Presentation Abstracts
Invited Presentation
A single discourse marker can
have multiple parallel functions at different levels of the speech
exchange, such as describing a causal event and labelling the
causal structure of the speech exchange itself; this multi-levelled
"polyfunctionality" is also characteristic of gesture
(Kendon, 1993; Sweetser, 1997). This paper examines differences
and similarities between the ways that spoken and gestural markers
regulate discourse structure. It argues that these contrasts largely
follow from the characteristics of spoken conventional language
and more iconic, less conventional gesture (conventional signed
languages will be briefly brought up as well). Interestingly,
some of the systematic parallelism between the two "channels"
lies in parallel use of shared metaphorical structures which systematically
structure both language and gesture (cf. Cienki, 1996, 1997; Sweetser,
1997).
Invited Presentation
A major issue for discourse studies
over the last three decades concerns the ways in which literacy
influences language use: Are there systematic linguistic differences
between spoken and written language that can be associated with
literacy as a technology? This paper argues that the written mode
provides the potential for styles of linguistic expression not
found in the spoken mode. Evidence for this claim will be presented
from a series of Multi-Dimensional studies of register variation:
synchronic, diachronic, and cross-linguistic. These studies conclude
that there are few (if any) absolute linguistic differences between
spoken and written language - rather, particular types of speech
and writing are more or less similar with respect to different
underlying 'dimensions' of variation. At the same time, these
studies show that there are important differences in the potential
range of linguistic variation found in each mode, with writing
permitting a much wider range of linguistic expression than speech.
PAPER SESSION 1
This study used an adaptation
of professional actors' learning strategy (previously employed
only for theatrical dialogue) for the acquisition of both expository
and narrative material by college students. Results showed that
students using this strategy (called active experiencing) retained
more of the essential content of each idea unit (and as many of
the exact words) as students using an intentional memorization
strategy. These results would indicate that active experiencing
can be of substantive benefit in typical college learning situations.
Readers who self-explain text
aloud understand more from the text and construct better mental
models of the content. This presentation summarizes two experiments
conducted to examine the benefits of providing extensive self-explanation
training, or practice, to middle-school children and adults. Effects
of prior knowledge and reading skill were also examined in relation
to the benefits of self-explaining and self-explanation training.
The benefits of self-explanation depended largely on prior knowledge.
The purpose of this study was
to investigate the recent view of long-term working memory during
reading proposed by Ericsson and Kintsch (1995). It is assumed
that adult readers can readily retrieve the episodic structure
of a text from long-term memory via the mediation of reinstated
content retrieval cues. A reading interruption procedure was used.
Cue sentences were inserted after the interruption but before
reading resumption. The features of the cue sentence were manipulated.
Is background knowledge solely
descriptive, or is there room for a perceptual component? The
indexical hypothesis suggests that perceptual components are crucial.
On this hypothesis, words and phrases are indexed to objects.
Then, affordances derived from the objects, not the words, are
used to guide the interpretation of the language. In fact, participants
given the opportunity to index read subsequent text more quickly
and followed instructions more accurately than those who could
not index.
PAPER SESSION 2
We use the Switchboard Telephone
Speech corpus to investigate the discourse-pragmatic contrast
between two fronting constructions: Topicalization (TOP) and Left
Dislocation (LD). Our data reveal that LD performs a topic promoting
function (Lambrecht 1994) for uniquely identifiable referents
(Gundel et al. 1993). TOP is not topic promoting, and is much
less constrained with regard to the activation status of the fronted
denotatum. While our study generally supports Prince's (1984,
1986) account of TOP, it provides a revised account of the functional
opposition between TOP and LD--one based on complementary rather
than inclusive functions.
This paper analyzes the use of
reference forms in an English telephone conversation between a
teenage girl and an adult friend. The analysis focuses on the
speakers' exploitation of these forms to negotiate their stances
toward and "ownership" of the topic of the conversation,
a favorite soap opera. It is argued that many of the referential
choices in the conversation are motivated by these stance-related
considerations, and that the constraints on referential choice
cannot be described without considering such socially situated,
"natural" data.
This paper investigates sentence
and attention unit length, independent constructions, types and
tokens, word order effects, and other linguistic attributes
cross-linguistically
and analyzes them quantitatively and qualitatively. Specifically,
it demonstrates how linguistic attributes lead to hypothesis about
conceptual organization and reflect cost of processing. It adds
to the literature in that it uses the descriptive tools of Cognitive
Grammar to analyze a corpus of 20 English and 20 Brazilian Portuguese
expository texts.
