cva-727.ppt
20070122
Contextual Vocabulary Acquisition:
From Algorithm to Curriculum
William J. Rapaport
Department of Computer Science & Engineering,
Department of Philosophy, Department of Linguistics,
and Center for Cognitive Science
rapaport@cse.buffalo.edu
http://www.cse.buffalo.edu/~rapaport
Contextual Vocabulary Acquisition
• Active, conscious acquisition of a meaning for a word in a
text by reasoning from “context”
• CVA = what you do when:
–
–
–
–
–
You’re reading
You come to an unfamiliar word
It’s important for understanding the passage
No one’s around to ask
Dictionary doesn’t help
•
•
•
•
No dictionary
Too lazy to look it up :-)
Word not in dictionary
Definition of no use
– Too hard
– Inappropriate
• So, you “figure out” a meaning for the word “from context”
– “figure out” = compute (infer) a hypothesis about
what the word might mean in that text
– “context” = ??
Overview of CVA Project
1. From Algorithm to Curriculum
– Implemented computational theory of how to
figure out (compute) a meaning for an
unfamiliar word from “wide context”
• Based on:
– algorithms developed by Karen Ehrlich (1995)
– verbal protocols (case studies)
• Implemented in a semantic-network-based
knowledge-representation & reasoning system
– SNePS (Stuart C. Shapiro & colleagues)
Overview of CVA Project (cont’d
2. From Algorithm to Curriculum
–
–
–
Convert algorithms to an improved, teachable
curriculum
To improve vocabulary & reading
comprehension
Joint work with Michael Kibby
•
Center for Literacy & Reading Instruction
Meaning of “Meaning”
• “the meaning of a word” vs. “a meaning for a word”
– “the” 
– “of ” 
– “a”

single, correct meaning
meaning belongs to word
many possible meanings
• depending on textual context,
reader’s prior knowledge, etc.
– “for” 
reader hypothesizes meaning
from “context”, & gives it to word
• “The meaning of things lies not in themselves
but in our attitudes toward them.”
– Antoine de Saint-Exupéry, Wisdom of the Sands (1948)
• “Words don’t have meaning;
they’re cues to meaning!”
– Jeffrey L. Elman, “On Dinosaur Bones & the Meaning of Words” (2007)
• “We cannot locate meaning in the text…; [this is an] active,
dynamic process…, existing only in interactive behaviors of
cultural, social, biological, and physical environment-systems.”
– William J. Clancey, “Scientific Antecedents of Situated Cognition”
(forthcoming)
CVA as Cognitive Science
• Studied in:
–
–
–
–
–
AI / computational linguistics
Psychology
Child-language development (L1 acquisition)
L2 acquisition (e.g., ESL)
Reading education (vocabulary development)
• Thus far: “multi-”disciplinary
• Not yet: “inter-”disciplinary!
What Does ‘Brachet’ Mean?
(From Malory’s Morte D’Arthur [page # in brackets])
1.
2.
3.
4.
10.
18.
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]
As the hart went by the sideboard,
the white brachet bit him. [66]
The knight arose, took up the brachet and
rode away with the brachet. [66]
A lady came in and cried aloud to King Arthur,
“Sire, the brachet is mine”. [66]
There was the white brachet which bayed at him fast.
The hart lay dead; a brachet was biting on his throat,
and other hounds came behind. [86]
[72]
Figure out meaning of word from what?
• context (i.e., the text)?
– Werner & Kaplan 52, McKeown 85, Schatz & Baldwin 86
• context and reader’s background knowledge?
– Granger 77, Sternberg 83, Hastings 94
• context including background knowledge?
– Nation & Coady 88, Graesser & Bower 90
• Note:
– “context” = text  context is external to reader’s mind
• Could also be spoken/visual/situative (still external)
– “background knowledge”: internal to reader’s mind
• What is (or should be) the “context” for CVA?
What Is the “Context” for CVA?
