Transcript Brachet - UB Computer Science and Engineering

Artificial Intelligence, Natural Language,
the Chinese Room,
& Reading for Understanding
William J. Rapaport
Department of Computer Science & Engineering,
Department of Philosophy, Department of Linguistics,
and Center for Cognitive Science
[email protected]
http://www.cse.buffalo.edu/~rapaport
What Is AI?
2 Contrasting Views
1. “The science of making machines do things that
would require intelligence if done by humans.”
–
–
Marvin Minsky
Using humans to tell us how to program computers
•
•
–
e.g., play chess, solve calculus problems, …
e.g., see, use language, …
“classical AI”
What Is AI? (continued)
2. “The use of computer programs
and programming techniques to cast light on
the principles of intelligence in general
and human thought in particular.”
–
–
–
Margaret Boden
Using computers to tell us something about humans.
Cognitive science
What Is AI? (cont’d)
• My view: AI “computational cognition” asks:
– How much of cognition is computable?
• 2 further questions:
– What is computation?
• Turing machine
– How will we know if (some aspect of)
cognition is computable?
• Turing test
Alan Turing
• British, 1912-1954
• “invented” the idea of
computation
– Turing “machine”
• cracked the Nazi
“Enigma” code
during WW II
• devised test for AI:
– Turing “test”
Turing Machine
• Everything that is (humanly) computable
can be computed using:
– a machine with:
• an “infinite” tape divided into squares (with ‘0’ on each)
• a movable read-write head
– a programming language with only:
• 2 nouns: 0, 1
• 2 verbs: move(left or right), print(0 or 1)
• 3 grammar rules:
– sequence (begin do this; then do that end)
» where “this”/“that” = print, move, or any grammatical instruction
– selection (if current square=0 then do this else do that)
– repetition (while current square=0 [or: current square=1] do this)
• Not everything can be computed!
– “halting problem”: can’t algorithmically detect infinite loops
– can cognition be computed?
(How) Can Computers Think?
(Human) cognition is computable (?)

(Human) cognitive states & processes
can be expressed as algorithms

They can be implemented
on non-human computers
How Computers Can Think (cont’d)
• Are computers executing such cognitive algorithms
merely simulating cognitive states & processes?
• Or are they actually exhibiting them?
– Do such computers think?
• Answer: Turing’s Test
Objection: Searle’s Chinese-Room Argument
My reply: Computers can understand just by
manipulating symbols
(like a Turing machine)
The Imitation Game
MAN
WOMAN
“I’m the woman”
“I’m the woman”
INTERROGATOR
The Turing Test
The Imitation Game
COMPUTER
MAN
WOMAN
“I’m the woman”
“I’m the woman”
INTERROGATOR
The Turing Test #2
The Imitation Game
COMPUTER
MAN
MAN
WOMAN
“I’m the woman”
man
“I’m the woman”
man
INTERROGATOR
The Turing Test #3
The Imitation Game
COMPUTER
MAN
HUMAN
WOMAN
“I’m the woman”
human
“I’m the woman”
human
INTERROGATOR
The Turing Test
I
Questions
Responses
H? / C?
“I believe that at the end of the century the use of words
and general educated opinion will have altered so much
that one will be able to speak of machines thinking
without expecting to be contradicted.”
- Turing 1950
The Chinese-Room Argument
• It’s possible to pass TT, yet not (really) think
story + questions
I
H
(in Chinese)
(who can’t
understand Ch.)
+
responses
(Eng.) program
(native Chinese
speaker)
(in fluent Chinese)
for manipulating
[Ch.] “squiggles”
Searle’s Chinese-Room Argument
(1) Computer programs just manipulate symbols
(2) Understanding has to do with meaning
(3) Symbol manipulation alone
is not sufficient for meaning
(4)  No computer program can understand
¬ (3): You can understand just by
manipulating symbols!
Contextual Vocabulary
Acquisition
• Could Searle-in-the-room figure out a
meaning for an unknown squiggle?
• Yes!
– Same way you can figure out a meaning
for an unfamiliar word from context
What Does ‘Brachet’ Mean?
