*!*Ghomeshi, Jila and Diane Massam. 1994. To Appear, Linguistic Analysis



tải về 302.24 Kb.
trang1/10
Chuyển đổi dữ liệu11.06.2018
Kích302.24 Kb.
  1   2   3   4   5   6   7   8   9   10
*!*Ghomeshi, Jila and Diane Massam. 1994. To Appear,
Linguistic Analysis.
Lexical/Syntactic Relations without Projection.

Allows meaning for grammatical constructions.


Syntactic canonical object: the DO in the prototypical
transitive clause (Hopper and Thompson).

Presentational aspect: e.g. English imperfect -ing

Aktionsart: Dowty/Vendler classes (applied to predicates
actually)

Lexical aspect: idiosyncratic properties of verbs

Argues that Persian NI is best explained by allowing the
base-generation of object NPs in more than one position.
Cites Mithun's 4 classes of NI.

Type 1: Lexical Compounding: V+N denotes a conventionalized activity


  Type 1a: composition by juxtaposition: V and O form a tight bond


but remain phonologically separate

  Type 1b: morphological compounding: V and O form a single word

Type II: manipulation of case is possible so that V+N unit
can function as a trasnitive unit with an oblique argument advancing
to DO status
Type III: manipulates discourse strucutre, such that known or
incidental information can be backgrounded by means of NI
Type IV: classificatory NI: N narrows the scope of the verb
such that a more specific object NP or part thereof can
also appear

Mithun proposes implicational hierarchy: lng that has Type IV will also


have all the others, lng that has Type III will have Type I and Type II.

``Persian has only Mithun's type I NI in that it does not allow


an oblique arguent ot be advanced intot he case position vacated by
the incorporated noun..." [WRONG!]

(aside: Mohamad and Karimi who treat compounds like dars xAndan


(compositional)
to be fundamentally different
than compounds such as rang zadan "to paint.")

G and M do not make this claim. Instead they argue that


there is an aspectual diffrence with CPs being less bounded.

They argue against a movement analysis since a number of


the compound verbs do not involve DOs, they coudl
not be derived by movement since they would involve a non-properly
governed trace.  Other args against movement analysis are
given in Heny and Samiian and Mohammad and Karimi.

They do not treat CPs as lexical because of the -esh


clitic: "Since compounding is generally assumed to be derivational
as opposed to inflectional, it must occur prior to
affixation of enclitics.  Threfore if the Persian process under
discussion was an instance of lexical compounding, the enclitic
could not occur inside the compound." pg 23
Also, compounds can be separated by phrasal material. They cite:

CHECK THESE:


{Hasan maSin az in mard xarid \\
Hasan car from thi sman bought+3sg \\
'Hasan bought a car from this man.'
(Heny and Samiian 195:8)

{gush be man ne-mi-kon-e\\


ear toe NEG+PROG+DO+3sg\\
'She doesn't listen to me'\\
(Mohammad and Karimi 197:7)

M &K reference cited as having othe reasons against a lexical


approach.

Heny and Samiian propose a Reanalysis soln.


G& M's solution:

*!*Givon, Talmy. 1991. Isomorphism in the Grammatical
Code: Cognitive and BIlological Considerations. In Studies
in Language. 15-1 85-114.

Peirce:  ``In the syntax of every language there are logical icons


of the kind that are aided by conventional rules." (1940:106)

Contrasts this view with Chomsky's that arbitrariness is hte hallmark


of human languages.

! The Quantity Principle

A larger chunk of info will be given a larger chunk of code
Less predictable info will be given more coding material
More important infrmation will be given more coding material

Motivated by: cognitive economy: processing time, cognitive complexity,


mental effort.

E.g., have as main clause verb vs have as grammtical


marker (stress differnce, 've).