Narrative representations in creole
discourse, which are computed by the right hemisphere of the brain,
are realized by five external units (quintuple-hierarchy) that
consist of three internal components (trinary branching). Creole
languages also demonstrate preference for a specific numerical
set of coherency-unit representations (numbering-preference).
This study suggests that these common features were derived from
the creolization process when the core part of discourse grammar
was being established according to universal principles.
PAPER SESSION 3
Spanish has been characterized
as a flexible word order language, in the sense that a particular
construction may exhibit different word orders. Constructions
formed by a conjugated verb, an argument np, and an adverb, occurring
in informal conversations are studied. There is a correlation
between word order and syntactic, semantic, cognitive, and pragmatic
factors, and factors pertaining to the general organization of
discourse. There is also a correlation between primary stress
placement and pragmatic factors.
This study tested the effect of
referent accessibility on pronominal anaphor processing when the
referent is located far away in the surface structure of the discourse.
More specifically, we investigated eye movement behaviors during
initial processing and re-processing of anaphoric sentences. According
to the memory-based text processing view proposed by McKoon, Gerrig,
and Greene (1996), automatic retrieval processes rely mainly on
the organization of the information encoded in memory.
This study deals with the time
course of lexical context effects on access to word meaning. In
three lexical decision experiments, a context formed by a word,
a homograph, and a target word expressing the homograph's dominant
meaning were presented in succession. The context word and two
temporal factors were manipulated. Immediate facilitation of response
time was noted when the context word was related to the target,
suggesting that access is initially constrained by lexical context.
This research examines the nature
of the associations in semantic memory when a reader views a word
or a picture. Since Collins and Quillian, Psychologists suppose
the existence of a semantic network for words into which propagation
of activation can occur. In our study, the existence of a semantic
network for pictures as it exists for words is examined. In three
experiments of association generation and semantic priming, we
compared the activation of semantic representations by written-words
and pictures.
SYMPOSIUM
The participants on this symposium have
been developing and testing AutoTutor, a fully automated computer
tutor that simulates dialogue moves of normal human tutors. There
is a tutorial dialogue of several turns as AutoTutor and the learner
collaboratively answer questions and solve problems. After the
learner types in the content of a turn, AutoTutor selects dialogue
moves that assist the learner in the active construction of knowledge.
The dialogue moves include immediate feedback, pumping, prompting,
splicing, hinting, elaborating, requestioning, and summarizing.
The tutor's moves are delivered by a talking head with appropriate
facial expressions and synthesized speech. This symposium will
demonstrate AutoTutor, describe its mechanisms, and report tests
of its performance.
AutoTutor simulates the dialogue moves of an unskilled human tutor. AutoTutor "comprehends" what the human learner types into the keyboard by using language modules, segmenting the content into speech act categories, and using Latent Semantic Analysis to evaluate the truth, relevance, and quality of student contributions. AutoTutor uses curriculum scripts to guide the tutorial dialogue at a macro-level. At the micro-level, appropriate dialogue moves are selected in a tutor's turn and are expressed by a talking head.
There will be a demonstration of AutoTutor. The demonstration will illustrate the range of contributions that typical learners type into the keyboard. The alternative dialogue moves of AutoTutor include the delivery of curriculum script subtopics, questions, immediate feedback (positive, neutral, and negative), pumps, prompts, elaborations, hints, splices, and summaries. Graphics, animation, synthesized speech, and the talking head will also be demonstrated.
A curriculum script was developed for 37 subtopics on computer literacy. In addition to an initial seed subtopic, there were 12 subtopics on computer hardware, 12 on operating systems and 12 on the Internet. The subtopics vary in difficulty and have one of four formats: Question+Answer, Didactic-information+Question+Answer, Graphic-display+Question+Answer, and Problem+Answer. The composition of these subtopics are described.
Language modules analyze the content of the message that the learner types into the keyboard during a particular conversational turn. The sequence of words and punctuation marks in a turn are segmented and classified into speech act categories (i.e., Question, Contribution, versus Short Response). The language modules include a lexicon, a connectionist network that identifies the syntactic classes of words, software agents implemented as "codelets" that sense surface linguistic features, and a connectionist network that formulates predictions about the next speech act category.
