Memory for Meaning and Surface Memory Research Paper

(a) The policeman chased the suspect.

(b) The officer chased the suspect.

(c) The suspect was chased by the officer.

Each of these sentences uses a different surface form to convey exactly the same meaning. Psycholinguistic research has shown that people recall the meaning of a sentence much better than its surface form, but that memory for form, under some conditions, is surprisingly robust.

1. Early Research on Memory for Meaning and Surface Form

Sachs (1967) carried out one of the first comparisons between memory for meaning and memory for surface form. She presented listeners with recorded passages that were interrupted zero, 80, or 160 syllables after a critical sentence. During the interruption, listeners were asked to judge whether a test sentence had occurred, in exactly the same form, earlier in the passage. Sometimes the test sentence was identical to the critical sentence (e.g., ‘He sent a letter about it to Galileo, the great Italian scientist.’), sometimes it was a paraphrase (e.g., ‘He sent Galileo, the great Italian scientist, a letter about it.’), and sometimes it altered the meaning of the critical sentence (e.g., ‘Galileo, the great Italian scientist, sent him a letter about it.’). On the immediate test, listeners correctly rejected both the paraphrases and the sentences with altered meanings. But after just 80 syllables of intervening text, performance on the paraphrases showed a precipitous decline. These results were taken as evidence that the surface form of a sentence is held in working memory long enough for the meaning to be extracted, but only the meaning is stored in long-term memory.

In a subsequent study, Jarvella (1971) interrupted recorded narratives and asked listeners to write down as much of the preceding narrative as they could recall ‘exactly, word-for-word.’ The two sentences preceding each interruption took one of two forms. The final two clauses were identical in both forms, but in one the next-to-last clause was part of the final sentence while in the other it was part of the preceding sentence. Consider the following examples:

(d) The confidence of Kofach was not unfounded. To stack the meeting for McDonald the union had even brought in outsiders.

(e) Kofach had been persuaded by the international to stack the meeting for McDonald. The union had even brought in outsiders.

About 80 percent of the words in the next-to-last clause (e.g., ‘to stack the meeting for McDonald’) were recalled verbatim when it was part of the final sentence as in (d), but fewer than 50 percent were recalled correctly when it was part of the penultimate sentence as in (e). These results suggest that sentence boundaries mark the point where the meaning of a sentence is stored in long-term memory and its surface form is irretrievably lost.

2. Surface Memory

In the late 1970s and early 1980s several experiments were published demonstrating reliable surface memory for sentences in natural settings. Kintsch and Bates (1977) tested memory for statements made during classroom lectures. They found that college students could reliably discriminate between statements they had actually heard (‘Galton was the brilliant younger cousin of Darwin.’) and meaning preserving paraphrases of those statements (‘Darwin was the older cousin of the extremely intelligent Galton.’) as much as five days later. Similar results were reported by Keenan et al. (1977) who investigated memory for statements made during a faculty lunchtime conversation, and by Bates et al. (1980) who tested memory for statements from a television soap opera.

Two factors appear to explain why these researchers found evidence for surface memory while Sachs (1967) and Jarvella (1971) did not. First, the experiments were conducted in natural settings where the surface form of a sentence can be socially relevant. This claim is supported by findings that surface memory is most robust for jokes (Kintsch and Bates 1977), mock insults (Keenan et al. 1977), and other socially significant utterances. Second, each of these studies included one or more control groups whose members completed the memory test without prior exposure to the lecture, conversation, or soap opera. The performance of these groups provides a baseline against which surface memory is more easily detected.

Additional evidence of long-term surface memory comes from studies that employ ‘indirect’ measures of memory. In one such study, Tardif and Craik (1989) created two versions of the same passage. Each sentence in one passage was paraphrased in the other. One week after reading the passage, participants were asked to read it again. Half saw the same version on both readings while half saw different versions. Rereading times were faster when the surface form remained constant.

3. Memory for Meaning

Van Dijk and Kintsch (1983) have proposed that the meaning of a discourse is represented in memory by two distinct, but interrelated, knowledge structures that they call the propositional textbase and the situation model.

3.1 The Propositional Textbase

Bransford and Franks (1971) argued that the meaning of a discourse is stored in memory as a network of ideas that transcend sentence boundaries. Consider the following:

(f ) The ants ate the sweet jelly that was on the table. (g) The ants in the kitchen ate the jelly.

(h) The ants in the kitchen ate the sweet jelly, which was on the table.

(i) The jelly was sweet.

These sentences convey four basic ideas: the ants ate the jelly, the ants were in the kitchen, the jelly was sweet, and the jelly was on the table. Bransford and Franks presented listeners with sentences like these, followed by a recognition memory test. The primary determinant of recognition performance was the number of ideas conveyed by each test sentence. Listeners were most likely to ‘remember’ sentence (h), which includes all four ideas, even if they never actually heard it. These results suggest that as each sentence is heard, the ideas are extracted and stored in long-term memory together with related ideas. Test sentences are then recognized by comparing them to the complete, integrated set of ideas in memory.

