I show that West
Greenlandic NI is better viewed as a kind of mixed category construction,
parallel to the English verbal gerund. The possessor and complement that
occur with a DV are not stranded by incorporation nor do they bear a
relation to the incorporated nominal directly. Instead, the DV inherits
its subcategorization requirements from both the verbalizing suffix and the
incorporated nominal. This is exactly parallel to mixed category
constructions in other languages, e.g., the English verbal gerund
"devouring" in "Pat's devouring the pancakes," which occurs with both a
specifier (like a noun) and a direct object (like a verb).
Verbalizing suffixes can be accounted for in HPSG by a lexical rule which
combines the valence properties of the verbal suffix with the valence
properties of the `incorporated' noun stem. A verb derived by this lexical
rule will project a verb phrase following general principles of X-bar
theory and argument saturation. In addition, any constraints which the
noun stem places on its specifier and complement will be inherited by the
DV. This accounts for the fact that the external specifier and complement
have the properties they would have had if they had appeared with the
incorporated nominal alone.
This analysis does involve a limited kind of mismatch: a DV projects a VP
but has noun-like valence requirements. However, the unusual properties of
DVs are restricted to the lexicon and HPSG's independently motivated theory
of lexical information places strong restrictions on the kinds of
mismatches that can be induced. So, there is no need for additional
construction-specific stipulations limiting the degree of mismatch between
syntax and morphology.
----------------------------------------------------------------------------
Understanding Mandarin \ba as a verb
Syntax
Many researchers (e.g., Li (1990)) have analyzed the morpheme \ba of
the Mandarin \ba construction as a direct object marker. Their
evidence is two-fold: \ba does not pass the three established tests
for verbhood, and, in examples like (\ref{core}), \ba does appear to
mark the direct object of the following verb.
(1)a.
L\v{\i} S\`{\i} b\v{a} n\`{a} ji\={a}n f\'{a}ngzi ch\={a}i le
Li Si \emph{ba} that CL(assifier) house demolish perf(ective)
`Li Si demolished that house'
b.
L\v{\i} S\`{\i} ch\={a}i le n\`{a} ji\={a}n f\'{a}ngzi
Li Si demolish perf that CL house
`Li Si demolished that house'
In this paper, I argue that \ba is a verb. First, the verbhood tests
are unreliable, since, for each one, there is a class of
uncontroversial verbs which also fail. Second, there is a wide range
of data that is highly problematic to other accounts, but expected if
\ba is a verb and its valence orchestrates the sentence. This data
includes: sentences where the subject of the sentence is not
interpreted as the subject of the post-\ba verb; sentences where the
NP following \ba is interpreted as the patient of a verb embedded one
clause down; and sentences where the post-\ba verb has another object,
in addition to the one that \ba is marking. (\ref{noncore}), from Li
(1990:184), gives an example of the final type.
(2) W\v{o} b\v{a} j\'{u}zi b\={o} le p\'{\i}
I {\em ba} orange peel ASPECT skin
`I peeled the skin off the orange'
(3) [_{\tn{IP}} [_{\tn{NP:subj}} w\v{o}]
[_{\tn{VP}} [_{\tn{\={V}}} b\v{a}
[_{\tn{NP:obj}} j\'{u}zi]] [_{\tn{IP:comp}} b\={o} le p\'{\i} ]]]]
I argue that \ba takes a subject, an object and a complement clause
(as in (3)) and stipulates that its object is interpreted as
the (sentential) topic of its complement clause. This analysis predicts
sentences like (\ref{noncore}), because \nba{'s} valence
and that of its complement are independent and may be instantiated
by different NPs. That \nba{'s} object is interpreted as the topic of
the complement clause explains why \ba has so many topic properties
(Tsao 1986) and why the two objects in sentences like (\ref{noncore})
cannot be completely independent: there are restrictions on how a
topic may be related to its clause (Tsao 1986).
This analysis brings into question conclusions about Mandarin syntax
based on treating \ba as a direct object marker, such as C.\ Li
and Thompson's (1974) claim that Mandarin is becoming an SOV language,
Huang's (1990) claim that Mandarin has prepositions, and Travis's (1989)
word order parameters.
{\bf Bibliography}
Li, Audrey Yen-Hui. 1990. {\em Order and Constituency in Mandarin
Chinese}. Boston: Kluwer.
