Thanks to visit codestin.com
Credit goes to www.scribd.com

0% found this document useful (0 votes)
3 views41 pages

04 Discourse Processing

Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
3 views41 pages

04 Discourse Processing

Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 41

Theories of Language Comprehension: India and Beyond

Dr Jayashree Aanand Gajjam

Centre of Excellence for Indian Knowledge System,


IIT Kharagpur

ELECTIVE COURSE
SPRING 2025

1
Discourse Processing

- Introduction - Models of Discourse Processing:


(i) Construction-Integration Model
(ii) Structure Building Framework
(iii) Even Indexing Model
- From Syllables to Mental Matrix:
(i) Working Memory
(ii) Anaphora Resolution
(iii) Process of Inferencing - Neural Basis for Discourse Processing

2
Discourse Processing: Introduction
Discourse= ‘Connected Sentences’
✓ Originally, it means ‘conversation’ or ‘dialogue’.
✓ Literal Meaning: ‘running back and forth’
✓ Discourse is any sustained linguistic interaction: a paragraph, a chapter, a 500-page novel, a face-to-face
conversation, a letter or an email, essays, stories, lectures, meetings, debates, instant messaging sessions,
OR simply a text or a narrative.
✓ It is a larger linguistic unit of collocated, structured, and coherent group of words.

✓‘Coherence’ is the central theme that binds the related sentences together into a discourse.
Thus, a discourse makes sense, and gives a single idea.

NOTE: Discourse Processing in Psycholinguistics draws insights from sociolinguistics, anthropological linguistics, and philosophy of language.

3
Discourse Processing: Introduction
Intrinsic Features of a Discourse:
1. Position: Opening and ending sentence
2. Order: There is an order between the events presented in the sentences.
3. Adjacency: Sentences are more meaningful due to the nearby sentences.
4. Context: Sentence meaning in discourse can be understood only in the context.

4
Discourse Processing: Introduction
Example:

"There are no colored bathrooms in this building or any building


outside the West Campus, which is half a mile away. Did you know
that? I have to walk to Timbuktu just to relieve myself. And I can’t
use one of the handy bikes. Picture that, Mr. Harrison. My uniform,
skirt below my knees, my heels, and a simple string of pearls. Well, I
don't own pearls. Lord knows you don't pay the colored enough to
afford pearls! And I work like a dog day and night, living on coffee
from a pot none of you want to touch! So, excuse me if I have to go
to the restroom a few times a day.“
- Hidden Figures

5
Understanding Discourse
1. Discourse goes beyond understanding mere words. Although, understanding individual words and
individual sentences is an absolutely necessary prerequisite.
2. A hearer takes the meanings of the individual sentences and integrates them into a coherent
framework by discovering the links between and among the meanings of the sentences.
3. The information is thus represented mentally in an elaborate matrix of existing knowledge, attitudes,
and emotional biases we might have about the topic/ writer or speaker/ the reason we are reading or
listening to it.
4. This requires an intricate orchestration of number of different processes such as storing the
information in the working memory, finding references for anaphoric elements, and building
inferences…
5. Thus, Discourse processing is an active cognitive processing.

6
Understanding Discourse:
Working Memory

7
Understanding Discourse:
Working Memory
Working memory= Short term memory
Stores information briefly (but longer than sensory memory).
Working memory provides a temporary space to retain this information (which is known as Working
Memory Span).
Duration: About 18 seconds

8
Understanding Discourse:
Working Memory
Sensory input → Working Memory → Long Term Memory

Coding in Working Memory:


1. Acoustic coding = letters are converted into acoustic codes,
2. Semantic coding = words are represented based on their meanings, and
3. Visual coding = preserves spatial characteristics.

Working memory plays a key role in discourse processing, as it provides a platform for all the ongoing
computations that perform different operations while processing a sentence (such as pre-lexical
recognition, lexical access, integrating lexical units into syntactic structures…).
Working memory helps in obtaining the basic building blocks of sentence meaning.

9
Understanding Discourse:
Working Memory
What keeps working memory active? 2. Chunking:
People can remember approximately 5 to 9 bits of
1. Rehearsal:
information for short periods of time.
Mental repetition or mental ‘practicing’ - George Miller (1956) ‘The Magical Number Seven, Plus Or Minus Two’
(e.g. talking to oneself)
• Keeps working memory active,
He calls these bits ‘chunks’ = a group of meaningful items.
• Prevents information from destroying.
These units are grouped into a single meaningful whole =
‘chunking’

The bigger the chunk, the greater the capacity.