• “context” ≠ textual context
– surrounding words; “co-text” of word
• “context” = wide context =
– “internalized” co-text …
• ≈ reader’s interpretive mental model of textual “co-text”
– involves local interpretation (McKoon & Ratcliff): proN resol’n, simple infs (prop
names)
– & global interpretation (“full” use of available PK)
– can involve misinterpretation
– … “integrated” via belief revision …
• infer new beliefs from internalized co-text + prior knowledge
• remove inconsistent beliefs
– … with reader’s prior knowledge:
• “world” knowledge
• language knowledge
• previous hypotheses about word’s meaning
– but not including external sources (dictionary, humans)
 “Context” for CVA is in reader’s mind, not in the text
Some Proposed Preliminary Definitions
(to extract order out of confusion)
• Unknown word for a reader =def
– Word or phrase that reader has never seen before
– Or only has vague idea of its meaning
• Different levels of knowing meaning of word
– Notation: “X”
Proposed preliminary definitions
• Text =def
– (written) passage
– containing X
– single phrase or sentence … several paragraphs
Proposed preliminary definitions
• Co-text of X in some text =def
– The entire text “minus” X; i.e., entire text surrounding X
– E.g., if X = ‘brachet’, and text =
• “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.”
Then X’s co-text in this text =
• “There came a white hart running into the hall with a white ______ next
to him, and thirty couples of black hounds came running after them.”
– Cf. “cloze” tests in psychology
• But, in CVA, reader seeks meaning or definition
– NOT a missing word or synonym: There’s no “correct” answer!
– “Co-text” is what many mean by “context”
• BUT: they shouldn’t!
Proposed preliminary definitions
• The reader’s prior knowledge =def
– the knowledge that the reader has when s/he begins to
read the text
– and is able to recall as needed while reading
• “knight picks up & carries brachet” ? small
• Warnings:
– “knowledge”  truth
• so, “prior beliefs” is better
– “prior” vs. “background” vs. “world”, etc.
• See next slide!
Proposed preliminary definitions
• Possible synonyms for “prior knowledge”,
each with different connotation:
– Background knowledge:
• Can use for information that author assumes reader to have
– World knowledge:
• General factual knowledge about things other than the text’s topic
– Domain knowledge:
• Specialized, subject-specific knowledge about the text’s topic
– Commonsense knowledge:
• Knowledge “everyone” has
– E.g., CYC, “cultural literacy” (Hirsch)
• These overlap:
– PK should include some CSK, might include some DK
– BK might include much DK
Steps towards a
Proper Definition of “Context”
•
Step 1:
–
•
The context of X for a reader =def
1.
The co-text of X
2.
“+” the reader’s prior knowledge
Both are needed!
–
After reading:
•
“the white brachet bit the hart in the buttock”
most subjects infer that brachets are (probably) animals, from:
•
•
–
That text, plus:
Available PK premise: “If x bites y, then x is (probably) an animal.
Inference is not an enthymeme! (because …)
Proper definition of “context”:
• But (inference not an enthymeme because):
– When you read, you “internalize” the text
• You “bring it into” your mind
• Gärdenfors 1997, 1999; Jackendoff 2002
– This “internalized text” is more important than the
actual words on paper:
• Text:
• Misread as:
“I’m going to put the cat out”
“I’m going to put the car out”
– leads to different understanding of “the text”
– What matters is what the reader thinks the text is,
• Not what the text actually is
• Therefore …
On Misinterpretation
• Sign seen on truck parked outside of cafeteria
at Student Union:
Mills Wedding and Specialty Cakes
On Misinterpretation
• Sign seen on truck parked outside of
cafeteria at Student Union:
Mills Welding and Specialty Gases
Proper definition of “context”:
• Step 2:
– The context of X for a reader =def
• A single KB, consisting of:
1. The reader’s internalized co-text of X
2.
“+”
the reader’s prior knowledge
Proper definition of “context”:
• But: What is “+”?
– Not: mere conjunction or union!
– Active readers make inferences while reading.
• From text = “a white brachet”
& prior commonsense knowledge = “only physical objects have color”,
reader might infer that brachets are physical objects
• From “The knight took up the brachet and rode away with the brachet.”
& prior commonsense knowledge about size,
reader might infer that brachet is small enough to be carried
– Whole > Σ parts:
• inference from [internalized text + PK]  new info not in text or in PK
• I.e., you can learn from reading!
Proper definition of “context”:
• But: Whole < Σ parts!
– Reader can learn that some prior beliefs were mistaken
• Or: reader can decide that text is mistaken (less likely)
• Reading & CVA need belief revision!