Latent Semantic Analysis (LSA, Landauer & Dumais, 1997) was used to represent the knowledge of computer literacy. Two books and 21 additional articles were compressed into space of 300 dimensions using LSA. This high-dimensional space is included in computations that evaluate the truth, relevance, and quality of the learner's contributions during tutoring. This presentation describes LSA, the corpus of texts on computer literacy, and the computational methods of evaluating learner contributions.
Latent Semantic Analysis (LSA) provides the representation that is needed for evaluating the meaning of learner contributions in AutoTutor. The LSA match scores are used to compute the truth, relevance, and quality of a learner's contributions during tutoring. It is important to assess the fidelity of these LSA scores in the knowledge domain of computer literacy. We compared LSA "match scores" for sentences and paragraphs that varied in length and quality.
After a subtopic is introduced by the tutor, there is a collaborative interaction between AutoTutor and the learner in an effort to answer a question or solve a problem. The dialogue moves of the tutor include immediate feedback (positive, negative, neutral), pumping, prompting, elaborating, hinting, splicing, requestioning, and summarizing. These dialogue moves were analyzed in a corpus of naturalistic tutoring protocols.
AutoTutor's dialogue moves within a turn are determined by a set of fuzzy production rules. Multiple rules are associated with each move category. The production rules are tuned to (1) the truth, relevance, and quality of the learner's recent contributions, (2) the phase of the tutorial dialogue within a subtopic, and (3) the learner's ability, initiative, and volume of verbal output. AutoTutor's generation of dialogue move categories has a close fit to the dialogue move patterns of human tutors.
Most of AutoTutor's dialogue moves are produced with synthesized speech and a talking head. The immediate feedback phase of the tutor's turn has facial expressions and intonation that signify positive, negative, or neutral feedback. The parameters of the facial expressions and intonation are based on a corpus of feedback expressions produced by human tutors. The extended information after the immediate feedback is also delivered with a talking head, but without precisely tuned evaluative parameters.
PAPER SESSION 4
We present a theoretical/computational
account of the cognitive processes during reading and of the role
that these processes play in the gradual emergence of a stable
episodic memory representation of the text. The presentation consists
of three parts. First, we describe the unique features of the
model. Second, we demonstrate the model's properties by projecting
a simulation onto the overhead screen. Last, we explore some of
the model's unique predictions and compare these to empirical
data.
In our research program we are
investigating whether information about social exchanges and the
detection of cheating behavior may be equally important for the
construction of situation models during text comprehension as
causal, temporal and spatial information. By performing causal
analyses of various scandals, a particular structure of cheating
behavior was identified in three-party interactions. With corresponding
texts, which were designed according to experimental criteria,
we investigated in three experiments under which conditions a
cheating inference is drawn by a reader when reading such a text.
In this study, we investigated
the multidimensional content of a situation model constructed
from a text in which spatial information are foregrounded. More
specifically, we examined what information relative to the characters
of a story readers preferentially integrate. As we expected, the
results to inferences judgment tasks indicate that readers not
only monitor character's spatial location but also information
relative to personality traits, emotional reactions, and intentions.
The effects of need for cognition,
text coherence, and prior knowledge on readers' situation models
were investigated in two experiments using different texts. Participants
were given original texts or their high-coherence revisions. Analyses
of answers to open-ended questions showed readers of high-coherent
text learned better in both experiments. In one experiment, there
was a text version x need for cognition interaction; low-nfc participants
learned better from revised text and high-nfc participants learned
better from original.
PAPER SESSION 5
Native speakers of three different
languages (Latvian, Mandarin, or English) rated collections of
their language's idioms for how transparent, mappable, or opaque
they perceived the relationship to be between the idioms' literal
and figurative meanings. In four experiments, native English speakers
classified these idioms according to their figurative meanings.
Response times and error rates indicate that speakers' ability
to interpret idioms depends on whether the phrases are Normally,
Abnormally, or Un- Analyzable. Based on these results, I argue
that universally-held conceptual metaphors motivate idiomatic
language in general, with further differentiation made based on
the cultural and linguistic subtleties unique to specific language
groups.
As predicted by the graded salience
hypothesis (Giora, 1997), 'less familiar' ironies were processed
only literally initially - 150 msec after their offset, regardless
of contextual bias. However, 1000 msec after their offset, the
ironic meaning became available and the literal meaning was still
as active. In the literally biasing context, only the literal
meaning was available (see also Giora et al., 1998). In contrast,
'familiar ironies' revealed no significant interaction. Both their
literal and ironic meanings were similarly active in both types
of contexts and under both interstimulus intervals, as predicted.