10
Understanding Discourse:
Working Memory
Working Memory to Long Term Memory:
When all the individual sentences are processed, a reader now needs to retain and integrate them. Then, he
associates that information with extra-linguistic information. All this is stored in Long Term Memory. It helps
in putting individual sentences to use.

11
Understanding Discourse:
Working Memory
While the information goes to long term memory, three things happen (Sachs, 1967):
1. Information about individual lexical items (words) and clause structure is lost. Only meaning is retained.
2. Individual sentences are lost. Only their integrated meaning is retained. (People recall the gist, not the
surface form. Human memory works fine in integrating and synthesizing the information, but not at
keeping individual bits of information. Individual sentences are used only to create more global
representation of meaning. )
3. Inferences are added to the representations or meanings.

Exception: When sentences carry high interactive content or great interpersonal import, even the syntax is
retained.

Ex: “Can’t you do anything right?”


Ex: “After all this time?” “Always!”

12
Understanding Discourse:
Anaphoric References

13
Understanding Discourse:
Anaphoric References
Ex:

14
Understanding Discourse:
Anaphoric References
Anaphor= An anaphor is a linguistic device that refers to someone or something that has been mentioned
in the previous context.
An anaphor can be either a pronoun or a definite noun phrase (a noun phrase introduced by a definite
article [such as ‘the’]).
Ex:
‘John went for a ride across the woods. He has been doing it every Sunday. He also visits a Church after.
Meeting with family and friends in the evening really cherishes him. It has been his routine for the last 35
years.’

Co-referencing = Locating a referent/ antecedent for the anaphor. It is an important task in Discourse
Processing. It helps a reader build a connected, semantically and pragmatically coherent mental
representation.

15
Understanding Discourse:
Anaphoric References
How does a reader resolve an Anaphor?
To resolve an anaphor, a reader uses both linguistic and non-linguistic knowledge.
‘John and Mary went for a ride across the woods. He has been doing it every Sunday. She wanted to visit a
Saint Anthony’s Church instead. It is famous for its ambience and musicals. But it was close due to the
maintenance issues which deeply disappointed her.’

A reader pays attention to the Discourse Focus. It is achieved by:

1. Recency (He → John) (She → Mary)


2. Long-term focus throughout the text. (instead= instead of riding)
3. Main Characters: here, John and Mary
4. Proper Names: here, John and Mary, Saint Anthony’s
5. Position in a prior sentence: especially, the subject in the previous sentence.

16
Understanding Discourse:
Anaphoric References
Some examples:
1. Joe removed the diskette from the computer, and disconnected it.
Joe removed the diskette from the computer, and copied it.

2. Vincent gave a book to Sody. Kim also gave him a letter.


Vincent gave a book to Sody. Kim also got a letter from him.

3. Ambiguous: Jenny went window shopping yesterday and spotted a nice cup. She wanted to buy it, but she
had no money with her. The following day, she went to the shop and bought the coveted cup. However, once
back home and in her kitchen, she put the cup on the plate and broke it.

17
Understanding Discourse:
Inferencing Process

18
Understanding Discourse:
Inferencing Process
Inferencing= Drawing the information from the text which is not explicitly stated, but is derived from the
knowledge of the listener and the information coded in his mental representations.
The formation and storage of inferences is a central feature of discourse processing. Inferencing is ‘core of
the understanding process’ (Robert Schank, 1976).
Ex:

‘My daughter just got married!’ ‘Do you like him?’ (him= son-in-law)
‘We went to a wedding.’ ‘The woman wore white.’ (woman= bride)
‘Go get bread packet from the market. If they have eggs, buy six of them.’ (them= eggs)
‘Hey! Pay attention to the baby. It’s eating the dust from the shoe.’ ‘Throw it away!’ (throwing= shoe)

19
Understanding Discourse:
Inferencing Process
A reader adopts a Minimalist view while Inferencing. This suggests that very few inferences are drawn
during the actual process of interpretation. These inferences are constructed by automated mental
processes. According to this view, inferences will be drawn under only two limited conditions:

1. If there is a need to establish a cohesion between two adjacent parts, and


2. If the information is quicky and easily available.

20
Understanding Discourse:
Inferencing Process
▪ A reader uses logic, sociocultural norms, and the real-world knowledge of the interlocutors to infer the
meaning. These are known as ‘Scripts’ (Schank and Abelson, 1977).
Scripts= general scenarios.
The more closer people are socially and culturally, the more ‘shared scripts’ they have.

Inferencing helps a reader to integrate the upcoming new information into his growing mental
representation, i.e., Global Discourse Structure. It is done more successfully if a reader is familiar with the
topic.