(From Malory’s Morte D’Arthur [page # in brackets])
There came a white hart running into the hall with a white brachet
next to him, and thirty couples of black hounds came running after
them. [66]
2. As the hart went by the sideboard,
the white brachet bit him. [66]
3. The knight arose, took up the brachet and
rode away with the brachet. [66]
4. A lady came in and cried aloud to King Arthur,
“Sire, the brachet is mine”. [66]
10. There was the white brachet which bayed at him fast. [72]
18. The hart lay dead; a brachet was biting on his throat,
and other hounds came behind. [86]
1.
Where is the “context”?
Reader’s Prior Knowledge
PK1
PK2
PK3
PK4
Text
Reader’s Prior Knowledge
PK1
PK2
PK3
PK4
Text
T1
Internalized (Co-)Text
integrated with
Reader’s Prior Knowledge
internalization
PK1
I(T1)
PK2
PK3
PK4
Text
T1
Belief-Revised
Internalized Text
integrated with
Reader’s Prior Knowledge
internalization
PK1
I(T1)
PK2
inference
PK3 P5
PK4
Text
T1
B-R Integrated KB
Text
internalization
PK1
I(T1)
PK2
inference
PK3 P5
PK4
P6
I(T2)
T1
T2
B-R Integrated KB
Text
internalization
PK1
I(T1)
PK2
T1
T2
inference
PK3 P5
PK4
I(T2)
P6
I(T3)
T3
B-R Integrated KB
Text
internalization
PK1
I(T1)
PK2
T1
T2
inference
PK3 P5
PK4
I(T2)
P6
I(T3)
T3
“Context” for CVA is the reader’s mind, not the (co-)text
Reader’s Mind
internalization
PK1
P7
Text
I(T1)
PK2
T1
T2
inference
PK3 P5
PK4
I(T2)
P6
I(T3)
T3
• “Words should be thought of not as having intrinsic meaning,
but as providing cues to meaning.”
– Jeffrey L. Elman, “On the Meaning of Words and Dinosaur Bones:
Lexical Knowledge Without a Lexicon” (2009)
• “Words might be better understood as operators,
entities that operate directly on mental states
in what can be formally understood as a dynamical system.”
– Jeffrey L. Elman, “On Words and Dinosaur Bones: Where Is Meaning?” (2007)
Computational CVA (cont’d)
• KB: SNePS representation of reader’s prior knowledge
• I/P: SNePS representation of word in its co-text
• Processing (“simulates”/“models”/is?! reading):
– Uses logical inference, generalized inheritance, belief revision
to reason about text integrated with reader’s prior knowledge
– N & V definition algorithms deductively search this
“belief-revised, integrated” KB (the wide context)
for slot fillers for definition frame…
• O/P: Definition frame
– slots (features): classes, structure, actions, properties, etc.
– fillers (values): info gleaned from context (= integrated KB)
Cassie learns what “brachet” means:
Background info about: harts, animals, King Arthur, etc.
No info about:
brachets
Input:
formal-language (SNePS) version of simplified English
A hart runs into King Arthur’s hall.
• In the story, B12 is a hart.
• In the story, B13 is a hall.
• In the story, B13 is King Arthur’s.
• In the story, B12 runs into B13.
A white brachet is next to the hart.
• In the story, B14 is a brachet.
• In the story, B14 has the property “white”.
• Therefore, brachets are physical objects.
(deduced while reading;
PK: Cassie believes that only physical objects have color)
--> (defineNoun "brachet")
Definition of brachet:
Class Inclusions: phys obj,
Possible Properties: white,
Possibly Similar Items:
animal, mammal, deer,
horse, pony, dog,
I.e., a brachet is a physical object that can be white
and that might be like an animal, mammal, deer,
horse, pony, or dog
A hart runs into King Arthur’s hall.
A white brachet is next to the hart.
The brachet bites the hart’s buttock.
