lektsii Toma Bever'a
- Московские лекции известного американского психолингвиста Томаса
Бевера состоятся 9 ноября в 16:00 в РГГУ (ауд 206, главного здания) и
10 ноября в 16:55 в МГУ на ОТиПЛе. Напоминаем, что для прохода в РГГУ
нужен пропуск, для получения коего просим обращаться в Центр
типологии (973-47-55, Вера Мальцева).
Аннотации лекций см.ниже.
Two Lectures on the Psychology of Language.
Thomas G. Bever
University of Arizona
The two talks are loosely connected by a common model of language
behavior on which acquisition, comprehension and production all integrate
local, discourse and statistical information with structural computations.
The first lecture reviews the history of 20th century psychology of language
primarily in America, including the last seven decades of structural linguistics
and associated psycholinguistics: it emphasizes the bilateral interaction
between theories of language structure and theories of language behavior,
and presents several examples of language universals which can be
explained as having a perceptual/cognitive basis.
The second lecture focuses more directly on a bi-phasic comprehension
model and evidence for it, with special attention to evidence that verb event
structure is accessed very early in sentence comprehension. Each lecture
is self-contained, and can be understood without the other.
The past and present in psycholinguistics (RGGU)
Psychological research and theory can contribute to understanding language: (1) It specifies how
language acquisition and behavior works in individuals. (2) It clarifies which properties of language
are unique and which reflect other mental structures. (3) It clarifies which properties of cognition
are unique and which derive from the existence of language in mind .
In the 1950s, contact with Jakobson and his student, Halle, convinced Chomsky to give up his own
behaviorism: he enriched the Transformational/ Generative model, now interpreted as a rationalist
mental model of linguistic knowledge. G. Miller started research on the “Psychological Reality of
Transformational Grammar,” a new rationalist/nativist . This phase lasted for 15 years – an exciting
period in which it seemed that linguistic theory could be directly confirmed by psychological data.
But there were several problems. First, real behavior is a blend of structural knowledge with
acquired habits. Second, there was no explanation of how or why innate knowledge becomes
reformed in each individual child into rules of hiser language. Third, Generative Syntactic Theory
kept changing. Nonetheless, there are several consistent ideas across all models: sentences have
computational derivations. This sets a conundrum: (1) Sentences are horizontal (unfold in time)
(2)Derivations are computationally vertical (span entire sentence structures).
An analysis by synthesis (AxS) model of language resolves this puzzle: a statistical pattern-based
phase accounts for the horizontal aspect of sentences, a computational phase accounts for
the derivational structure. Children acquire linguistic constructions in such phases: they alternate
between relying on computational structures and statistical generalizations, which give the child
a way to understand sentences for which it does not yet have a computational analysis.
This model has important consequences for language universals. (1) it offers an explanation
of the “derivational opacity” constraint - a given representational level cannot be inferred
from a more superficial one. Discovering “derivations” is characteristic of natural problem solving
in humans: On this view, derivations present puzzles, FUN TO LEARN. (2) A related prediction
is that every language must have a “canonical sentence form”, to provide a basis for the child’s
learning statistical generalizations. We can state this as the “Canonical Form Constraint” (CFC):
To be learnable, languages must have a canonical statistically dominant form.
The model may also apply to a structural syntactic universal – the Extended Projection Principle
(EPP) (every sentence has a “subject”): the CFC requires every sentence to have at least one
argument, with “subject” as default. Thus EPP is not a syntactic universal, but a behavioral
THE CAUSAL DIRECTION BETWEEN COGNITIVE AND BEHAVIORAL UNIVERSALS: KTO
These ideas are examples of how psychological theory can offer an explanation of particular
universal properties of attested languages. The potential virtue of such explanations is that
they narrow the focus of our search for “true” innate linguistic universals. There is evidence
that the AxS system is typical of human (and possibly animal) perception. Hence, its application
to language offers an explanation for certain linguistic properties, as opposed to the reverse.
It will take more information and more thought to resolve these puzzles. The overarching motive
in this paper has been to outline perspectives on the possible explanatory relations between
psychological and linguistic research.
We understand everything twice – The Comprehension role of Verb Event Structure
Language comprehension presents a puzzle: discourses are serial in time, but their
structure and meaning are timeless mental objects. A sentence is immediately understood
as it is heard serially, but understanding it involves pairing local superficial information with
an internal representation of great complexity. We have suggested that listeners actually
integrate the two kinds of knowledge in two phases of an analysis-by-synthesis framework:
an initial preliminary assignment of meaning is based on local information and statistical
strategies; a subsequent confirmation of the meaning is based on a complete assignment
of syntactic and semantic structures. That is, we understand everything twice. The
reason we do not normally notice this is that the second representation “absorbs” the initial
representation, and we perceive it as a unified and vivid mental object.
This model makes a number of surprising predictions – notably, that full syntax is assigned late,
and certain abstract semantic structures are assigned early. A range of existing experimental facts
support that prediction: these include probe studies showing that NounPhrase trace is accessed
after WH trace; ERP studies showing that local phrase violations are detected quickly, together
with semantic violations, but global syntax violations are detected slowly.
A new prediction of the model is that lexically based semantic structures should be accessed early
in processing. In the experimental section of this talk, I concentrate on a new series of studies
which demonstrate the immediate perception of verb event structure (AktionsArt): in particular,
we show that verb telicity (having a goal, or specific event) is immediately accessed in the initial
phase of comprehension, prior to accessing verb syntactic subcategorization information.
The studies utilize the “reduced relative clause” effect to probe for the immediate activation
of telicity. The paradigm is:
VERB TYPE VERB SENTENCE
Telic, transitive: arrested the soldier (who was) arrested by the sailor protested
Atelic, transitive: chased the soldier (who was) chased by the sailor protested
Telic, potentially intransitive: tripped the soldier (who was) tripped by the sailor protested
Atelic, potentially intransitive: fought the soldier (who was) fought by the sailor protested
Reduced relative clauses have a common mis-parse when the initial portion of the sentence
could be analyzed as a simple declarative sentence. This is often exemplified in the sentence:
“The horse raced past the barn fell”, in which “the horse raced past the barn” is mis-parsed
as a main clause with “the horse” serving as agent rather than patient. Telicity and transitivity
both signal that a patient is semantically/syntactically required. This will tend to reduce the
misparse of the initial portion of reduced relatives as a simple active construction. Thus, the
difference in processing complexity between the unreduced relative (with the phrase “who was”)
and the reduced relative clause, indicates whether telicity or transitivity has been accessed from the
initial N+verb sequence: the smaller the difference, the greater evidence for accessing
We have re-analyzed previous studies of self-paced reading, as well as our own studies; we have
also run a study with eye-movement recording, and also an auditory complexity study. All studies
are consistent with an early access of telicity, certainly by the “by phrase”: the effect of transitivity
is later, at the main verb. These results are consistent with the AxS model, again reflecting the idea
that locally available semantic information is accessed prior to syntactic structures.