Top Banner
Talking heads!4/10 Easy and hard problems Artificial intelligence Turing test Loebner prize The real thing: human language comprehension
30

Talking heads!4/10

Dec 30, 2015

Download

Documents

lysandra-foley

Talking heads!4/10. Easy and hard problems Artificial intelligence Turing test Loebner prize The real thing: human language comprehension. Overview of comprehension. Overview of comprehension. Vast database. lexicon. - PowerPoint PPT Presentation
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: Talking heads!4/10

Talking heads!4/10Easy and hard problems

Artificial intelligence

Turing test

Loebner prize

The real thing: human language comprehension

Page 2: Talking heads!4/10

Overview of comprehension

Page 3: Talking heads!4/10

Overview of comprehension

parsing SemanticRepresentation(labeled phrase “tree” with Morphemes)

interpretation

lexiconLocal context

Vast database

• (production runs the arrows the other way- creating a tree that determines movements)

Page 4: Talking heads!4/10

detailed sketch

Page 5: Talking heads!4/10

The waveform of the utterance

• Continuous movements convey discrete thoughts.

Page 6: Talking heads!4/10

Lexicon

• function morphemes

• Content words (nouns, verbs, adjectives…)

• Word formation rules

• Example: (adj -->(prefix)+V+ -able) unfixable

• Other info on morphemes: phonology, syntax, meaning and use

Page 7: Talking heads!4/10

Sketch of the lexicon

Page 8: Talking heads!4/10

Parsing -- computing a structural description (labeled tree) using:

• Acoustic cues in time-pressure wave

• Function words and inflections

• Lexical guidance

• Word and phrase order cues

• Determines grammatical relations and input to semantics (recall the aphasic woman given “The bird that the cat watched was hungry.”

Page 9: Talking heads!4/10

Lexical guidance, inflections and parsing

-1• In languages with I. E. inflections, subject and object NPs would be

indicated directly and thus the syntactic role of an NP is clear. In English only the pronouns retain these inflections. Compare the following she/her/her and he/his/him.

• In the following examples, we see that the English loss of inflections and optional that puts an extra load on a word-order dependent parser. Notice that if that -- especially unstressed -- follows directly after the verb, it predicts a complement clause.

Page 10: Talking heads!4/10

What is “lexical guidance?”

• A general term for the role lexical information plays in building trees containing that morpheme.

• Lexicon gives possible or required role players for verbs.

• Recall discussions of “eat” and “dine” from Pinker.

• “Leave” from “The dog barked at the girl that Otto wanted to ()leave()().”

Page 11: Talking heads!4/10

Lexical guidance, inflections and parsing -2

• Some verbs take only simple NP objects (a)

• 11a- The doctor visited the child/her/*she.

• others take complement clause NPs (b)

• 11b- The doctor insisted the child/*her/she/take the pill.

Page 12: Talking heads!4/10

Lexical guidance, inflections and parsing

-3• Some take either (c).

• 11c- 1. The doctor remembered the child/her.)

• 11c- 2. The doctor remembered the child/she/ had an allergy.

Page 13: Talking heads!4/10

The semantic representation of the

utterance• Much of the conventional meaning of

an utterance derives from its hierarchical structure and morpheme meanings or “senses.”

• This is the assumption of combinatorial or compositional semantics.

Page 14: Talking heads!4/10

The listener's interpretation of the utterance is based

on:• A Semantic representation of the

sentence

• Local context– This includes determining referents of referring

expressions, antecedents to anaphors and definite descriptions, nonverbal cues, etc.

• The "vast database”– Anything else!

Page 15: Talking heads!4/10

Reference and local context

• Reference and co-reference– Significant aspects of meaning come from the

context of the utterance. These include deictic words (e.g. now, this, here) and other pronouns, as well immediately preceding conversation in “working memory.”

• Gaps and traces (Pinker, p.219;482)– The policeman saw the boy that the crowd at the

party accused (TRACE) of the crime.

Page 16: Talking heads!4/10

Complex sentences - ambiguity

• Example 1 can be interpreted as a relative (2) or complement clause (3)– 1. “ The fact that Otto knew was surprising.”– 2. “The fact that Otto knew () was surprising.” (Otto

knew some fact that was surprising. Note the gap in 2 but not in 3 below.)

– 3.”(The fact) that Otto knew was surprising.” (A complement clause names explicitly the surprising fact, namely “that Otto knew (something) was surprising.”