• operation “+”:
– input: PK & internalized co-text
– output: “belief-revised integration” of input, via:
• Expansion:
– addition of new beliefs from ICT into PK, plus new inferences
• Revision:
– retraction of inconsistent prior beliefs together with inferences from them
• Consolidation:
– eliminate further inconsistencies
Prior Knowledge
PK1
PK2
PK3
PK4
Text
Prior Knowledge
PK1
PK2
PK3
PK4
Text
T1
Integrated KB
internalization
PK1
I(T1)
PK2
PK3
PK4
Text
T1
B-R Integrated KB
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
Note: All “contextual” reasoning is done in this “context”:
B-R Integrated KB
internalization
PK1
P7
Text
I(T1)
PK2
T1
T2
inference
PK3 P5
PK4
I(T2)
P6
I(T3)
T3
Note: All “contextual” reasoning is done in this “context”:
B-R Integrated KB
(the reader’s mind)
internalization
PK1
P7
Text
I(T1)
PK2
T1
T2
inference
PK3 P5
PK4
I(T2)
P6
I(T3)
T3
Proper definition of “context”:
• One more detail: X needs to be internalized
• Context is a 3-place relation among:
– Reader, word, and text
• Final(?) def.:
– Let T be a text
– Let R be a reader of T
– Let X be a word in T (that is unknown to R)
– Let T-X be X’s co-text in T.
– Then:
• The context that R should use to hypothesize a meaning for R’s
internalization of X as it occurs in T =def
– The belief-revised integration of R’s prior knowledge with R’s
internalization of T-X.
This definition agrees with…
• Cognitive-science & reading-theoretic views of text
understanding
– Schank 1982, Rumelhart 1985, etc.
• & KRR techniques for text understanding:
– Reader’s mind modeled by KB of prior knowledge
• Expressed in KR language (for us: SNePS)
– Computational cognitive agent reads the text,
• “integrating” text info into its KB, and
• making inferences & performing belief revision along the way
– When asked to define a word,
• Agent deductively searches this single, integrated KB for
information to fill slots of a definition frame
– Agent’s “context” for CVA = this single, integrated KB
Distinguishing Prior Knowledge from Integrated Co-Text
• So KB can be “disentangled” as needed for
belief revision or to control inference:
• Each proposition in the single, integrated
KB is marked with its “source”:
– Originally from PK
– Originally from text
– Inferred
• Sources of premises
Some Open Questions
• Roles of spoken/visual/situative contexts
• Relation of CVA “context” to formal theories of
context (e.g., McCarthy, Guha…)
• Relation of I(T) to prior-KB; e.g.:
– Is I(Ti) true in prior-KB?
• It is “accepted pro tem”.
– Is I(T) a “subcontext” of pKB or B-R KB?
• How to “activate” relevant prior knowledge.
• Etc.
Background of CVA Project
•
People do “incidental” (unconscious) CVA
–
Possibly best explanation of how we learn vocabulary
•
•
•
Given # of words high-school grad knows (~45K),
& # of years to learn them (~18) = ~2.5K words/year
But only taught ~10% in 12 school years
Students are taught “deliberate” (conscious) CVA
in order to improve their vocabulary
1. Computational CVA
• Implemented in SNePS (Shapiro 1979; Shapiro & Rapaport 1992)
– Intensional, propositional semantic-network
knowledge-representation, reasoning, & acting system
• Indexed by node: From any node, can describe rest of network
– Serves as model of the reader (“Cassie”)
• 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 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.
General Comments
• Cassie’s behavior  human protocols
• Cassie’s definition  OED’s definition:
= A brachet is “a kind of hound which hunts by scent”
How Does Our System Work?
• Uses a semantic network computer system
– semantic networks = “concept maps”
– serves as a model of the reader
– represents:
• reader’s prior knowledge
• the text being read
– can reason about the text and the reader’s knowledge
Fragment of reader’s prior knowledge:
m3 = In “real life”, white is a color
Member(Lex(white),Lex(color),LIFE)
m6 = In “real life”, harts are deer
AKO(Lex(hart),Lex(deer),LIFE)
m8 = In “real life”, deer are mammals
AKO(Lex(deer),Lex(mammal),LIFE)
m11 = In “real life”, halls are buildings
AKO(Lex(hall),Lex(building),LIFE)
m12 = In “real life”, b1 is named “King Arthur”
Name(b1,”King Arthur”,LIFE)
m14 = In “real life”, b1 is a king
Isa(ISA,b1,Lex(king),LIFE)
(etc.)
m16 = if
v3 has property v2 & v2 is a color & v3  v1
then v1 is a class of physical objects
all(x,y,z)({Is1(z,y),Member1(y,lex(color)),Member1
(z,x)}
&=>
{AKO1(x,lex(physical object))})
Reading the story:
m17 = In the story, b2 is a hart
ISA(b2,lex(hart),STORY)
m24 = In the story, the hart runs into b3
Does(b2,into(b3,lex(run)),STORY)
(b3 is King Arthur’s hall) – not shown
(harts are deer) – not shown
A fragment of the entire network showing the reader’s
mental context consisting of prior knowledge, the story,
& inferences.