Size and Stress gradient of:

full NP $>$ indpendent pronoun $>$ unstressed pronoun $>$ zero anaphors

Stress on contrastive, contrary to expected info

Deletion under identity: Mary wanted to 0 leave

! Proximity Principle

Entities that are closer together functionally, conceptually,
or cognitively will be placed closer together at the code level,
i.e. temporally or spatially
Functinal operators will be placed closest, temporally or
spatially at the code level, to the conceptual unit to which they
are most relevant (cf. Mirror Principle)

\eenumsentence{ She let go of him


She let him go
She wanted him to go
She wished that he would go
She said, "He's gone."

[\eenumsentence{ So come talk to me


Come to talk to me
Come 'n talk to me
Come and talk to me ]

Proximity reflected in:


Non-restrictive relative clauses have a separate intonational contour.
Restrictive REL clauses have smae intonation contour as head
noun:
{The old woman, who I told you about before,...\\
The old woman with grey hair sat next to the old woman with black hair.

Operators that are conceptually relvant to the verb cluster near


the verb or cliticize on it: tense, aspect.

Focus markers go directly before the focused consituent:


{Only John met Mary \\
John only met Mary. \\
John met only Mary.

! Semantic principle of linear order:


THe order of clauses in coherent discourse will
tend to correspond to th etemporal order
of occurence of the depicted events.

[Give example of Durie]

Cause before effect; condition before entailment (Greenberb 1966, Haiman 1978)

Sometimes the terminative-perfective marker  (``finish") comes after


the verb, when other tense, aspect, modal morphemes come before the verb.

! Pragmatic Principle of linear order


More important or more urgent information tends to be placed
first in the string.

Less accessible or less predictable information tends omtbe placed


first in the string.

Clause initial indefinite or importnat full NPs in lngs


with flexible word order: Bagels, I like;
It's the bagel that's good.
What did he do?

[Note diffrence with Chafe or MacWhinney: starting points]


[to Tom: extra clausal position is different than first in
regular clause]

*!*Gleitman, Lila, Gleitman Henry, Carol Miller and Ruth Ostrin.


Similar and Similar Concepts. April 1995. The Instute for Research
in Cognitive Science. U of PA. IRCS Report 93-22.

I liked that you emphasized the role of the


construction and its interaction with lexical meaning:
if it's possible, the construction wins.  
That's true for other constructions as well.  E.g., the
ditransitive forces a construal in which "Chicago"
stands metonymically for certain people in Chicago in:

 They sent Chicago the letter.

(If you can't get an animate construal, the sentence is out:

 *They sent the border the letter. )

 I thought you might want to address Dowty's discussion
of symmetric predicates, since it I think it could be
used to support your point.  (Not that doing 5 experiments
isn't enough for one paper!)  He also argues that the
difference between the subject and object arguments
in e.g. The tree is near the bicycle
is not a lexical difference, but an effect of subjecthood.
"...there are clear asymmetries in meaning
brought about by interchanging arguments of be-involving
a difference in "perspective"-which we otherwise need not
attribute to be's lexical meaning, though we may need independently
to characterize the subject vs. nonsubject NP's in discourse." (pg506)
(I had his quote on line for another reason).

*!*Gleitman, Lila. 1994.   The Structural Sources of


Verb Meanings. In Language Acquisition:
core readings P. Bloom (Ed.). MIT Press.

Factors that pose challenges to children learning word meanings


blind children who are told to "touch but don't look at.."


a table woud merely bang or tap it, but if
told "now you can look at it" will explore all surfaces systematically
with her hands.

"look up!" the child keeps head immobile and searches space above


head.

Traditional idea: The blind child learns that look involves haptic


exploration because it is that verb which is
used most reliably in contexts in which haptic exploration is
possible and pertinent to the discourse.

But this cann't be right: look is used only 72\% of the time


when the relevant object was near enough to teh child for
her to explore it manually.  see only occured 39\% of the time
in such contexts. (compare put, give, hold which appeared
over 95\% of the time in such contexts.

Just as surprising, they knew green was only attributable


to phsyical objects (syntactic bootstrapping doesn't tell them this,
but coocurrence might).

They may well pay attention to more factors, but that doesn't


necessarily help.  "The trouble is that an observer who
notices everything can learn nothing" pg 181.