Li, Charles N.\ and Sandra A. Thompson. 1974. Historical change of
word order: A case study in Chinese and its implications. In
Anderson, J.\ and J. Charles, eds. {\em Historical Linguistics}.
Amsterdam: North-Holland. 199--217.
Travis, Lisa. 1989. Parameters of phrase structure. In Baltin, M.\
and A.\ Kroch, eds. {\em Alternative Conceptions of Phrase
Structure.} Chicago: Univ.\ of Chicago Press. 263--279.
Tsao, Feng-fu. 1986. A topic-comment approach to the {\em ba}
construction. {\em Journal of Chinese Linguistics.} 15(1):1--55.
---------------------------------------------------------------------
{\bf `Extraction' Without Traces}
Although more than one traceless analysis of unbounded filler-gap
constructions has been presented in the recent syntactic literature
(Steedman 87,88; Kaplan and Zaenen 89; Pollard and Sag 93), it is
widely believed that the phonetically empty categories -- traces --
that are profligate in modern GB analyses of the same phenomena (inter
alia) have empirical motivation outside the technical apparatus of GB
Theory. In this paper, we critically examine essentially every
independent argument offered for the existence of traces, arguing that
none of these is satisfactory from the perspective of modern
linguistic theory. We also offer positive arguments for the position
that filler-gap dependencies are terminated not by empty constituents,
but rather by the lexical heads that are normally thought of as
properly governing the trace.
(i) It is generally assumed that the impossibility of auxiliary
contraction in examples like *How tall do you think he's? is to be
explained by some appeal to the presence of a trace in the
post-auxiliary position. However, in the analysis of these phenomena
provided by Selkirk (84), a morpheme cannot contract if it is needed
to bear stress, a matter determined by the ABSENCE of other
constituents in the prosodic phrase, not the PRESENCE of empty
constituents. This approach is clearly superior to trace-based
accounts in that it extends to further data, e.g. contraction failure
around parentheticals.
(ii) Wanna contraction is also thought to motivate traces whose
presence blocks application of a contraction rule. But, quite apart
from issues raised by Postal and Pullum (86) and Lightfoot (86), wanna
is best treated (synchronically) as a defective verb, not the output
of a contraction rule (see Fodor (in pr)).
(iii) Floated quantifiers cannot appear at extraction sites (*What
color are the cars all?), and this too has been accounted for by
appeal to traces (Sag 78,80). Yet Dowty and Brodie (84) defend an
analysis of floated quantifiers as base-generated VP modifiers that
immediately explains this fact, under the crucial assumption that
there are no traces, i.e. no following phrase for the quantifiers to
be attached to in these examples.
(iv) Experimental psycholinguistic data have shown that the meaning
of a filler phrase is mentally activated at a gap position (Bever and
MacElree (88); MacDonald (89)), and the appropriate theory of this has
been assumed to require traces. But this finding is compatible with
ANY theory that establishes a semantic link between a filler and the
position of a missing argument - empty constituents are not required.
Re-activation studies can show that dependencies are computed on-line
but no experimental technique is yet known that can distinguish
between empty and absent constituents.
By contrast, Pickering and Barry 91 present data suggesting that
hearers seek verbs to terminate the processing of filler-gap
sentences, not traces. For this reason, the heavy NP in (1)
encountered before the preposition terminator causes processing
difficulty, but the verb terminator encountered before the heavy NP in
(2) causes no such difficulty:
(1) Which box did you put the very large and beautifully decorated
wedding cake bought from the expensive bakery in?
(2) In which box did you put the very large and beautifully
decorated wedding cake bought from the expensive bakery?
In the lexically based, traceless analysis of extraction constructions
we present, such contrasts are precisely the expected result. A
lexical rule applies to a lexical head, moving a complement from its
COMPS list to its SLASH list. `Slashed' verbs project feature
specifications about missing elements up the tree in accordance with
universalprinciples of HPSG theory. We treat extraction irregularities
in the lexicon, e.g. unextractable objects (Who did you allow to/*let
leave early?) and Kayne's (80): Who did you assure us to be good?/*I
assure you them to be good (see Postal 93). Our analysis of `crossover'
phenomena also makes no appeal to traces.
More information about the HPSG-L
mailing list