The definition algorithm searches this entire network,
abstracts parts of it,
& produces a hypothesized meaning for ‘brachet’.
Implementation
• SNePS (Stuart C. Shapiro & SNeRG):
– Intensional, propositional semantic-network
knowledge-representation & reasoning system
– Formula-based & path-based reasoning
• I.e., logical inference & generalized inheritance
– SNeBR belief revision system
• Used for revision of definitions
– SNaLPS natural-language input/output
– “Cassie”: computational cognitive agent
How It Works
• SNePS represents:
– background knowledge + text information
in a single, consolidated semantic network
• Algorithms deductively search network for slotfillers for definition frame
• Search is guided by desired slots
– E.g., prefers general info over particular info,
but takes what it can get
The Algorithms
1. Generate initial hypothesis by
“syntactic manipulation”
•
•
Algebra: Solve an equation for unknown value X
Syntax: “Solve” a sentence for unknown word X
–
–
“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.
I.e., “define” node X in terms of immediately connected nodes
2. Deductively search wide context for more information
•
I.e., “define” word X in terms of some (but not all) other connected nodes
3. Return definition frame.
Noun Algorithm
• Generate initial hypothesis by syntactic manipulation
• Then find or infer from wide context:
– Basic-level class memberships
(e.g., “dog”, rather than “animal”)
• else most-specific-level class memberships
• else names of individuals
–
–
–
–
–
–
–
Properties of Xs (else, of individual Xs) (e.g., size, color, …)
Structure of Xs (else …) (part-whole, physical structure…)
Acts that Xs perform (else …) or that can be done to/with Xs
Agents that do things to/with Xs
… or to whom things can be done with Xs
… or that own Xs
Possible synonyms, antonyms
Verb Algorithm
• Generate initial hypothesis by syntactic manipulation
• Then find or infer from wide context:
– Class membership (e.g., Conceptual Dependency)
• What kind of act is X-ing (e.g., walking is a kind of moving)
• What kinds of acts are X-ings
(e.g., sauntering is a kind of walking)
– Properties/manners of X-ing (e.g., moving by foot, slow walking)
– Transitivity/subcategorization information
• Return class membership of agent, object, indirect object, instrument
– Possible synonyms, antonyms
– Causes & effects
• [Also: preliminary work on adjective/adverb algorithm]
Computational cognitive theory of how to learn word meanings from context (cont.)
• 3 kinds of vocabulary acquisition:
– Construct new definition of unknown word
• What does ‘brachet’ mean?
– Fully revise definition of misunderstood word
• Does “smiting” entail killing?
– Expand definition of word used in new sense
• Can you “dress” (i.e., clothe) a spear?
• Initial hypothesis;
Revision(s) upon further encounter(s);
Converges to stable, dictionary-like definition;
Subject to revision
56
QuickTime™ and a
GIF decompressor
are needed to see this picture.
Belief Revision
•
To revise definitions of words used inconsistently
with current meaning hypothesis
•
SNeBR (ATMS; Martins & Shapiro 1988, Johnson 2006):
–
If inference leads to a contradiction, then:
1.
SNeBR asks user to remove culprit(s)
2.
& automatically removes consequences inferred from culprit
Belief Revision
•
Used to revise definitions of words with different sense from
current meaning hypothesis
•
SNeBR (ATMS; Martins & Shapiro 88):
–
If inference leads to a contradiction, then:
1.
2.
•
SNeBR asks user to remove culprit(s)
& automatically removes consequences inferred from culprit
SNePSwD (SNePS w/ Defaults; Martins & Cravo 91)
–
–
•
Previously used to automate step 1, above;
Now, legacy code
AutoBR (Johnson & Shapiro, in progress [?])