Gleitman compares word learning to syntax learning, noting that


the problem is at least as difficult.  Just as we are assumed to
have innate syntactic knowledge to help us, we need some kind
of constraints on word learning.

Golinkoff (1986): communicative episodes betewen mothers and


11-18 month old children: even in later stage, the mothe
only
immediately comprehends the child's desires about half the time.
Other times, the mother misunderstands or ignores his signals altogether.

! Multiply interpretable events: push implies move and


slide, clank, roll or crawl;

buy/sell, chase/flee, give/receive: are ``thrusts to teh heart


of the observational learning hypothesis"

Pinker (1987): ``the child could learn verb meanigs by (a) sampling


on each occasion in which a verb is used, a subset of the features..., (b)
adding to the tenative defn for the verb its current value for that feature,
and (c) permanently discarding any feature value that is contradicted by
a current situation." (54)

Gleitman: can't be right, because it assumes the mother and child are


always attending to the same event.

[Also, can't deal with polysemy: climb]


G: ``the verbs seem to describe specific perspectives taken on those events by the


speaker, perspectives that are not ``in the events" in any direct way.

Also, ``since verbs represent not only events but the intents, beliefs,


and perspectives of the speakers on those events, the meanings of the
verbs cannot be extracted solely by observing the events." (think,
hope, suppose, understand): children may not understand these well
at 2 or 3, but they do eventually learn them at 4 or 5 so G thinks
this is still a problem.

Subset problem: how does the child figure out duck instead of


animal/mallard. 

The basic level idea and no synonymy ideas together might help, but


she is skeptical of those ideas.  [Syntactic bootstrapping doesn't
help either]

Gleitman notes that if uses of look and see in


different constructions are distinguished, you can formulate the
haptic perception.  See Table 2.3.2.

See Table 2.3.3


Syntactic Bootstrapping:  the range of subcategorization frames has considerable


potential for partitioning the verb set semantically, and that
language-learners have the capacity and inclination to
recruit this information source to redress the insufficiencies of
observation.