To account for results such as these, psychologists borrowed the idea of a proposition from linguistics and philosophy. Propositions are sometimes defined as the smallest units of meaning to which we can assign a truth value. It makes sense to ask whether ‘the ants ate the jelly’ (a complete proposition) is true, but not to ask whether ‘ants’ by itself is true. Propositions are also defined by their structure. Each proposition consists of a single predicate term (a verb, adverb, adjective or preposition) and one or more arguments (nouns or other propositions). By either definition, each of Bransford and Franks’ ‘ideas’ corresponds to a single proposition that we can represent as follows:

( j) EAT (ANTS, JELLY)

(k) IN (ANTS, KITCHEN)

(l) SWEET (JELLY)

(m) ON (JELLY, TABLE)

Most psychologists now agree that the meaning of a discourse is represented in memory as a propositional textbase: a network of propositions connected by shared arguments. Thus, EAT (ANTS, JELLY) is connected to SWEET (JELLY) by JELLY and to IN (ANTS, KITCHEN) by ANTS. This claim is supported by over 25 years of research on the comprehension and recall of discourse (for a review, see van Dijk and Kintsch 1983, Chap. 2). Among the most compelling is a study by McKoon and Ratcliff (1980) who presented readers with brief stories followed by a speeded recognition task. Readers were shown a series of probe words and asked to judge, as quickly as possible, whether each word had occurred in one of the stories. Probe words were recognized more quickly when the preceding test word was from the same story. This facilitation effect became more pronounced as the distance between the words in the propositional textbase decreased.

3.2 The Situation Model

During comprehension, listeners and readers use their knowledge of the world to go beyond the propositional textbase and create a representation of the situation described by a discourse, a representation that is as much as possible like the representation that would result from direct experience. Van Dijk and Kintsch (1983) refer to this as a situation model. A situation model can take many forms and often includes sensory-motor information that is not normally associated with strictly linguistic representations. Among the most studied types of situation models are those associated with spatial descriptions and simple narratives.

Mani and Johnson-Laird (1982) demonstrated that a propositional textbase does not adequately describe how spatial descriptions are represented in memory. They presented readers with two types of descriptions:

(n) The bookshelf is to the right of the chair. The chair is in front of the table. The bed is behind the table. (determinate)

(o) The bookshelf is to the right of the chair. The chair is in front of the table. The bed is behind the chair. (indeterminate)

These descriptions differ by a single argument to a single proposition: BEHIND (BED, TABLE) versus

BEHIND (BED, CHAIR). This suggests that there should be little or no difference in how these descriptions are understood and remembered. But the situation associated with the indeterminate description (o) is ambiguous. The bed could be either in front of or behind the table. Because of this, Mani and JohnsonLaird hypothesized that readers would have greater difficulty understanding and remembering indeterminate descriptions. Their experiments confirm this prediction.

Trabasso et al. (1984) were among the first to propose that causal connections play a central role in the mental representation of narratives. They claim that each event in a narrative is understood by determining its causes and consequences, and that the narrative as a whole is understood by finding a chain of causally related events that connect its opening to its eventual outcome. In support of this claim they showed that events with many causal connections to the rest of a narrative are remembered better and rated more important than otherwise similar events with fewer causal connections. They also showed that events on the causal chain that connects the opening of a narrative to its outcome are recalled better and rated more important than causal ‘dead ends.’ These results suggest that narratives are represented in memory by a complex situation model that includes a representation of each event and the causal connections that tie the events together.

4. Current and Future Directions

A major goal of psycholinguistic research is to create computer models that simulate the comprehension and recall of discourse (e.g., Kintsch 1998). Achieving this goal will depend on understanding, in detail, how the surface forms and meanings of sentences are represented in memory. Another significant research goal is to identify the specific brain regions involved in understanding and remembering discourse. Up to the end of the twentieth century, most research of this type focused on isolated words and out-of-context sentences. Research on how, and where, the brain creates a coherent propositional textbase or situation model is still in its infancy (e.g., Beeman and Chiarello 1998).

Bibliography:

1. Bates E, Kintsch W, Fletcher C R, Giuliani V 1980 The role of pronominalization and ellipsis in texts: Some memory experiments. Journal of Experimental Psychology: Human Learning and Memory 6: 676–91
2. Beeman M, Chiarello C (eds.) 1998 Right Hemisphere Language Comprehension: Perspecti es from Cogniti e Neuroscience. Erlbaum, Mahwah, NJ
3. Bransford J D, Franks J J 1971 The abstraction of linguistic ideas. Cogniti e Psychology 2: 331–50
4. Jarvella R J 1971 Syntactic processing of connected speech. Journal of Verbal Learning and Verbal Beha ior 10: 409–16
5. Keenan J M, MacWhinney B, Mayhew D 1977 Pragmatics in memory: A study of natural conversation. Journal of Verbal Learning and Verbal Beha ior 16: 549–60
6. Kintsch W 1998 Comprehension: A Paradigm for Cognition. Cambridge University Press, Cambridge, UK
7. Kintsch W, Bates E 1977 Recognition memory for statements from a classroom lecture. Journal of Experimental Psychology: Human Learning and Memory 3: 150–9
8. Mani K, Johnson-Laird P N 1982 The mental representations of spatial descriptions. Memory & Cognition 10: 181–7
9. McKoon G, Ratcliff R 1980 Priming in item recognition: The organization of propositions in memory for text. Journal of Verbal Learning and Verbal Beha ior 19: 369–86
10. Sachs J 1967 Recognition memory for syntactic and semantic aspects of connected discourse. Perception & Psychophysics 2: 437–42
11. Tardif T, Craik F I M 1989 Reading a week later: Perceptual and conceptual factors. Journal of Memory and Language 28: 107–25
12. Trabasso T, Secco T, van den Broek P 1984 Causal cohesion and story coherence. In: Mandl H, Stein N L, Trabasso T (eds.) Learning and Comprehension of Text. Erlbaum, Hillsdale, NJ
13. van Dijk T, Kintsch W 1983 Strategies of Discourse Comprehension. Academic Press, New York