Joined up cognition: scripts, frames and schemas

From Psy241wiki

Jump to: navigation, search


The given new contract: Clark 1977

When we make an utterance we present new information (the point of the utterance) and given information (what we assume the listener already knows). We don’t give more information than is necessary, for example when giving directions to a stranger it would be odd to use points of reference such as “that house over there next to my best mate Jacks house”

We also rely on people to make bridging inferences, which is where they work out ambiguous words such as pronouns based on shared knowledge. If I say “Jack has a banana. It is yellow” we use bridging inferences to determine that ‘it’ refers to a banana. Inference bridging include understanding puns and jargons (medical and legal).

Clark and Murphy 1982: we have a need for audience design. We must tailor our speech to suit the audience we are presenting it to.

The speech accommodation theory (Giles, 1984) is similar to Clark and Murphy's 'Audience Design'. According to Giles, people modify their speech style to fit the context of the conversation. For instance, depending on the circumstances, we use 'speech convergence' to be friendly to our audience and 'speech divergence' is used when we do not want to be as friendly e.g. the way we talk might emphasise the in-group and out-group differences between us and the audience.

The importance of context

Bransford and Johnson (1972) gave participants an ambiguous paragraph to read. Participants given the title of the paragraph before it was read had increased comprehension and better recall. Being given the title after the paragraph is read does not improve comprehension or recall. They found the same thing with ambiguous paragraphs and a descriptive picture. Being given the title or a picture of the paragraph creates a schema of what is happening in the paragraph, and allows participants to structure the information in order to better remember and then recall it.

Grice’s Maxim’s of Conversation (1975)

Grice argued we tend to follow several rules when engaging in conversation.

1. Maxim of quantity: don’t say more than is necessary.

2. Maxim of quality: don’t lie for no reason or give statements with no evidence.

3. Maxim of relation: all utterances will be in some way relevant to the conversation

4. Maxim of manner: speak clearly and briefly and make yourself understood

If we don't follow these rules, people tend to notice, and it can be sign of disorders (for example autistic people may not follow these maxims, and persistent unneccessary lying is characteristic off psychopathy and borderline personality disorder).

Schemas and scripts and how we use them to understand language

Bartlett (1932) argued that we can’t simply understand sentences in terms of individual words, there must be schemas involved; generalised descriptions of categories, actions, events or situations. While quite obviously correct researchers have never been able to agree on the specific characteristic of schemas. Bartlett tested the idea of schemas by reading participants a story called 'The War of the Ghosts', and participants then had to recall it in to the next participant and so on. Bartlett found that each person recalled the story differently, such as rationalising some of the details. Bartlett argued that memory is not exact and can be distorted by our individual schemas.

Sachs (1967) tested the recall for the wording of a sentence at different delays. Participants listened to a tape and then were given a sentence and asked to recall whether they’d heard that exact sentence before. The sentences were either semantically different from the original text, changed from active to passive voice or syntactically different from the original text. After delays, syntactic and voice changes were not detected, but semantic ones were. This shows that even after about 5 seconds, we have already lost the detail of something we hear but have committed the ‘gist’ to memory

Frames: Minsky 1976

A frame is a schema for organising information on a single concept. It is a blank template that has slots that can be filled with variables. Variables can be compulsory or optional, and optional slots are not filled with information they may be filled with defaults, which are flexible and can be re-written in light of new information (the default that bananas are edible raw could be overwritten when we find out about species of banana that must be cooked). As we learn more about the frame, we instantiate it or fill it in, though it’s likely that frames are never fully instantiated. The problem with frames is it not clear how context interacts with them. This idea follows Bartlett's idea of schema, but has more recently been descried as a 'script'.


Script theory came as an elaboration of Minsky’s frames; they are sequences of events and actions stored together in our head. Schank and Abelson proposed that we can complete complex action from a combination of simpler scripts. This is similar to the idea of chunking of responses in automaticity. Scripts allow us to complete sequences of routine actions without thinking too much about them. Bower Black and Turner (1979) described people’s typical scripts about certain situations such as going to a restaurant. They found lots of commonality between people. They also then gave pps 1-3 different scripts for the same thing, before presenting them with a final one which they were told to memorise. Participants who read more scripts were more likely to falsely recall typical information about the script that those who read less. When people were given scripts to remember in a random order, they tended to recall them in order, suggesting we organize scripts in our head.

Computer models of scripts and language comprehension Schank and Abelson 1970-1982

SAM (script applier mechanism)

SAM is a computer program that can ‘read between the lines’ of conversation and make bridging inferences. SAM can take a simple sentence “that morning Jack got ready for uni” and use scripts to produce “Jack may have exercised. He may have showered. he put on his clothes. He placed his books in his bag. He checked his timetable. He ate breakfast. He brushed his teeth. He did his hair..” and then answer questions such as “Did Jack brush his teeth that morning?” it does this by filling in defaults Unfortunately SAM could not understand some sentences, because his scripts couldn’t comprehend all possibilities. PAM was developed to cover this

PAM (plan applier mechanism)

PAM was developed to understand more disparate sentences. It constructs likely ‘plans’ that people will follow to make bridging inferences in conversation. It attempts to link sentences together using plans of what people might do. Unfortunately it invents endless amounts and has trouble selected the best one.

Knowledge and reading

Voss Spilich and Chiesi (1978,1979,1980) found that participants high in baseball knowledge understood baseball sentences faster, better and remembered them better and forgot less. This is because they already had rich baseball schemas to add the new information to.

Knowledge and chess

De Groot (1965) showed the correct position of chess pieces on a chess board for 10 seconds then removed them. Participants were required to replace the chess pieces in the exact spot. Experts performed better than average and beginner players. Suggesting that this is also because experts have excellent schemas in chess.

Putting it all together

Links to expertise: Chase and Ericsson (1981) describe the superior processing capabilities of experts as a result of being able to predict what’s coming. Argueably this is what we’re doing with speech comprehension; we’re experts of our own social world. Anderson’s ACT-R provides an integration; it uses existing declarative memory to disambiguate sentences. But do we need scripts and frames if we already have models of semantic memory?

Personal tools