Two obstacles in attempts to evaluate
the psychological reality of conceptual metaphor in figurative
language comprehension are the lack of extra-linguistic evidence
and the presence of more parsimonious accounts. To address these
obstacles, two extra-linguistic paradigms not usually associated
with conceptual metaphor--category accessibility and reading inferences--were
used to evaluate four accounts for the comprehension of proverbs:
conceptual metaphor, the standard pragmatic model, association
and conceptual base. The results supported the conceptual metaphor
view.
The extent to which comprehension
is affected by different types of explanatory introductions was
investigated. Findings suggest that introductions that are
"functional"
in emphasis facilitate comprehension of subsequently read texts
to a significantly greater extent than introductions that are
"mechanistic" in emphasis. These introductions were
found not to differ on obvious superficial characteristics but
rather on deeper or conceptually based characteristics. These
findings are taken to reflect differences in conceptual organization
and its influence on the processing of discourse.
POSTER SESSION 1
The present work focuses on the
stage of recognition of intertextuality in a text. Intertextuality
is an operative semiotic mechanism seen as a hybrid area between
semiotics and pragmatics; it is the way we relate textual instances
among themselves and the way we recognize them as signs which
make us recall areas of our previous textual experience or pre-text.
Every text contains tangible elements that acts as intertextuality
indicators which can be grouped according to a typology.
This research deals with the distinction
between on-line and backward processes in the updating of a situation
model. We assumed that readers use one of these temporal components
according to their knowledge specificity of the situation. The
results to on-line and backward tasks showed that readers with
specific knowledge preferentially update their situation model
in a backward way, compared to subjects with general knowledge
who used equally well the two temporal components.
In this study, we investigated
whether readers' prior knowledge on a spatial layout determine
the content of a situation model constructed from a descriptive
text. As we expected, our results showed that subjects with specific
knowledge process equally well layout and characters information
and constructed a more precise and a more available situation
model than subjects with general knowledge who focused more on
layout information.
The goal of this research was
to study the changes in knowledge structure novices and experts
have on a specific domain. Subjects had to perform a key-word
sorting task before and after reading a text either coherent or
not coherent in terms of the temporal-causal sequence of information.
Our results showed a differential effect of textual coherence
on subjects' knowledge structure and emphasize the importance
of the semantic nature of information.
Three experiments examined memory
for sentential metaphors ("Playful monkeys are clowns")
and similes ("Playful monkeys are like clowns") presented
auditorily either with or without a meaningful discourse context.
Results using both recognition and recall showed metaphors to
be remembered better than similes, with the latter helped only
slightly by the presence of a meaningful context. Concrete sentences
were recalled and recognized better than abstract ones, with or
without the discourse context.
Two experiments investigated the
effect of prior knowledge of orientational metaphors (e.g. MORE
IS UP) on text comprehension. Participants read texts that were
consistent or inconsistent with MORE IS UP. Reading time for a
critical sentence was slowed for inconsistent texts. For texts
describing horizontal arrangements, no evidence of a preferred
mapping was found. The data suggest that readers' knowledge of
orientational metaphors can influence text comprehension.
In this paper the use of the concept
'situation' (eventualities and thematic roles) is discussed. It
will be claimed situations should replace the concept of 'proposition'.
The role of situations will be presented in a theoretical discussion
of a semantic representation of linguistic information, of language
evolution and language acquisition, and in a discussion of the
results of a connectionist model of text summarization and the
results of an eye movement research.
This study investigates the properties
of events in a memory representation as a network. We evidenced
using a recognition priming task on pairs extracted from narratives,
that the connection strength has a greater influence on the representation
of events in memory than the connectivity strength. Moreover,
we showed that a global activation of the network (general primes)
can slow down memory retrieval of information compared to a local
activation, even with non-causally related primes.
Some verbs, such as "blame"
and "confess", implicitly impute causality to one or
other of the participants in the actions they describe. Previous
studies have shown that readers have more difficulty processing
passages containing "because" clauses when the explicit
and implicit causes differ than when they are the same, but the
content of the "because" clauses also varied. In a series
of experiments, we used materials in which identical "because"
clauses were congruent or incongruent with different main clauses.
