Week 13: Discourse Processing

More discourse comprehension: Construction-Integration Model

A model of story (discourse) comprehension is the Construction-Integration Model. According to this model: Propositions from what we read in the text enter WM. We make inferences based on background knowledge/schemas and add these propositions to WM. We try to link together the propositions in WM in a coherent way.

Clinical considerations for reading

Different levels of representation and different amounts and types of inferences may be made depending on: The type of text The reader's goals / motivation The reader's reading proficiency The reader's background knowledge The reader's WM capacity

Why is discourse difficult to study in a controlled manner?

Discourse is difficult to study in a controlled manner because it involves high-level cognition. Discourse also depends quite heavily on prior knowledge, which varies greatly between individuals.

What is discourse?

Discourse is written text or speech that is at least several sentences in length. It includes conversation, lectures, stories, essays. Discourse is structured systematically.

In which groups may discourse production be impaired?

Discourse production may be impaired due to a variety of reasons, including: Aphasia TBI Dementia

How can discourse samples be elicited?

Discourse samples can be elicited through: Picture description (single picture) Picture sequences Narrative retell Narrative completion Personal narratives Joint narrative with spouse Semi-structured interview

What types of questions do we ask in the study of discourse?

How we know to what or whom a pronoun refers? What is conveyed by using the article a vs. the? How we know the meaning of ellipsed statements? How we know what sense of a word is intended?

What do we need to do as we form mental models of discourse?

In order to form mental models from written or spoken discourse, we need to make inferences. Schemas are important for forming inferences. -->During discourse processing, top-down processes lead to the generation of numerous inferences based on schemas

How does inference assist with discourse comprehension? How is knowledge of discourse structure related to inference?

Inference allows the reader or listener to fill in gaps in discourse Knowledge of discourse structure can support our ability to make inferences.

Local vs. global coherence

Inference may assist with local coherence and global coherence. We need to establish local coherence between pieces of information in working memory. We need to establish global coherence between pieces of information at different points in the text (separated)

Logical inferences vs. Bridging inferences vs. Elaborative inferences

Inferences can be divided into three broad categories: Logical inferences: inferences is made based on the the meanings of words (e.g. we make the logical inference that a widow is a woman). Bridging inferences: establish coherence between the current portion of text and the preceding text (includes anaphor resolution- pronouns referring back to nouns) Elaborative inferences: embellish or add details to the text (e.g. the girl running for ice cream is young.)

Strong vs. weak inference

Inferences may be classified as strong or weak. Strong inferences: Obvious, drawn immediately Weak inferences: Non-obvious, drawn later in the text

Analysis of discourse production

Informativeness (e.g., CIUs) Lexical and syntactic complexity Referential clarity Discourse structure Overall coherence Turn-taking Criterion-referenced measures

Discourse comprehension

Memory for details of a text fades over time, but memory for the underlying schema is more robust If the details of a story are in conflict with an existing schema, prior knowledge will result in distortions in recall that make it more similar to the schema

Micro structure and macro structure in the Construction Integration Model

Micro-structure level: propositions are linked to each other within the text. Macro-structure level: The overall gist or message of the story. (We integrate our existing schemas/background knowledge with the propositions in the text to form a coherent mental model of the text.)

What are other confounding factors in discourse research?

Other confounding factors in discourse research include: Memory (working memory required to hold multiple sentences in mind) Executive function (must monitor own behavior in order to achieve a goal: extract meaning from text, communicate information to others. We must use attentional control, planning, switch between ideas and concepts, think about several concepts simultaneously, integrate information about prior knowledge of others).

Constructing propositions. What is a proposition?

Propositions can be used to describe any kind of relationship. Propositions can be combined to represent more complex relationships or ideas. Propositions are the "briefest unit of language that can be independently found to be true or false." (Sternberg, 2006, p. 379) (propositions represent basic relationships between words. A proposition is the smallest unit of language which can be judged as true or false)

How is our semantic knowledge/ existing schemas are related to our understanding of a text.

Readers with limited general knowledge will have less efficient and thorough comprehension We also build vocabulary through reading. We are more likely to remember things that fit with existing schemas. If we don't have an existing schema related to the content of the text, it is hard to connect new information with old information.

What are examples of other types of inferences?

Referential: To what previous word does this apply? Case structure role assignment: What is the role? Causal antecedent: What caused this? Supraordinate goal: What is the main goal? Thematic: What is the overall theme? Character emotional reaction: How does s/he feel? Causal consequence: What happens next? Instrument: What was used to do this? Subordinate goal-action: How was the action achieved?

What are schemas?

Schemas are "well-integrated chunks of knowledge about the world, events, people, and actions." (Eysenck & Keane, 2005, p. 383) They include: Scripts: contain knowledge of events and their consequences. eg. going to restaurant, party Frames: hold knowledge about features, functions, etc. of objects and things in the world. e.g. a coat has arms, keeps you warm, can be different colors

What does discourse comprehension require?

Semantic encoding : Figuring out what words mean by drawing from semantic memory or using context. Constructing propositions : We find out key ideas and relationships between these ideas and store them as propositions. Forming mental models : of characters, settings and events described. Extracting key information : Based on our goals as a reader. Context and our point of view affects our goals.

What are the 3 levels of text representation in the construction-integration model?

The Construction-Integration Model leads to 3 levels of text representation: 1. Surface representation: similar to semantic 2. Propositional representation 3. Situational representation: mental model

Why do we remember propositions better that are central to the story's plot?

The WM "buffer" is periodically emptied of all but a few key propositions that are central to story development More central propositions are remembered better than less central ones because they are held in WM longer and are more likely to be included in the macro-structure

What is semantic encoding?

Vocabulary very closely relates to the ability to comprehend a text This is a two-way street: Semantic knowledge is used to understand discourse, and discourse is often used to build semantics We learn most of our vocabulary indirectly

Do we hold propositions or words in working memory during reading?

We hold propositions rather than words in WM during reading to be able to store more information (analogous to chunking). The longer time in working memory, the better comprehension/recall will be later.

How are schemas related to discourse?

We integrate information from discourse into existing schemas and use our schemas to make sense of discourse. Schemas help us with inferencing. This is an example of top-down processing, as prior knowledge/ existing schemas help us make sense of the text. We are reading/ listening to discourse with our own goal and POV. We will activate different schemas (burglary and home buying) if reading or listening with different goals.

How we extract key information

What key information is extracted from a discourse depends on the reader's / listener's: Goals Point of view

Could some texts lead to multiple mental models?

Yes, some texts can lead to multiple mental models, and these are trickier to comprehend.