Ambiguity in Language
Models of ambiguity resolution
-Garden Path Model -Mixed Model of Ambiguity Resolution
Garden Path Model problems
-Mackay: takes longer to finish ambiguous fragments
Clause boundary criticality: results
-ambiguous/before clause boundary: slowest -after clause boundary: no difference between amb/unamb
mean response latencies to the third word: conclusions
-body interpretation is selected in controls, so doesn't prime piano (same as for ceiling) -people with alzheimers: organ as a musical instrument interpretation hangs around
relation to Neely
-fast acting modular lexical processor first, independent of context/expectations -followed by a slower more attention demanding integration process
language ambiguity levels
-lexical (word level) -syntactical (noun phrases, verb phrases, etc.)
Garden Path Model
-one initially only computes a single interpretation of an ambiguous utterance --we can use context to override/decide it -if it is wrong, we have to reanalyse surface structure to recompute new meaning
Mackay study: results
-took longer to begin completing ambiguous fragments -took longer to complete ambiguous fragments -there were more stutters, dysfluencies, etc. in ambiguous fragments > but participants were not aware of the ambiguities
Garden Path Model: 2 observations
1. we rarely notice ambiguities 2. there is a startle response, when one initially interprets the wrong meaning
Mixed Model of Ambiguity Resolution: support
Bever, Garrett and Hurtig -clause boundary criticality Swinney -multiple interpretations are computed
Clause boundary criticality
Bever, Garrett, Hurtig sentence completion task -complete a fragment before clause boundary, or a full clause -within that, it is either ambiguous or not
lexical ambiguity examples
Child's stool great for use in gardens -stool: chair, pee [see slides]
Schizophrenia and ambiguity
Cohen & Servan-Schreiber schizophrenics (vs controls) are more likely to retain the multiple interpretations for longer
syntactic ambiguity examples
For sale: mixing bowl set designed to please a cook with round bottom for efficient beating [see slides]
Late Closure
Frazier we attach each phrase to most recent phrase (as opposed to an earlier phrase) -minimize strains on working memory ex: (mixing bowl): "efficient beating" refers to "round bottom" which refers to "a cook"
Good vs Poor Comprehenders (of text)
Gernsbacher and Faust a low scorer on comprehension test is more likely to retain multiple interpretations for a long time -show priming for 'spy' at 3 syllables -have selection deficit -larger Stroop effects
Mixed Model of Ambiguity Resolution
Mackay 1. compute all interpretations of ambiguous utterances 2. based on the available evidence, attempt to select an interpretation 3. if insufficient information is available, must select by the clause boundary 4. if wrong selection is selected, then attempt to recover surface structure to reinterpret
Alzheimers and ambiguity
Mean Response Latencies to the Third Word Balota and Duchek
Swinney's cross-modal priming: example
Rumor had it that, for years, the government building had been plagued with problems. The man was not surprised when he found several spider roaches and bugs(|) in the cor(|)ner of the room -visual stim at one of the (|)
Mackay study: example
ambiguous: -after taking the right turn at the intersection.... -knowing that visiting relatives could be bothersome.... unambiguous: -after taking the left turn at the intersection... -knowing that some relatives could be bothersome...
Clause boundary criticality: examples
before clause boundary: -"although flying planes can..." (amb) -"although some planes can..." (unamb) after clause boundary: -"although flying planes can be dangerous, he..." (amb) -"although some planes can be dangerous, he..." (unamb)
startle response and humor
for some reason we think the startle response is funny -punchlines -[A man and his pet giraffe walk into a bar and start drinking. As the night goes on, they get drunk, and the giraffe finally passes out. The man decides to go home. As he's leaving, the man is approached by the barkeeper who says, "Hey, you're not gonna leave that lyin' here, are ya?" "Hmph," says the man. "That's not a lion -- it's a giraffe."]
group differences in selection process
if it's attentional selection like Swinney suggests, people with poor attentional control may show priming for both meanings for a longer period of time -good vs bad comprehenders -schizophrenics -alzheimers
Mackay study: conclusion
individuals ARE computing ambiguities (all interpretations), but aren't aware of them
Mackay study: sentence continuation
participants are given a fragment and must complete it -fragments are either ambiguous or unambiguous (control) -if garden path is correct: no difference in time to complete un/ambiguous fragments because you only choose 1 interpretation either way
Swinney's cross-modal priming
participants listen to stuff, and are presented words on a screen (decide if word/nonword) -we synchronize screen to what they're hearing -bugs example -we can present the stim either right after 'bugs' or 3 syllables later
mean response latencies to the third word
present the words surgeon, organ, and piano prime 1 | prime 2 | target | controls | alz -surgeon | organ | piano | 567 | 705 -surgeon | ceiling | piano | 566 | 725
Garden Path Model: startle response
sometimes we compute the wrong interpretation, so we have to recompute -[slides example, Muhammad Ali, "punch": the drink, not the action]
Swinney's cross-modal priming: resutls
visual stim: 0 syllables (immediately)/3 syllables ANT: 708/795 SPY: 715/849 SEW: 746/848 -there is priming for both 'ant' and 'spy' at 0 syllables -only 'ant' has priming after 3 syllables -'spiders and roaches' drives interpretation after 3 syllables (2nd step)
Garden Path Model: rarely notice ambiguities
we only compute one meaning, so we don't notice others -"The farmer put the straw by the threshing machine": can also be drinking straw