Although the previous effects of congruity on reading time were
not replicated, congruity did have other effects on other dependent
variables.
In this paper we focus on how
verb bias information facilitates clausal integration. We describe
how knowledge about likely causes (implicit causality) and consequences
(implicit consequentiality) affects the way in which locally ambiguous
relations between clauses are understood. We show that the language
processor is influenced by this implicit information independently
of the actual plausibilities of the described events.
Text comprehension of causally
complex texts were examined in two experiments. Materials outlined
two protagonists who had independent subgoals and a shared superordinate
goal. The superordinate goal could only be satisfied when both
subgoals were satisfied. In experiment 1, the reading times of
spillover sentences were longer when they followed previously
satisfied subgoals compared to when one of the subgoals was unsatisfied.
In experiment 2, this finding was replicated, and then reversed
when the position of the subgoal was brought closer to the target
sentence region.
Three experiments were conducted
to determine (a) whether adults make inferences about fictional
characters' emotional states both when they are and are not expected
to write continuations to stories, and (b) the extent to which
adults make such inferences automatically, particularly under
the no-continuation task conditions. The results of two reading
time experiments and one divided attention task experiment indicate
that adult readers do make emotion inferences and do so automatically,
both under elaborative and non-elaborative task conditions.
The role that reactions play in
adult readers' understanding of story characters' emotions is
explored. Goals, s/u outcomes and reactions were found to be considered
the most important story grammars, however emotional reactions
as a unique category were considered the single most important
story grammar. When emotional reactions were absent from the narratives,
the adult readers were found to increasingly rely on goals to
understand the characters' emotions. The presence of emotional
reactions increased the readers' perceived intensity of anger
in aversive episodes and sadness in loss episodes.
The instrument inference is one
of the kinds of elaborative inferences. Although most research
on inference showed that elaborative inferences were not generated
on-line during comprehension, they have several methodological
problems. This study was conducted to investigate whether instrument
inference was generated during comprehension using various on-line
measures. Lexical decision time and reading time data provided
convergent evidence for on-line generation of instrument inference.
The results were consistent with the prediction of constructionist
view on inference.
In this research, we studied the
role of semantic associations provided by a predictive text in
the generation process of a prediction. Readers had to perform
an on-line lexical decision task or a plausibility judgment task
at two different delays. Our results suggest that semantic associations
increase the generation degree of the prediction, but must add
new causal relations between the text content and the prediction,
to increase its integration degree in the final representation.
This paper raises three theoretical
insufficiencies with the research on inference generation. First
I question if measure of comprehension of narrative texts is satisfactory
considering that the primary function of narratives seems to be
entertainment, affective responses and pleasure. Secondly, I claim
that knowledge based inferences are given far too little attention.
Thirdly, I maintain that the number of on-line inference types
in narrative comprehension is impossible to set if based only
on studies of reception. The text itself and the narrational strategies
employed are paramount influentials and have to be carefully considered.
Since some of these issues are studied within the cognitive strands
of cinema studies, this might be a good partner for psychological
studies of narratives.
Culture affects recall of cultural
texts (e.g., Reynolds, Taylor, Steffensen, Shirey, & Anderson,
1977) and moral judgment development affects recall of moral texts
(Narvaez, 1998). Does culture affect the online processing of
moral texts? Participants took an inventory of individualism-collectivism
and read stories about requests for help in which the protagonist
did or did not help a relative. Results indicate significant differences
in individualism-collectivism and in moral inference generation.
This study was designed to examine
student writers at different levels of instruction, to describe
their cognitive cooperative strategies as shown in the cooperative
method. A teaching-learning module was designed within a cooperative
context and then was experimentally applied to sixth-grade students
(10-12 year-old children). An experimental group (who worked cooperatively),
and a control group were given the same tasks behaviours so as
to detect their composing process. In this paper the application
stage is revised.
The aim of this research was to
conduct an exploratory study of the connections between reading
and writing of argumentative texts, based partly on van Dijk and
Kintsch's theory (1978; 1983), van Dijk (1985, 1990)and important
contributions from Kucer (1985), Reuter (1994, 1995) , Eisterhold
(1994) and Nystrand (1990). It is argued that the processing of
information in reading and writing must share some psycholinguistics
strategies, knowledge, Two tests were administered to a group
of students . One on reading comprehension and the second on composition.