[PK: Only animals bite]
--> (defineNoun "brachet")
Definition of brachet:
Class Inclusions: animal,
Possible Actions: bite buttock,
Possible Properties: white,
Possibly Similar Items: mammal, pony,
A hart runs into King Arthur’s hall.
A white brachet is next to the hart.
The brachet bites the hart’s buttock.
The knight picks up the brachet.
The knight carries the brachet.
[PK: Only small things can be picked up/carried]
--> (defineNoun "brachet")
Definition of brachet:
Class Inclusions: animal,
Possible Actions: bite buttock,
Possible Properties: small, white,
Possibly Similar Items: mammal, pony,
A hart runs into King Arthur’s hall.
A white brachet is next to the hart.
The brachet bites the hart’s buttock.
The knight picks up the brachet.
The knight carries the brachet.
The lady says that she wants the brachet.
[PK:
Only valuable things are wanted]
--> (defineNoun "brachet")
Definition of brachet:
Class Inclusions: animal,
Possible Actions: bite buttock,
Possible Properties: valuable, small,
white,
Possibly Similar Items: mammal, pony,
A hart runs into King Arthur’s hall.
A white brachet is next to the hart.
The brachet bites the hart’s buttock.
The knight picks up the brachet.
The knight carries the brachet.
The lady says that she wants the brachet.
The brachet bays at Sir Tor.
[PK: Only hunting dogs bay]
--> (defineNoun "brachet")
Definition of brachet:
Class Inclusions: hound, dog,
Possible Actions: bite buttock, bay, hunt,
Possible Properties: valuable, small, white,
I.e. A brachet is a hound (a kind of dog) that can bite, bay, and hunt,
and that may be valuable, small, and white.
Algorithms (for Computers)
1.
Generate initial hypothesis by
“syntactic manipulation”
•
•
Algebra: Solve an equation for unknown value X
Syntax: “Solve” a sentence for unknown word X
–
–
2.
Deductively search wide context to update hypothesis
•
•
3.
“A white brachet (X) is next to the hart”
 X (a brachet) is something that is next to the hart and
that can be white.
“Define” node/word X
in terms of immediately connected nodes/words
Look for: class membership, properties, structure, acts, agents, etc.
Define” node/word X
in terms of some (but not all) other connected nodes/words
Output definition “frame” (schema)
A Computational Theory of CVA
1.
2.
A word does not have a unique meaning.
A word does not have a “correct” meaning.
Author’s intended meaning for word doesn’t need to be known by reader
in order for reader to understand word in context
Even familiar/well-known words can acquire new meanings in new contexts.
Neologisms are usually learned only from context
a)
b)
c)
3.
Every co-text can give some clue to a meaning for a word.
•
4.
Generate initial hypothesis via syntactic/algebraic manipulation
But co-text must be integrated with reader’s prior knowledge
Large co-text + large PK  more clues
Lots of occurrences of word allow asymptotic approach to stable meaning hypothesis
a)
b)
5.
CVA is computable
CVA is “open-ended”, hypothesis generation.
a)
•
b)
6.
7.
CVA ≠ guess missing word (“cloze”);

CVA ≠ word-sense disambiguation
Some words are easier to compute meanings for than others (N < V < Adj/Adv)
CVA can improve general reading comprehension (through active reasoning)
CVA can & should be taught in schools
CVA as Symbol Manipulation
• We “solved” each sentence for the unknown word
– in terms of the rest of the text
together with our background knowledge
• Just as we can solve an algebra problem
– in terms of the rest of the equation
• By manipulating symbols
– which is what computers do!
Computational
Natural-Language Understanding
• What else is needed?
– besides symbol manipulation
Mind as a Symbol-Manipulation System
To understand language, a cognitive agent must:
• Take discourse as input
• Understand ungrammatical input
• Make inferences & revise beliefs
• Make plans
– For speech acts
– To ask/answer questions
– To initiate conversation
• Understand plans
– Speech-act plans of interlocutor
•
•
•
•
Construct user model
Learn (about world, language)
Have background/world/commonsense knowledge
Remember
– What it heard, learned, inferred, revised
= have a mind!
• All of this can be done by computers manipulating symbols!