She cites: laugh, smack, put as 1 2 and 3 argument verbs

She cites Zwicky 1971 on greem

also cites Bowerman for overgeneraliations

``no one can doubt that, at least sometimes, the context of use is
so rich and restrictive as to make a certain conjecture about
interpretation overwhelmingly likely"

```putting" logically implies one who puts, a thing put, and a place into which


it is put...In contrast, because one cannot move objects from place to place
by the perceptual act of looking at them, the ocasion for using look
in such a structure hardly, if ever, arises (*John looked
the ball on the table).'

``the component `transfer' is inserted into a verb's semantic entry in


case it is observed to occur in three NP structures. This happens for
/put/ but not /look/" pg 198.

the component `mental' is added to the entry for see (which


she takes to mean perceive, visually or haptically)  when it appears
as Let's see if there's cheese in the refridgerator (``since
events as well as entitites can be perceived") pg 198.

``each verb is associated with several of htse structures. Each


such structure narrows down the choice of interpretations for the verb." pg 198

She claims a virtue is that this procedure is categorical and not


probablistic.
``The child can take one or two instances of a verb
in some frame as conclusive
evidence that it is licensed in thi ssyntatic environment."

are the semantic/syntactic relations the same cross-linguistically?

Yes in GB (please).

Every NP in a sentnece must receive one and only one thematic role (theta criterion)

The theta criterion will hold at every level of derivation (projection principle)

BUT, N incorporation, PRO drop, difference in NP or PP.


Assumes ``at least some of hte mapping rules have to be in place before


the verb meanings are known, or else the whole game is over." pg 203

Infants parse by means of fundamental frequencies, silent intervals,


syllabic length [this would only be evidence that they can detect phrase
boundaries, not that they know what the phrases are or how they
are put together]

Golinkoff et al. (1987: adapted preferential looking paradigm


(Spelke 1982) for use iwth language.

Hirsh-Pasek (1985): 17 month olds, who knew few


if any verbs, lookd longer at BB is tickling CM than
CM is tickling BB.

Hirsh-Pasek et al. 1988: BB turns, CM turns BB


Unknown verbs were also used with 2 year olds (27 months)
on BB is flexing with CM, BB is flexing CM

Brown (1957): some dax, a dax, daxing

Katz, Baker and Macnamara 91974): a gorp vs Gorp

Naigles (1990): 24 month olds: preferential looking paradigm with


novel verbs The rabbit is gorping the duck; The rabbit and
duck are gorping

:these show the zoom lens effect

Older children:
Fisher et al. (1989): looked at 1) number of args (1 or 2);
structural positions (ride/carry); structural positions and
preps (give/take).

*!*Green, Georgia and Jerry Morgan. 1996.  Auxiliary Inversions and


the notion
`default specification'.  Journal of Linguistics 32:43-56.

Green argues that auxiliary inversions to not require reference


to a default feature, and that more generally, it is not
necessary to stipulate a default value when languages
exhibit different values for a feature in different contexts.

Inverted structures are more diverse structurally than is generally recognized,


and can be captured in a constraint-based grammar with monotonic multiple
inheritance and no overridable default specifications.

(HPSG does not allow global feature-specific default specifications.)

Default1: (Chomskyan): stipulates a property that inheres in UG,
e.g. a positive value on PRO-drop (Hyams 1983).

Default2: (e.g. as used in GPSG): a (learned) language specific property;


features, not grammars are marked or unmarked

A sign is an abstract, structured object with phonological, synsem


and context attributes.  Words and phrases are subsorts of signs.
(A sort is defined by which attributes it has and possibly what
values those attributes have.  Sorts are organized in a (monotonic) multiple
inheritance network).

Construction Grammar's ``construction" $\approx$ ``sign"

``The fact of the matter is,....that while the univerted structure
is supposedly the typical, functionally neutral, unmarked
one for clauses as a whole, the inverted structure is the typical,
functionally neutral, unmarked one in clauses with the illocutionary
force of questions." pg 48

Defaults are relative to a subset of structures.

Aux inversions occur in questions, exclamations, if-less
conditionals, negative existentials, and some ``extraction'' constructions.
(see paper for attributions)

\eenumsentence{ Will the next four months be like the last three? (question)


God have I seen attitudes change!  (exclamation)
Should I leave this job or go to the bathroom I risk being fired. (conditional)
Didn't nobody teach me this. (AAVE) (negative existential)
Not a bite did she believe he had eated. (extraction, negative filler)
So gorgeous was the spectacle on the May morning...that the crowd,
waiting in hushed, black clad awe, could not keep back gasps of
admiration. (extraction, comparative filler)
Thus did Illini hammer Indiana a third straight time.  (extraction, deictic filler)

A closer look at conditionals (from Green 1985):


\eenumsentence{ Were we in Boston, we coud go to Fenway Park.
Had I been in Boston, I would have visited them.
Should he leave early, I'll let you know.
*Will I be in Boston, I will buy one.
*Was I in Boston, I would have bought one.
*Can I be in Boston, I will go there.