An exploratory study of different variables was made using canonical
correlational analysis. Specific relations are examined and research
implications are drawn.
Twenty writers planned aloud for
five minutes, then wrote a short narrative on an assigned topic.
We expected that more developed writing plans and more causally
connected plans would result in more plan elements being included
in the final story. However, on a relative basis, more developed
and causally connected plans had fewer plan elements in the final
story. Yet, individual story elements with more causal connections
did appear more often in the final story.
In this study, the effects of
questioning during and after reading on memory were compared to
a recall without questioning. The results indicate that questioning
during reading increases proficient readers recall but decreases
the comprehension of less proficient readers. Furthermore, questioning
after reading not only does not benefit proficient readers but
also may decrease young readers‚ comprehension. The effects
of questioning depend on timing and readers abilities, and they
do so by directing readers‚ attention to specific text information
targeted by questions.
The importance of causal structure
has been well-documented in text comprehension research. This
study investigated how both easy and difficult texts can be improved
using the causal network theory and how these repairs can differentially
impact comprehension for more- and less-skilled readers. Results
indicated that all readers, especially less-skilled, benefited
from the revisions but only for the difficult text. The causal
network theory provides an appropriate and systematic method for
revising texts.
Fifty-nine third through fifth
grade poor comprehenders but good decoders were assigned to three
conditions: a visually-based remediation, a verbally-based remediation,
or an untrained control. Thirty-six of the students were further
split into groups that crossed low and high visual skills, with
low and high verbal skills. The results indicated that an effective
comprehension remediation program must take into consideration
the student's profile of cognitive strengths and weaknesses. In
particular, programs designed to remediate weaknesses are more
effective than those designed to build on strengths. Thus, a student
who is weak in visual skills but strong verbally should be
remediated with a program that
strengthens visual skills.
The experiment explored the role
of working memory capacity on higher-level text processing. Readers
studied an outline, read a text, and then answered questions based
on the text information. Working memory effects were obtained
for inference and problem solving questions such that a larger
working memory capacity corresponded to improved performance.
No working memory differences were obtained for text based questions.
It was concluded that working memory plays a role in higher-level
text processing.
Psychology and geography specialists
were asked to search psychology and geography texts in order to
answer four content-related questions. Search time and search
patterns showed a limited influence of discipline on students'
online search strategies. However strategies were consistent within
question types and participants. Participants had a better incidental
memory for the structure of the document in their specialty. The
data support a model of document search as an generalized process
with a limited influence of domain-related knowledge.
This study examines the influence
of literacy, international, national and local television news,
and hard vs. soft nature of news on recall. Subjects were asked
to watch a fabricated news program and assessed on their ability
to recall the news. Results indicated that (a) literate audience
recalls better, (b) hard and national news are better recalled
than soft and local news, (c) the difference between hard and
soft news is smaller among illiterate women.
This study aimed to identify the
suppression mechanism as a basic process involved in the relationship
between working memory capacity and text comprehension in 4th
grade children. First, the contribution of working memory capacity
to reading comprehension was established with correlational methods.
Second, using the Hartman & Hasher's (1991) paradigm, differences
in the ability to suppress previously relevant information were
investigated in good and poor comprehenders who differ in working
memory capacity.
The present experiment was performed
in order to evaluate the influence of various factors in sentence
comprehension by young readers. It appeared that 1) the syntax
played no significant role; 2) complex grapho-phonological structures
delayed correct sentence understanding, and 3) sentence comprehension
was primarily dependent on pragmatic factors directly related
to the children world knowledge. These data will help to develop
a model of sentence comprehension by young readers.
Invited Presentation
University of Sussex, UK
In this talk, I consider the difficulties
of children who have problems with reading comprehension, even
when they are competent at single-word recognition. I shall discuss
the relative contribution of several theoretically relevant skills
and abilities to the prediction of children's reading comprehension.
The aim of a current longitudinal project is to assess which skills
and abilities might play a causal role in the development of reading
comprehension. I shall present some recent data from this project,
and discuss the implications of the findings for our understanding
of children's problems in text comprehension.
PAPER SESSION 6
In previous work we have argued
that an adequate theory of discourse should explain the fact that
the similarity between coherence relations varies. A classification
based on concepts like polarity (positive-negative) has been proposed.
If these concepts indeed have a cognitive status, then the order
of acquisition of coherence relations should reflect the classification
principles. This hypothesis was tested in an investigation of
primary school children's utterances and in an experiment using
a sentence completion task.
