Post on 17-Dec-2015
Lewis’ Model of Sentence Comprehension
Figuring out at arrived that its subject is the toy - By finding item in memory whose features best match retrieval cue’s - Memory representations & cues both = feature bundles = Content-Addressable Memory (& parallel search)
- Multiple items in memory with similar features make it harder to be sure to get the right one = Similarity-Based Interference at retrieval
Working Memory
• Engle (2010) argues it’s just stuff that’s active because attention is focused on it– Attention required to keep stuff active
• Activation decays when not attended
– People vary in how good they are at focusing and maintaining attention on some things while inhibiting other things
• WM = Interface between memory & attention
Evidence for general purpose WM
• Different kinds of measures of WM capacity produce highly correlated results, as long as the task is “complex”– Verbal/mathematical
• Operation span, counting span, reading span– Spatial
• Navigation span, symmetry span, rotation span
– It doesn’t add much to predictive power to make even this broad distinction
• What does seem necessary is that test requires not being distracted by one thing while trying to retain something else– e.g., between comprehending sentence & remembering
post-sentence letter– Attentional switching & inhibition
Unsworth, Schrock, & Engle (2004)• WM score predicts even low-level control
of eye movements in prosaccade/antisaccade task
Conway, Cowan, & Bunting (2001)
• “Cocktail party” phenomenon weaker for those with high WM scores
• Dichotic listening– Different words in 2 ears– Instructions to attend 1 and ignore other– When own name is in ignored ear, often hear it
• It seems to pop out
– People with higher WM scores less likely to detect own name than those with lower WM scores• Because they’re better at ignoring, i.e., less
distractible
10/27/10 Psyc / Ling / Comm 525 Fall 2010
Nature of Mental Representationof Discourse
• Not a literal record of the input– At least, not for very long– So what is it, then?
• Some kind of representation of meaning abstracted away from form – Two candidates (Probably both right, to some
extent)• Propositions• Mental Models
10/27/10 Psyc / Ling / Comm 525 Fall 2010
Propositions(Kintsch)
• Proposition = a "meaning unit", "idea unit"– Every sentence can be represented by 1 or more
propositions– The stupid man bought the wrong car.
• Proposition 1: BOUGHT (MAN, CAR)• Proposition 2: STUPID (MAN)• Proposition 3: WRONG (CAR)
• BOUGHT, STUPID, WRONG = Predicates– Action, state, relationship, property, ...
• MAN, CAR = Arguments– Entities participating in the action, state, ...
• Roughly, the same propositions could represent other ways of saying the same thing– The wrong car was bought by the stupid man.
10/27/10 Psyc / Ling / Comm 525 Fall 2010
Propositional Representationof Discourse
• Argument Overlap– Sentences that talk about same entities or
situations• Mostly by using anaphors
– Are linked up in memory via overlapping arguments
– The discourse is reorganized in memory
10/27/10 Psyc / Ling / Comm 525 Fall 2010
The librarian reshelved some books that were scattered all around the room. He was growing more and more annoyed. Some of them were on the floor under the tables.
RESHELVED (LIBRARIAN, BOOKS) ANNOYED (LIBRARIAN) SCATTERED (BOOKS)
ON (BOOKS, FLOOR)
UNDER (BOOKS, TABLES)
• Propositions about the same thing stored "near" each other– Passage is reorganized in representation
– ON (BOOKS, FLOOR) is "closer to" SCATTERED (BOOKS) than to ANNOYED (LIBRARIAN) in the propositional representation
• Even though in the original passage, the words Some of them were on the floor... are closer to the words He was growing more and more annoyed
10/27/10 Psyc / Ling / Comm 525 Fall 2010
Evidence for Propositional Representation
• Ratcliff & McKoon (1978)– People read series of unrelated sentences like– The mausoleum that enshrined the czar overlooked
the square.
– Propositions: ENSHRINED (MAUSOLEUM, CZAR) OVERLOOKED (MAUSOLEUM, SQUARE)
– After reading a series of sentences, recognition task for words from sentences
• Words presented in a list• Sometimes words from same sentence next to each other
in list
• Idea: words that are "close" in memory should prime each other
10/27/10 Psyc / Ling / Comm 525 Fall 2010
ResultsWord sequence Sentenc
ePropositio
nRT
(msec)
Priming
desk-square Different Different 671
czar – square Same Different 571 100
mausoleum-square
Same Same 551 20
• When 2 words came from the same sentence• Faster to respond to 2nd one than when came from different sentences
• When they came from the same proposition• Faster to respond to 2nd one than when came from diff. propositions
• Notice, better priming from words that were farther apart in sentence• Supports reorganization idea
•This study used single sentences - what about discourse?
