1998: Annotated Bibliography
Adele Goldberg
*!*Afarli, Tor. 1987. Non-Subject Pro-Drop in Norwegian. LI
18 2: 339-345.
Element can be missing from the second conjunct of a coordinate verb phrase:
Jens hugg ved og stablet op
Jens chopped firewood and piled (it) up.
"She seized them and put (them) in the oven."
The construction is acceptable in most places except southeastern Norway.
Also exists in Icelandic.
The omitted element need not be a DO:
"just I take a newspaper and read in (it)" (example 4)
may occur within subordinate clauses:
"some workers who hauled snow from the streets and tipped (it) into the sea."
*!*Ahrens, Kathleen. 1995. The Mental Representation of Verbs.
UCSD Dissertation.
Discusses Shapiro's (1987) claim that it is hte number of arg struc
frames that was the crucial metric in determining the representational
copmlexity of the verb, not the maximal number of arguments
within any particular frame.
Schmaumder (1991) failed to replicate. Shapiro et al. (1991) did
replicate.
KA points out that not all subcat frames and arg structures were
included for each verb. return is classified as non-alternating
but it is listsed as alternating in the Oxford Advanced Larner's
Dictionary. Fix me a sandwich was not included for fix,
secured her the tickets was not included for secure.
There wree only 5-6 verbs per class so these types of errors were
very problematic.
KA's chart is on page 69ff. When length and frequency is controlled
for as best as possible, there is no difference between verbs
with two subcat frames and those with 3 subcat frames.
KA shows that the participant role complexity metric does
account for the difference (although it boiled down to 12 ms,
it was statistically reliable).
KA shows that 2-participant role verbs are faster to integrate
into a sentence than 3-part. role verbs. Both verb types
integrate fastaer into the 2-fronted-arg sentence than into
the fronted adjunct+arg sentence.
*!*Allen, Shanley E.M. 1999? MS Learning about Argument Realization in Inuktitut and
English: Graduate development in the use of non-ellipsed forms. Max Planck
Institute for Psycholinguistics
Children tend to use lexical NPS for arguments they consider most informative
and to ellipse those argumetns they consider redundant (Greenfield \& Smith 1976;
Clancy 1980, 1993, 1996; Givon 1983; Chafe 1987; Du Bois 1987; Hirakawa 1993)
Examples of what counts as inforative are cited (including contrasting args,
referents not present in physical context, answers to specific queries, newness
in discourse, etc.)
*!*Allen, Shanley E.M. and Heike Schroder. to appear (2000?) Preferred Argument Structure in Early Inuktitut Speech Data
In
J DuBois, L Kumpf and W Ashby,
Prefered Argument Structure: Grammar as Architecture for Function. Amsterdam: Benjamins.
Full lexical NPs appear relatively frequently in the S and O roles, but
relatively infrequently in the A role.
*!*Anderson, John R. 1991. The Adaptive Nature of Human Categorization.
Psychological Review Vol 98 N 3. 409-429.
Origins of categories:
linguistic labels
feature overlap
similar function
Three views need not be in opposition.
A focuses on categorization of living objects: calculation
assumes objects are only classified one way.
Coupling parameter: c: fixed probablity c that two objects come from the same
category (not dependent on the number of objects seen so far).
P(k) = cn$_k$ / (1-c) + cn
n$_k$ is the number of objects assigned to the category k so far
n is the total number of objects seen so far
For large n, this closely approximates n$_k$/n, which means that there is a strong
base rate effect with a bias to put new objects into large categories.
P(0): = probablity that the new object comes from an entirely new category.
P(0) = (1-c) / (1-c) + cn
For large n this closely approximates (1-c)/cn, which means that the prob ofa brand
new category depends on the coupling probability and the number of objects seen so far:
the greater the coupling probability and the more objects, the less likely it is that
the new object comes from an entirely new category.
[could be used to explain the stabilization of language]
Empirical findings: review
Central tendencies
the reliabilty with which an instance is classified decreases as a function
of its distance from the central tendency of the category.
Reed (1972): subjects categorized 10 faces into two categories:
their categorization varied with distance of the face from the prototype
Rational model: identified two or more categories,
depending on the presentation order.
The categories chosen in original experiment were
preserved as separate categories by the model.
Influence of particular instances
Medin and Schaffer (1978); Nosofsky (1988) : subjects were trained to
classify 12 colors that varied in brightness and saturation.
Showed a sensitivity to the manipulation of the frequency of a stimulus.