In this study, we examine how
children use various dimensions of coherence in constructing narrative
texts for younger students. Dyads of middle school children saw
an animated video and then constructed a printed text. We report
on dyads' use of prosodic (e.g., rhyme, repetition) and causal
dimensions in the printed texts, their production strategies,
and quality ratings of their texts. Results suggest that children
use multiple dimensions of narratives, including the prosodic,
to construct coherent representations.
We examined the effects of text
organization, introduction and delay on 6th and 4th graders' comprehension
of a text describing a controversial protest. The students read
the text and then answered comparative and integrative questions
immediately and after a one-week delay. There was a main effect
of grade on both types of questions, but the argument version
did not improve students' answers to comparative questions. Moreover
a rhetorical introduction interfered with the comprehension of
the source version. The relevance of causal-temporal vs. other
sources of coherence are discussed.
This research investigated the
effect of occurrence and co-occurrence frequencies of concepts,
computed from a corpus, on knowledge organization. Our main hypothesis
was that knowledge organization of three groups (Students, 7th
grade and 8th grade) were calibrated to inputs they received from
textbooks. This hypothesis predicts that knowledge organization
could be accounted by quantitative factors such as occurrence
and co-occurrence frequencies. A lexical decision task, an association
task and simulations with LSA model were performed.
PAPER SESSION 7
We investigated how temporal information
is understood and retained in situation models of narratives.
In Experiment 1, comprehension of temporal information was slowed
down and its retention impaired, if the information was inconsistent
with temporal order information given earlier. This was true for
absolute as well as relative temporal order information. Experiment
2 indicated that temporal inconsistencies increase reading time
even if readers are unable to report the inconsistencies.
This study investigated the role
of verb aspect and world knowledge in conveying the duration of
narrative events. Stories contained an aspect sentence describing
an event that was either completed or in progress and three post-aspect
sentences which could either occur concurrently or subsequently
to the aspect event. The perception that the aspect event was
still ongoing and its accessibility to working memory were both
assessed across the four test sentences. The results indicate
that readers rely on verb aspect to construct the temporal components
of a situation model.
The present study examined comprehension
changes across two readings of texts that described simple machines.
After each reading, participants drew and labeled the machines
that the texts depicted. Sentence reading times were predicted
by variables assessing the textbase, the situation model, and
subjects' memories for the text. The results indicated that readers
reread the texts strategically, allocating fewer resources to
proposition assembly and more resources to updating their situation
models.
The translation from a non-native
language into individuals' native language reveals the mediation
of situation models during text comprehension. Sixteen native
speakers of English differing in their level of mastery of French
translated French sentences into English, one sentence at a time.
The results suggest that situation models (1) are constructed
during skilled translation of texts and (2) facilitate translation
performance.
Invited Presentation
Elizabeth Bates
University of California, San Diego
In interactive, connectionist
or constraint-satisfaction theories of language processing, lexical,
grammatical and discourse information are used together, as soon
as they are available, to access words and construct sentences
in comprehension and production. This approach contrasts markedly
with modular accounts, in which each source of information is
evaluated (at least at first) within a separate processor. In
this presentation, I will review evidence from several different
languages in favor of the interactive approach, and I will also
show how claims about modular dissociations in aphasia can be
reconciled with the interactive view.
PAPER SESSION 8
Analysis of unwitting gestures
and co-occurring speech in videotaped, unrehearsed conversation
reveals the on-line origins, evolution, and organization of linguistically
communicative productions. Three Chinese and three English speakers
tell the story of a cartoon to a listener. Comparisons of the
meanings and synchrony of their gestures in relation to speech
reveal differences in how speakers of these typologically different
languages negotiate with the formal structures of their languages
to organize comparable narrative sequences.
We describe here a study on how
gestures are used as a resource in the course of articulating
one's knowledge. Our data consists of a corpus of videotaped interaction
of medical students participating in a Problem-Based Learning
(PBL) curriculum. Our analysis involves a number of issues including:
the form of the gestures observed, how gestures are located within
their material environment, how gestures are coordinated with
speech in ways that provide for their mutual interpretation, how
gestures are coordinated with other nonvocal behaviors, and how
gestures are used to organize social interaction whereby participants
negotiate mutual understandings.
This study investigates whether
prosodic characteristics of a text correlate with text structure.