Reading for Understanding
Research Initiative
• Institute of Education Sciences
– US Department of Education
• 5 or 6 “R&D Core Teams”
• 1 or 2 “Network Assessment Teams”
• “Manhattan Project” / “Apollo Mission”
– to improve (the teaching of)
reading for understanding/reading comprehension
– $20 million over 5 years
A Center for Reading for Understanding
• A Research and Development Core Team to Integrate
Vocabulary, Writing, Reasoning, Multimodal Literacies,
and Oral Discourse to Improve Reading Comprehension
• Principal location: UB
• Satellite locations: Niagara University,
Penn State University
• Affiliated school districts: Niagara Falls CSD,
Cleveland Hill USD,
State College Area SD
Projects
• Writing Intensive Reading Comprehension
– Jim Collins (UB/LAI)
• Contextual Vocabulary Acquisition
– Bill Rapaport (UB/CSE & CCS)
• Multimodal Literacies
– Kathleen Collins (Penn State)
• Virtual Interactive Environments to improve vocabulary and reading
comprehension
– Lynn Shanahan & Mary McVee (UB/LAI)
• Interactively Modeled Metacognitive Thinking
– Rob Erwin (Niagara U)
• Bilingualism and Basic Cognitive Processes
– Janina Brutt-Griffler (UB/LAI)
o Experimental Design & Statistical Analysis
– Ariel Aloe (UB/CSEP)
Contextual Vocabulary Acquisition
— Bill Rapaport (CSE & CCS)
1. Previous research:
•
2.
A reader's understanding of a word's meaning in a context
is a function of both the context (the surrounding words)
and the reader's prior knowledge.
Already accomplished:
a)
A procedure for successful CVA can be expressed in terms so precise that
they can be programmed into a computer.
b) That computational procedure can then be converted into a strategy
teachable to human readers
3.
We propose to embed this procedure in a curricular
intervention that can help readers improve both
a) their vocabulary and
b) their reading comprehension.
CVA (cont’d)
• Our goal is not to improve vocabulary per se,
but:
– to improve reading comprehension by
• active thinking about the text
• with a specific goal of vocabulary enrichment in mind
– and to provide readers with a method that can be
used independently…
• e.g., when they are reading on their own
…to learn new vocabulary
and to improve comprehension.
CVA (cont’d)
• GOFAI:
– if we know how to explicitly teach some
cognitive task to humans
• e.g., play chess, do calculus, prove theorems
– then we can explicitly program a computer to
do that task pretty much as humans do it.
• Our CVA algorithms fall into this category
– Can we do the converse?
– Can a human reader learn vocabulary
& improve reading comprehension
by carrying out our algorithm?
CVA (cont’d)
• Not “think like a computer”
– I.e., rigidly, mechanically, uncreatively
• But:
– what we have learned by teaching a computer to do CVA
can now help us
teach human readers who need guidance in CVA.
CVA (cont’d)
•
Research questions:
1. Can a computer algorithm be translated into a
successful curricular intervention?
a) Does this computer-based curriculum improve
(meaning) vocabulary?
CVA students vs. “typical context-based” students
vs. “no treatment” control
– Each read same passage with single unfamiliar word
& figure out a meaning from context
ii. CVA students vs. “direct-method” control
at time t
– Tested on meaning at time t > t
i.
CVA (cont’d)
1. Can a computer algorithm be translated
into a successful curricular intervention?
b) Does the algorithm-based curriculum improve
reading comprehension?
•
•
CVA students vs. “typical context-based” students
vs. no-treatment students
Test for reading comprehension on passages…
– containing unfamiliar words
– with no unfamiliar word
CVA (cont’d)
2.
Other issues:
•
•
“teacher training”/professional development
oral discussion:
–
•
•
use thinksheet as “detective”/“scientist” notebook
role & nature of prior knowledge
–
•
•
explicit instruction on using language to reason is valuable
what is needed, how to identify it, how to elicit it, etc.
test at different ages & in STEM vs. ELA
develop software for use in classroom
–
student can ask Cassie how she figured it out