[abstracting away from the different mathematical implimentations]
Shepard (1987)
Anderson (1991): training on one stimuls generalizes to other stimuli as a function of
similarity and of the size of the original category.
*!*Barddal Johanna, 1999. The Dual Nature of Icelandic Psych-Verbs.
Working Papers in Scandinavian Syntax 64. 79-101.
*!*Barddol, Johanna. Ms 2000. Case Assignment of Nonce Verbs in Icelandic.
Argues for importance of isolate attraction, citing Goldberg and Sethuraman and Osherson.
She cites Cruse and Croft (in prep book on Cognitive Linguistics: chapter 10, 12) as observing
that low-frequency
constructions often exhibit low degrees of productivity (not none).
that this is expected because high type frequency leads to more general or schematic
constructions while high token frequency leads more specific or substantive
constructions as entrenched.
*!*Barddal Johanna, 1999. Case in Icelandic--A Construction Grammar Approach. To appear in Tijdschrift Joorykandinavistiek.
Some confusion about my analysis of negative evidence :
"..the learner gradually distinguishes the different
meanings of different constructions and infers
that if a given verb is not used in an optimal construction,
then such a usage is not warrented." (pg 5)
Narrowly defined subclasses are discussed as a second learning
mechanism.
913 verb uses were collected, most of them novel.
39 sentential patterns.
Borrowing from English provides cases of many novel verb uses.
Cluster attraction (Goldberg 1995) and Isolate attraction.
[V+st \'a]
(e.g., Netast \'a, ``to write to each other
on the internet)"
is cited as a productive construction
but -st used to have reflexive meaning and
\'a may mean to(?).
Boltast um is cited as novel, "to move around heavily"
but it too may be compositional ( um = "around"?)
Isolate attaction: new verb formed analogically to one
existing verb with the same or similar meaning.
The forms are given without glosses, page 15, so I can't
tell what exactly the analogy is.
JB suggests that the argument structure of the target language
is sometimes borrowed, citing t\'ekka inn (``check in"),
brenna \'ut (burn out); but since these are complex predicates
in English, it could be said that normal lexical borrowing
is involved.
Is generous about my lack of informative discussion about cases
in Constructions
Most novel verbs take nominative subjects, accusative direct objects.
Many take accusative prep objects as well.
A few (7) take dative subjects, many take dative objects and many
take prepositional objects.
There are also 3 accusative subjects, 2 genitive objects in the
database.
*!*Benmamoun, Abbas. Ms 2000. The Role of the Imperfective Template in Arabic Morphology.
Abbas argues that Vs are not created by combining roots and templates,
but rather from word to word mappings (more like English).
" If the active form has the vowel /u/ initlaly, the last vowel of hte stem is
realized as /a/. For this analysis to work, however, one must know what the
vowel melody of the active verb is, a possibility that is not allowed
under the root to template account" pg 7
McCarthy and Prince (1990) and McCarthy (1993) also suggest a partially
word to word derivation, but their basic form is hte perfective.
Abbas argues the base form is the imperfective (it also has hte widest distribution, he notes).
*!*Bowerman, Melissa. 1996. Argument Structure an Learnability: Is a Solution in Sight?
BLS 454-468.
Novel causatives appear around age 2, flourished ...between 3-5, and then continue at a lower
level until 12, when they end. (C: 225 tokens, 79 types; E 92 tokens, 54 types). errors
are summarized in appendix.
Errors with come, go, disappear, stay are unlikely to involve incorrect word meanings
as Pinker had proposed.
One shot innovations for lack of better alternative: stay used causatively 43 times,
long after she knew and usually used hte more appropriate verbs keep and leave.
Go used causatively 28 times long after she knew verbs take and send.
Sometimes the erroneous causativiszation did not set in until well after the corect form had
already been established in the child's speech.
B argues that C and E routinely violated semantic constraints on the construction, causitiviziting
non-dynamic verbs (like be), when there was no "act" (with be, climb), and
when the causation was indirect: I want to watch you this book! Everybody makes me cry...you just cried me.
(Most of the latter errors involved dolls. But the rarity of occurence of these forms with
"make someone V" meanings was arguably due to a lack of opportunitites to make such errors. That
is,they seldom say "make me laugh" OR "laugh me".
So, we need to explain how children in fact do retreat from a causativizing operation that is
overly general.