21
Understanding Discourse:
Inferencing Process
Some types of inferences:
1. Causal: The delicate bowl fell off the shelf. Inference: It broke.
(Determining the actual effect of a particular phenomenon.)

2. Bridging: The car smashed the truck. Inference: The windshield shattered.
(Objects/ events in the sentences are connected in a meaningful way.)

3. Instrument: The woman stirred her coffee. Inference: She used a spoon.
(About the tools and methods used in the event.)

4. Elaborative: Dave ate four pounds of crab. Inference: Dave likes crab.
(Bridging the current text idea to prior world knowledge.)

5. Goal: Susan left early for the birthday party. She stopped at the mall on the way. Inference: Susan
wanted to buy a present.
(Determines the purpose of a certain event.)

22
Understanding Discourse:
Inferencing Process

The general consensus is that:


- bridging inferences are routinely drawn,
- elaborative inferences are rarely drawn, and
- causal and instrument inferences are drawn under very limited circumstances.

23
Discourse Processing:
Theories / Models

PSYCHOLINGUISTICS (ELECTIVE, SPRING 2023-2024, IITKGP) 24


Discourse Processing:
Theories / Models
There are primarily three prominent theories or models of Discourse Processing:
1) Walter Kintsch’s Construction-Integration Theory
2) Morton Ann Gernbacher’s Structure Building Framework
3) Rolf Zwaan’s Event Indexing Model

PSYCHOLINGUISTICS (ELECTIVE, SPRING 2023-2024, IITKGP) 25


Construction-Integration Theory
According to this theory, a reader interprets a text by building three
Discourse distinct kind of mental representations.
1. Surface Model
Processing
2. Text Base
3. Situation Model
Walter Kintsch’s
Construction-
Integration Model
(1998)

26
Construction-Integration Theory
1. Surface Model (Linguistic Representation)
- It captures the exact words in the text along with their syntax.
- The least abstract mental representation (= clearer).
- Provides the inputs to build the second model. A reader does some
work on that input. He then outputs a set of propositions
represented by the surface model.

Proposition= a predicate and its arguments (verb and actors)


Proposition= the smallest unit of meaning that has truth value.

Word by word information may be retained for a short span.

27
Construction-Integration Theory
2. Text-base (Propositional Network)
-The mental representation that describes the propositions.
- It is close to verbatim form of text itself but it does not include the
precise wording of the original text- some surface information is lost.
- It may include some information that was not explicitly mentioned
in the text.
Ex:
Lack of toilet facility, the lady being a black person, has to walk far,
many times a day, does not get paid much, humiliated, alienated,
expressing disappointment and hurt….

28
Construction-Integration Theory
3. Situation Model (Elaborated Propositional Network)
-The highest level of abstraction (=lack of exact words)
- It describes the ideas/ events that the text is about.
- It is a mental simulation that captures a number of different
features of the real or imaginary world.
- It contains information directly stated in the text + inferred
information.

29
Construction-Integration Theory

Discourse A reader:
Processing 1. creates a set of
productions that
manipulates the
content of working
Walter Kintsch’s memory, and
Construction- 2. builds a coherent,
structured model
Integration Model representations,
(1998) 3. that can be stored in a
stable form in his long-
term memory.

30
Construction-Integration Theory
The ultimate goal of the
Construction-Integration system is Thus, a text is processed in cycles.
to build a Situation Model that can
Construction: In each cycle, comprehenders input a few propositions’
describe the relevant aspects of worth of text. Knowledge associated
what a text is about. At this state, with the inputted text becomes automatically activated.
the model uses the general world
knowledge. Integration: In this phase comprehenders connect new propositions
to previously processed propositions, draw inferences, and update
their situation model. Situation model contains ‘Tokens’ or Mental
Experimental Research: Place-holders: for characters, objects, and concepts previously
introduced by the speaker.
Participants never made mistakes
about the Situation Model, they
made some mistakes about the The situation model is the strongest and longest lasting
representation, the text-base is the next strongest, and the surface
specific Propositions in the story, form is the weakest and shortest lived.
and they often made mistakes
about the specific wording of the
story. (The surface form
representation is fairly weak and
decays very rapidly.)
31
Construction-Integration Theory

Try!
Check your understanding of a certain movie/ specific episode in the
movie.

Check your understanding of a news item.

Check your understanding of a course content/ specific topic/ specific


chapter.