& new default reasoner (Bhushan & Shapiro)
–
Will replace SNePSwD
Revision & Expansion
• Removal & revision being automated via SNePSwD by ranking all propositions
with kn_cat:
most
certain
intrinsic
story
least
certain
info re: language; fundamental background info
(“before” is transitive)
info in text
(“King Lot rode to town”)
life
background info w/o variables or inference
(“dogs are animals”)
story-comp
info inferred from text (King Lot is a king, rode on a horse)
life-rule.1
everyday commonsense background info
(BearsLiveYoung(x)  Mammal(x))
life-rule.2
specialized background info
(x smites y  x kills y by hitting y)
questionable already-revised life-rule.2; not part of input
Belief Revision: “smite”
•
•
Misunderstood word; 2-stage “subtractive” revision
Background knowledge includes:
(*) smite(x,y,t)  hit(x,y,t) & dead(y,t) & cause(hit(x,y,t),dead(y,t))
P1: King Lot smote down King Arthur
D1: If person x smites person y at time t, then x hits y at t, and y is dead at t
Q1: What properties does King Arthur have?
R1: King Arthur is dead.
P2: King Arthur drew Excalibur.
Q2: When did King Arthur do this?
• SNeBR is invoked:
– KA’s drawing E is inconsistent with being dead
– (*) replaced: smite(x,y,t)  hit(x,y,t) & dead(y,t) & [dead(y,t)  cause(hit, dead)]
D2: If person x smites person y at time t, then x hits y at t & (y is dead at t)
P3: [another passage in which ~(smiting  death)]
D3: If person x smites person y at time t, then x hits y at t
Belief Revision: “dress”
•
•
“additive” revision
Background info includes:
(1) dresses(x,y)  z[clothing(z) & wears(y,z)
(2) Spears don’t wear clothing (both kn_cat=life.rule.1)
P1: King Arthur dressed himself.
D1: A person can dress itself; result: it wears clothing.
P2: King Claudius dressed his spear.
[Cassie infers: King Claudius’s spear wears clothing.]
Q2: What wears clothing?
•
SNeBR is invoked:
–
–
–
KC’s spear wears clothing inconsistent with (2).
(1) replaced: dresses(x,y)  z[clothing(z) & wears(y,z)] v NEWDEF
Replace (1), not (2), because of verb in antecedent of (1) (Gentner)
P3: [other passages in which dressing spears precedes fighting]
D2: A person can dress a spear or a person;
result: person wears clothing or person is enabled to fight
CVA as Science & Detection
• CVA = hypothesis generation & testing
– scientific task:
• develop theory of word meaning
• not guessing, but…
– “In science, guessing is called ‘hypothesis formation’ ” (Loui)
– detective work:
• finding clues
• not “who done it”, but “what does it mean”
– susceptible to revision upon further evidence
2 Problematic Assumptions
• CVA assumes that:
– reader is consciously aware of the unfamiliar word
– reader notes its unfamiliarity
• CVA assumes that, between encounters:
– reader remembers the word
– reader remembers hypothesized meaning
I. Are All Contexts Created Equal?
• Beck, Isabel L.; McKeown, Margaret G.; &
McCaslin, Ellen S. (1983),
– “Vocabulary Development:
Not All Contexts Are Created Equal”
– Elementary School Journal 83(3): 177-181.
• “it is not true that every context is an appropriate
or effective instructional means for vocabulary
development”
Role of Prior Knowledge
• Beck et al:
– co-text “can give clues to the word’s meaning”
• But “clue” is relative:
– clues need other info to be interpreted as clues
• Implication A1:
– textual clues need to be supplemented with other information
to compute a meaning.
• Supplemental info = reader’s prior knowledge
– has to be available to reader
– will be idiosyncratic
Do Words Have Unique, Correct Meanings?
• Beck et al. (& others) assume:
– A2: A word has a unique meaning
– A3: A word has a correct meaning
• Contra “unique”: A word’s meaning varies with:
– co-text
– reader(’s prior knowledge)
– time of reading
• “Correct” is a red herring (in any case, it’s fishy):
– Possibly, words have author-intended meanings
• but these need not be determined by co(n)text
– Misunderstandings are universally unavoidable
– Perfect understanding/dictionary definition not needed
• “satisficing” understanding for passage comprehension suffices
• reader always has opportunity of revising definition hypothesis
Beck et al.’s
Categories of Textual Contexts
• What kinds of co-texts are helpful?
• But keep in mind that we have different goals:
– Beck et al.:
• use co-text to teach “correct” word meanings
– CCVA:
• use context to compute word meaning for understanding
Beck et al.’s Textual Context Categories
Top-Level Kinds of Co-Text
• Pedagogical co-texts:
– artificially constructed, designed for teaching
– only example is for a verb:
• “All the students made very good grades on the tests, so
their teacher commended them for doing so well.”