Preemption: `it has been proposed that the child's causative might be preempted by the corresponding preiphrastic causative, e.g. make disappear. Bu the extension of preemption to these
cases seems somewhat dubious (Bowerman 1988): lexical and periphrastic causatives are, as constructions, systematically associated with different meanigs, soa child sould not readily allow one
to be supplanted by the other." pg 461 [but see Brooks and Tomasello: they DO preempt]
For C: forms with and without suppletive counterparts declined largely in parallel
Role of semantic subclasses: If the induction of semantic categories is important in children's
retreat from causative overgeneralizations, semanticaly distant verbs should decline faster than semanticlaly close verbs. But
errors with verbs of volitional action, emotional expression and psychological events (all of which
violate teh directness constraint) hold their own over time against semantically close verbs.
*!*Brown, Penelope. 1998. Verb specificity and argument realization in Tzeltal child
language. draft 1998
Three issues:
whether semantic factors also play a role in argument representation
verbs outnember new noun types: effect of verb specificity?
[X, Gelman and Shatz]
Is there any evidence that children initially prefer 'light verbs'
with more general meanings or light meanings for their specific verbs?
[discuss role of frequency]
*!*Butt, Miriam and Tracy Holloway King. ms-2000. Null Elements in
Discourse Structure.
Cites B. Hoffman 1995 Penn dissertation on variable salience of backgrounded
information. The computational analysis of the sntax and interpretation
of 'free' word order in Turkish.
In Hindi: topics appear intially, foci immediately before the verb, backgtrounded material post-verbally
Only continuing topics and backgrounded info can be dropped.
New topics cannot be dropped:
She will live life according to her own wishes. *[I] was wrong.
"Most of the instnaces of continuing topic drop" involve the 1st person pronoun, but they note that this may be an artifact of the discourse contexts
they looked at. *Verb agreement is not necessary*, since verbs in Hindi
do not agree with non-nominative subjects and yet those subjects can
still be dropped.
More than subjects can be dropped:
"Becuase man dopes not ever make a sacrifce for a woman. nor will (he)topic
ever make (a sacrifice for a woman)backround"
*!*Bybee, Joan. 1995. Regular Morphology and hte Lexicon. In
Language and Cognitive Processes 10 (5), 425-455.
Source-oriented lexical schemas: relate two forms, e.g., present and past tense forms ( wait-waited)
product-oriented lexical schemas: relate surface forms ( strung, stung, flung, hung)
high token frequency words: have high lexical strength: easy to access, serve as the bases of morphological relations and exhibit an autonomy that makes them resistant to change and prone to semantic independence.
The autonomy is also said to lead to weaker connections to other words: are learned on their own, not in relation to other terms.
" Forms of high token frequency will be more autonomous and more likely to be unanalysed, and less likely to participate in schemas; high token frequency forms will thus not contribute to the productivity of a pattern (Bybee 1985; Guillaume 1927/1973; Moder 1992)" (pg 434)
In fact, she argues that high token frequency can actually detract from productivity.
Moder (1992: SUNY Buffalo dissertation): subjects were less likely to produce nonce verb forms with irregular past-tense when primed with a high-frequency prime as compared iwth a low frequency prime. [did all have a family resemblance?] Again, the idea is that high token frequency leads to storage in unanalysed form.
She speculates that it was the incrase in frequency of went that led
to its ability to separate from wend in order to become the past tense of go
Productivity of a schema depends on: 1) the defining property of the schema (how "open" it is) and 2) its strength (type frequency).
Within paradigms, words with higher lexical strength serve as the basis for the formation of new words (Bybee 1985) [what does this mean?]
If patterns have an open schema, they gradually increase in type frequency.
German past participles (clahsen and rothweiler 1992): productive affix is t, despite, they claim the fact that it doesn't have higher type fequency than en. BUT, bybee points out, they count uses with productive separable prefixes as different
types (the equivalent of write down, write off, write on).
German and Arabic plurals (Marcus et al. 1993)
German -s plural an Arabic "sound" plural have open schemas (are relatively unrestricted), but have low type frequency. Bybee argues that they are indeed default or "emergency" morphemes, but that they actually have lower productivity than English -s or -ed, since most often another plural form is used. Moreover their use IS affected by existing lexical itema.
Grm
-en is used with Fem nouns, -e is used with masc nouns, and er with neuter nouns.
-en has hte highest type frequency, and children generalize it the most freuqently. Fem loan words also use it.
-s was used productively most often in nounds ending in a full vowel, and environment that almost always takes -s in the German lexicon (janda 1990).