Page 17: Talking heads!4/10

Complex sentences - ambiguity

• Multiple gaps allow multiple interpretations– “The girl that Bill wanted () to leave () wore a

blue dress.”– Is “that” coreferent with the first or second gap

-- the subject or object of leave?– Compare “The girl that Bill wanted () to leave

Sam wore a blue dress.”

Page 18: Talking heads!4/10

Acquisition video examples

– When did the boy say () that he hurt () himself?

– What do you think Cookie Monster eats ()?– You think Cookie Monster eats (what)?– *What do you think what's in here?

Page 19: Talking heads!4/10

Complex sentences - “idea density”

• Packaging several propositions (sentences) into one sentence increases complexity of processing.

• The most extreme case is:– The player kicked the ball kicked him.– The player (that was) thrown the ball kicked him.– The player kicked the ball (that was) thrown him.

Page 20: Talking heads!4/10

• Many aphasics cope by using several simple reduced complexity sentences.

• Kemper et al (1997) report low idea density predicts Alzheimers disease decades before other symptoms!.

Complex sentences “idea density” 2

Page 21: Talking heads!4/10

“Vast database” contributions to meaning

• Inference -- going beyond the given propositions -- is part of comprehension.

• Pinker: She: “I’m leaving.” He: “Who is he?”

• “In 1950, I was the tallest kid in sixth grade.”

Page 22: Talking heads!4/10

“Vast database” 2

• Some inferences are “presuppositions” -- implicit statements assumed true by speaker and inferred by listener.

• “Bill knows that the world is flat.”

• “When did you stop drinking?”

Page 23: Talking heads!4/10

• “The earthquake destroyed all the buildings in the town except the mint.”

• “Her gift is too tall for her bedroom.”

• She fell off the first/top rung of the ladder. (primes ok/dead)

• Our brain tends to activate likely consequences as it comprehends L

Vast database 2.5 - inferences from anywhere

based on the semantic interpretation of the

sentence

Page 24: Talking heads!4/10

“vast database” 3

• “sportugese” - Sportsfans predict scores from verbs better than non-fans? No kidding!

• “Lakers crush Celtics” “Celtics hold off Timberwolves”

• What is the margin of victory?

Page 25: Talking heads!4/10

• Read and summarize:• "The procedure is actually quite simple. First you arrange

things into different groups depending on their makeup. Of course one pile may be sufficient depending on how much there is to do. If you have to go somewhere else due to lack of facilities that is the next step, otherwise you are pretty well set. It is important not to overdo any particular endeavor. That is, it may not seem important, but complications from doing too many can easily arise. A mistake can be expensive as well....."

“vast database” 3.5

Page 26: Talking heads!4/10

“vast database” 4

• For most of us, this is a parsable text, using known words, but with little overall meaning unless we were given a title "Washing Clothes." Then it becomes meaningful and much more memorable as Bransford and Johnson, (1972) demonstrated.

• Bransford, J. D., & Johnson, M. K. (1972). Contextual prerequisites for understanding: Some investigations of comprehension and recall. Journal of Verbal Learning and Verbal Behavior, 61, 717-726.

Page 27: Talking heads!4/10

Do hemispheres comprehend differently?

• The LH does much of the language processing in comprehension. However there is some suggestion the RH gives a unique perspective.

Page 28: Talking heads!4/10

RH contribution to comprehension?

• Using similar materials, with and without titles, St George et al (1999) show differential hemisphere involvement -- the RH working hardest when there are no titles.

Page 29: Talking heads!4/10

RH solves the puzzle better

• St George, M., Kutas, M., Martinez, A., & Sereno, M. I. (1999). Semantic integration in reading: engagement of the right hemisphere during discourse processing. Brain, 122(7), 1317-1325.

Page 30: Talking heads!4/10

“vast database” 5

• The listener’s perception of the purpose of the utterance influences one’s interpretation of it.

• "I now see that my husband was simply engaging the world in a way that many men do: as an individual in a hierarchical social order in which he was either one-up or one-down… conversations are negotiations. Life is a contest, a struggle to preserve independence and avoid failure..

• I, on the other hand, was approaching the world as many women do: as an individual in a network of connections.. conversations are negotiations for closeness… they try to protect themselves from others' attempts to push them away.." p.331 712 notes. (Deborah Tannen, 1990)