10/27/10 Psyc / Ling / Comm 525 Fall 2010
Dell, McKoon & Ratcliff (1983)• Context passage:
– A car sped around a corner out of control. A grocer was outside a store arranging apples. A little puppy was dragging its owner on a leash.
• Then, there was a final sentence either without or with an anaphor:– A bicycle smashed into some boxes.– The vehicle smashed into some boxes.
• Task = Probe Recognition:– While people were reading the final sentence– a word flashed on the screen– & they said whether it had appeared earlier in the passage.
• Types of probes:– Antecedent of anaphor (when anaphor was present) car– Another word from the same proposition as antecedent corner– A more recent word from a different proposition apples
10/27/10 Psyc / Ling / Comm 525 Fall 2010
• Logic– If you retrieve the antecedent of an anaphor to
understand it– Then the antecedent should be active right after reading
an anaphor, compared to when there was no anaphor
• So,– If the word car flashes on the screen right after the
anaphor vehicle in:• A car sped around the corner. .... The
vehicle ... ,
– saying yes car did appear earlier in story should be faster than
– saying yes to car after the non-anaphor bicycle in• A car sped around the corner. ... A bicycle ...
10/27/10 Psyc / Ling / Comm 525 Fall 2010
Conclusions
• People retrieve antecedents of anaphors very quickly – There’s significant priming immediately after the
anaphor
• They seem to retrieve the whole proposition introducing the antecedent– There’s just as much priming for corner as for car
immediately after the anaphor
• But they quickly discard the non-antecedent parts of the proposition – Priming for corner disappears by the very next word
10/27/10 Psyc / Ling / Comm 525 Fall 2010
Memory for LanguageBransford, Barclay, & Franks (1972)
• Probe recognition test for whole sentences
• Original sentence:– Three turtles rested on a floating log and a fish swam
beneath them.or– Three turtles rested beside a floating log and a fish swam
beneath them.
• Test sentence:– Three turtles rested on a floating log and a fish swam
beneath it.– or– Three turtles rested beside a floating log and a fish swam
beneath it.
• Results:– People wrongly said “yes” to on test sentences more than to beside
ones
10/27/10 Psyc / Ling / Comm 525 Fall 2010
Memory for Inferences• Why were people more likely to think they saw the
original on sentences?– Because original & test on sentences are consistent
with each other– i.e., they can both describe the same situation
• People seem to build a kind of Mental Model of the situation described– Also called Situation Model
• & remember in that form the meaning of what was said
• Model often requires making inferences beyond what was said
• We’re quite bad at remembering what was said vs what we inferred– Advertisers and lawyers rely on this!
10/27/10 Psyc / Ling / Comm 525 Fall 2010
Mental ModelsGlenberg, Meyer, & Lindem (1987)
• People read passages like– John was preparing for a marathon in August. After
doing a few warm-up exercies, he { put on / took off } his sweatshirt and went jogging. He jogged halfway around the lake without too much difficulty. Further along his route, his muscles began to ache.
• Probe Recognition Task– At the end of the passage, people decided whether the word
sweatshirt had been in it
• Results– People were faster to say "yes" to sweatshirt
• after the put on version of the story• than after the took off version
– Because John is wearing the sweatshirt in the mental image you have at the end of the put on version?
• So you don't have to remember back to when it was mentioned?
Issues about Memory & Language
• Is there language-specific working memory (WM), or is there just one general WM that language shares with other processes?
• Caplan says there are 2 separate kinds of WM resources used in language comprehension
– One for automatic aspects of language comprehension• e.g., word recognition, parsing• Completely language-specific resource
– Another for later less automatic components• e.g., resolving anaphors, linking discourse elements across
sentences• General resource
– Only the second is interfered with by additional demands on WM from outside the comprehension process
• Fedorenko et al. ( 2006) are testing this claim
10/27/10 Psyc / Ling / Comm 525 Fall 2010
Similarity-Based Interference• In Lewis’, Gibson’s, and Gordon’s Parsing Models
– When multiple entities must be held in WM until their role is determined by later words
• e.g., Relative Clause Head– SubRel: The witness who ___ saw the murderer testified …– ObjRel: The witness who the murderer saw ___ testified …
– The more similar the multiple entities are• e.g., both proper names, or both definite descriptions, or
both animate• The harder it will be to retrieve & link the correct one
– The slower reading times will be at the retrieval point– The more likely comprehension errors will be
– Van Dyke & McElree (2006) are testing the nature of Similarity-Based Interference
10/27/10 Psyc / Ling / Comm 525 Fall 2010