32
Structure Building Framework
Model:
Discourse 1. Laying a foundation:
Processing To understand a story, readers lay a foundation. The foundation is
based on the information that arrives first. Just like when we build
a house, laying a foundation is the first thing.
2. Mapping and Shifting:
Morton Ann Two additional processes are used to continue building the
Gernbacher’s structure: mapping and shifting.
The mapping process connects the incoming information to the
Structure Building foundation as long as it is relevant, and coherent with the earlier
Framework information. If not, then the readers build a new substructure by
process of shifting. The branching substructures terminate at the
foundation.
Thus, foundation and appropriate number of connected
substructures- represent any mental representation of a story/
picture.
Professor at the University
of Wisconsin-Madison, USA

33
The model applies to both Structure Building Framework
verbal and non-verbal (pictures Summary:
etc.) materials.
The structure building framework says that we use general purpose
According to Gernbacher, the cognitive mechanisms to process and understand discourse. It is
processes that are responsible related to the idea that human mind is based on the connectionist
architecture.
for Discourse Processing are also
responsible for other cognitive - Whenever the new information is coming, some memory nodes are
tasks. activated.
- Initial activation forms the foundation of mental structures.

According to this theory, the - Once the foundation is laid, all the upcoming input is either mapped
goal of comprehension is to to it (if it activates same memory nodes), or is shifted to make a new
build coherent mental substructure (if it activates different memory nodes).
representations= structures. - Once the memory nodes are activated, they send processing signals
to enhance or suppress other memory nodes’ activation. When the
information represented by them is necessary, they are enhanced.
When the information represented by the memory nodes is not
longer necessary, their activation is suppressed.

34
Structure Building Framework
Try!
Check your comprehension of a movie consisting of different short
stories.

35
Event Indexing Model:
According to this model, the purpose of the discourse comprehension
Discourse system is to understand the “goals and actions of protagonists and
events that unfold in the real world or some fictional world.”
Processing To represent these story elements, five core aspects of stories are
tracked, and each event in the story is indexed (or tagged) according to
each of the five core features:
Rolf Zwaan’s Event 1. The time frame over which the event occurs (time),
Indexing Model 2. The characters that are involved in the event (protagonists),
3. The causal connection of the current event to preceding and
following events (causation),
4. The spatial location(s) where the events occur (space), and
5. How the event relates to a protagonist’s goals (motivation).

36
Event Indexing Model:
Similar to the structure building framework, the EIM conceptualizes
events as activated memory nodes.
The representation of a story consists of a set of memory nodes and the
connections between them.
Each memory node is coded for the five features, and as each new
piece of the text is processed, it is evaluated as to how it relates to
previously activated memory nodes.
So, each time a new piece of text is processed, the comprehender
updates the situation model to reflect the information provided by the
text. Different pieces of text can require updating of different features
of the event index.

37
Event Indexing Model:
Some cues that a reader uses:
➢ Virtual proximity: Objects that are in close virtual proximity to a currently
focused character have a higher degree of activation.
Some researchers have
therefore proposed that a ➢ Temporal Iconicity: A reader indexes the events in order that they are
critical element of discourse described.
coherence and discourse ➢ Mental Flip: If Temporal Iconicity is violated, a reader flips the order of the
understanding is the events mentally. It uses working memory processes.
discovery of the ➢ Emotional States: A reader also indexes emotional states and personalities of
causal structure of the the characters and uses them to infer characters’ responses in the next events.
story.
➢ Tracking Goals: A reader tracks multiple goals of multiple characters in
narratives. This is a sophisticated process. A character is always a central to the
narrative. A reader also recognizes when the goals of different characters are in
conflict.
➢ Causal Connections: A reader applies a causal connections between
propositions. It is the glue that binds the pieces of a narrative together. Also
known as, cohesion. It contributes to textual coherence.

38
Yujie Wang, Damminda
Alahakoon, and Daswin De
Silva , 2018

39
Discourse Processing:
Neural Basis
Some Evidence:
1. More coherent text produces different brain wave activity than less coherent texts.
2. Left hemisphere: what sentences mean (semantics). It is involved in describing causal connections,
especially when it requires a bridging inference.
3. Right hemisphere: establishing a coherence between sentences, drawing inferences (coherence). It is
more activated when text lacks cues for coherence. (Right Superior Temporal Lobe)
4. Patients with dementia exhibit impairments in anaphora processing. Their speech lack cohesive
sentences. They drop some crucial components in their speech. Usually some other components from
other narratives intrude their speech.

40
Discourse Processing:
Neural Basis
Some Evidence:
Working memory in discourse processing:
Left frontal and parietal lobes
Maintenance, monitoring and
manipulations in the working memory.

41

You might also like