Negative Fillers (from Green 1985 see also Horn 1985):


\eenumsentence{ Not a page did he turn.
Not the page did he turn.

Klima (1964), Ladusaw (1982): inversion is correlated with semantic/pragmatic


property: same as polarity licensing [which is also actualy very complicated!]

Green notes that in GPSG, semantics of constructions follow from general


rule of combination and are predictable [in current HPSG, this is
no longer necessarily true].  %She notes the possibility of  a special
%f-INV combinator.

Constructions that require a [+INV] feature must specify it (it is often not


a property specific to particular words).

E.g. Negative-topic filler-head clauses Not a soul did I see specify [INV]


Lexical items which can be used to reference or implicate questions allow


the complement clause to be inverted:
\eenumsentence{ Someone must know is he sick.
They never found out did anyone stay awake.
Those which can't think, after could subcategorize for [-INV]
complements:

 think [SUBCAT $<$ NP, S[-INV]$>$

 after [SUBCAT $<$ S[-INV] $>$

Or, the property could be stated as a sort declaration on a subtype of


lexical entries.

Conclusions:


 Facts are quite complex

 Features apply to construction types as well as words

 Syntactic features like [+INV] correspond imperfectly to semantic


features.

 Generalizations often apply across a subgroup of constructions not


across all of the grammar.  These can be stated simply in multiple
inheritance hierarchy by having subtypes inherit from supertype
with the relevant feature.
General architecture of HPSG is analogous in many ways
to that of Construction Gammar

*!*Green, Georgia. 1985. The Description of Inversions in GPSG.


BLS 11.

*!*Green, Georgia. 1994.  Assessing TEchniques for Analysis


of Natural Language Use.  Cogsci Scie TEch Repoert UIUC-BI-CS-94-08

Argues that both corpus and introspective data are useful.

*!*Green, Georgia. 1996.  Modelling Grammar Growth;
Universal grammar without innate principles or parameters.
U of Illinois. ms

*!*Grimshaw, Jane and Sten Vikner.  1993. Obligatory Adjuncts


and the Structure of Events. In Knowledge and language
eric Reuland and Werner Abraham (eds). Kluwer Academic Publishers.

They look at passives that require the by adjunct:


\eenumsentence{ *This house was built/designed/constructed.\\
This house was built/designed/constructed by a French architect.
*Tomatoes are grown.\\
These tomaties are gorwn by organic farmers.
By phrases are not the only means by which to make
short passives acceptable:
\eenumsentence{ This house was built yesterday/in ten days/
in a bad part of the town/only with great difficulty.
The best tomatoes are grown in Italy/organically.

We know these are adjuncts and not arguments because they


are highly variable in form and semantic type, they are clearly
adjuncts in other contexts (particuarly yesterday, in ten days, etc, passive by is treated as an adjunct by many theories, and
obligatoriness of the adjunct is affected by the progressive
and perfect, and we don't expect argument structure to vary
under tense/aspect differences.

Verbs which require obligatory adjuncts are passive, never


active ``constructive" ACCOMPLISHMENTS.  Acommplishments have two subevents,
and in the passive, the theme only "serves to identify" one
subevent (the state).  The activity is not identified.
This is true of verbs of creation (draw, knit, dig) and
change of state verbs (cook (a turkey), paint (a house),
fix, freeze, broil, fry, saute, develop (a film)).
[Are the change of state verbs supposed to be out as short
passives?]

G&V note that verbs of destruction destroy, kill, shoot,


ruin, break, arrest never require obligatory adjuncts, however.
That is, they make fine short passives.
So do Dowty's "verbs that create a performance object" record,
transcribe.
G&V account for this by claiming that the process as well
as the state
destroy is "identified" by the theme argument in the case of
``destroy", but not ``build."

G&V note that the process component of build can be


identified by an expression that specifies manner, time, duration,
place or purpose of the process or reason for the process:
{This example was constructed by two linguists/yesterday/in 10 minutes/in Geneva/with difficulty/to prove our point/in order to show that we are right/using an IBM/for a good reason. pg 147.
Adverbials which have no relationship to the event do not save the
constructions:
\eenumsentence{ *This example was probably constructed.
*Fortunately, this example was constructed.

G&V note that a state reading is possible for \ex{0a-b:


the event structure for these states is simple:

[$_{event$  state ]

They note that this reading is generally only possible with
a contrastive reading:
\eenumsentence{ The example as (probably) constructed.
??The bridge was (probably) constructed.
The paper was written.
They note that Danish makes a morphological distinction between
stative passives and eventive passives.
Un seems to rescue a poassive with no obligatory
adjunct from ill-formedness:
\eenumsentence{ The paper was unwritten.
The film was undeveloped
"It does not seem likely that the negative can identify
a sub-event...they are not well-formed as events, but only
as states.

Obligatory adjunct is not necessary in perfect and progressive


``at least for some of the constructions discussed above."
\eenumsentence{ *This film was developed.
  1   2   3   4   5   6   7   8   9   10


Cơ sở dữ liệu được bảo vệ bởi bản quyền ©hocday.com 2016
được sử dụng cho việc quản lý

    Quê hương