The hierarchical structure of the texts was described by two theories.
The texts were read aloud; the duration of pauses and the pitch
of text segments were measured. There was a strong linear relationship
between pause duration and pitch on the one hand and level in
the hierarchy on the other hand. Prosodic characteristics play
a role in expressing the structure of the information and in packaging
the information in meaningful units.
PAPER SESSION 9
While most people consider hostile
electronic mail to be one of the hazards of the Internet, "flames"
are a boon to scholars wishing to better understand insulting
speech. We analyzed 1222 messages sent to controversial Web sites
and developed a successful set of syntactic and semantic rules
to automatically classify whether messages were flames. We discuss
our most and least successful rules, our methodology, and directions
for future research on text classification and computer-mediated
communication.
Essays provide a rich representation
of a reader's knowledge of a topic, but can be difficult to grade.
Expert graders graded essays using several criteria and we used
Latent Semantic Analysis (LSA) to automatically determine several
measures of the quality of the essays. Results indicated that
LSA graded as accurately as the human graders. A grading application
will be described which was used in an Psycholinguistics course
to provide students with instant feedback on the quality of their
essays and an indication of what key pieces of information were
missing from their essays.
To interpret natural language
at the discourse level, it is very useful to recognize dialogue
acts. We have implemented Transformation-Based Learning, because
this machine learning method has a number of advantages over alternative
approaches for computing dialogue acts. We extended and modified
our system to address the limitations of the original algorithm
and the particular demands of discourse processing, and our experiments
show that our system performs as well as two benchmark algorithms.
PAPER SESSION 10
Narrative as argumentation was
studied. A "prosecuting attorney" presented a narrative
stating how and why the defendant performed the crime. Four conditions
of narrative were presented, a baseline, and three conditions
of poorer narrative quality, as defined by historians' conceptions
of quality narratives. Judgments of guilt, confidence judgments,
and judgments of narrative quality and convincingness were lower
when the narratives were deficient in chronology/coherence or
causation. Narrativity as an argument form is discussed.
We investigated college students'
representation and use of source information when learning from
multiple texts. The importance and uniqueness of the items presented
in a pair of texts was manipulated. We found that readers were
able to identify the source of the propositions, and that both
reading ability and verbal reasoning ability were related to their
accuracy on this task. Representing source information and connecting
it to the situations described in texts becomes important as we
increasingly expect students to learn from multiple sources.
The effects of delay on essay
revision were investigated across four delay periods. At one and
two weeks, writers made more meaning changing revisions and essay
quality tended to be higher than at no delay or a three week delay.
Analyses of essay quality across delays showed that higher quality
essays contained more meaning than surface changing revisions.
Delayed revision may lead to better revision, and emphasis on
meaning revisions may account for increased quality.
PAPER SESSION 11
Films are also texts. This paper
tries to support this assertion by applying the constructivist
model (Graesser, Singer & Trabasso, 1994) to cinematic comprehension.
The features of coherence, explanation, situation model, information
sources and reader's goals are found to have equivalents in cinema,
although the contents and background knowledge of the processes
might vary. Although it is concluded that verbal and visual comprehension
overlap to considerable degree, an empirical investigation into
cinematic comprehension has to be sensitive to the specificity
of text and background knowledge. Throughout, excerpts from films
are used to support the argumentation.
Boundary crossing entails a shift
from one scope of activity to another, often with increased tension
or resistance, and unwelcome or adverse consequences. There was
considerable support for hypotheses that acts of boundary crossing
would intensify risk, heighten uncertainty, reveal deficient verbal
skills, and lead to signs of increased frustration and confusion
over what transpires. A climate of protracted disagreement and
misunderstanding produced a surplus of negative affect (anger,
sadness, fear) and more frequent complaints about performance
defects and critical appraisals than social settings where a climate
of agreement and understanding promoted a surplus of positive
regard (love, joy, surprise) and minimal boundary crossing issues.
This paper describes a "dialogic"
approach to analyzing autobiographical narrative. By drawing on
and systematizing Bakhtin's dialogic theory of narrative discourse,
the paper develops conceptual and methodological tools for analyzing
both denotational and interactional functions of autobiographical
narrative. It illustrates these tools with an analysis of one
life history interview. The analysis reveals a systematic parallel
between the events described in the narrative and the interactional
positions adopted by the narrator while telling that narrative.