• Natural co-texts:
– “not intended to convey the meaning of a word”
– 4 kinds (actually, a continuum)
Beck et al.’s Textual Context Categories
1. Misdirective (Natural) Co-Texts
• “seem to direct student to incorrect meaning for a word”
• sole example:
– “Sandra had won the dance contest and the audience’s cheers brought her to
the stage for an encore. ‘Every step she takes is so perfect and graceful,’
Ginny said grudgingly, as she watched Sandra dance.”
– [[grudgingly]] =? admiringly
• Is this a natural context?
• Is this all there is to it?..
– A4: Co-texts have a fixed, usually small size
– But larger co-text might add information
– Prior knowledge can widen the co(n)text
• ‘grudgingly’ is an adverb!
– A5: All words are equally easy to learn
– But N easier than V, V easier than Adj/Adv! (Granger/Gentner/..Gleitman..)
• A6: Only 1 co-text can be used.
– But later co-texts can assist in refining meaning
Beck et al.’s Textual Context Categories
2. Nondirective (Natural) Co-Texts
• “of no assistance in directing the reader toward any
particular meaning for a word”
• sole example is for an adjective:
– “Dan heard the door open and wondered who had arrived. He
couldn’t make out the voices. Then he recognized the lumbering
footsteps on the stairs and knew it was Aunt Grace.”
• But:
–
–
–
–
Is it natural?
What about larger co-text?
An adjective!
Of no assistance? (see next slide)
Syntactic Manipulation
• Do misdirective & nondirective contexts yield no (or
only incorrect) information?
• Cf. algebraic manipulation (brings x into focus):


2x + 1 = 7
x = (7 − 1)/2 = 6/2 = 3
• Syntactic manipulation (bring hard word into focus):
• “ ‘Every step she takes is so perfect and graceful,’ Ginny
said grudgingly.”
• ‘Grudgingly’ is the way that Ginny said “…”
• So, ‘grudgingly’ is a way of saying something
• In particular, ‘grudgingly’ is a way of (apparently) praising
someone’s performance
• “he recognized the lumbering footsteps on the stairs”
• ‘lumbering’ is a property of footsteps on stairs
Beck et al.’s Textual Context Categories
3. General (Natural) Co-Texts
• “provide enough information for reader to place word in a
general category”
• sole example is for an adjective:
– “Joe and Stan arrived at the party at 7:00. By 9:30 the evening
seemed to drag for Stan. But Joe really seemed to be having a
good time at the party. ‘I wish I could be as gregarious as he is,’
thought Stan”
• Same problems, but:
– adjective is contrasted with Stan’s attitude
– contrasts are good (so are parallel constructions)
Beck et al.’s Textual Context Categories
4. Directive (Natural) Co-Texts
• “seem likely to lead the student to a specific, correct meaning
for a word”
• sole example is for a noun:
– “When the cat pounced on the dog, he leapt up, yelping, and knocked
over a shelf of books. The animals ran past Wendy, tripping her. She
cried out and fell to the floor. As the noise and confusion mounted,
Mother hollered upstairs, ‘What’s all the commotion?’ ”
• Natural? Long!
• Noun!
– note that the sole example of a directive context is a noun, suggesting
that it might be the word that makes a context directive
Beck et al.’s Experiment
• S’s given passages from basal readers
• Researchers categorized co-texts & blacked out words
• S’s asked to “fill in the blanks with the missing words or reasonable synonyms”
• Results confirm 4 co-text types
• Independently of results, there are methodological questions:
– Are basal readers natural contexts?
– How large were co-texts?
– Instruction on how to do CVA?
• A7: CVA “comes naturally”, so needs no training
– A8: Fill-in-the-blank tasks are a form of CVA
• No, they’re not! (see next slide)
Beck et al.’s Experiment
CVA, Neologisms, & Fill-in-the-Blank
• Serious methodological problem for all of us:
– What if S knows the unknown word?
– Filter out such S’s and words?
• hard to do; what about testing familiar words?
– Replace word with made-up “neologism”?
• must be carefully chosen
– Replace word with blank?
• both kinds of replacement mislead S to find
“correct missing/hidden word”
• ≠ CVA!
• Our (imperfect) solution:
– use plausible-sounding neologism
– tell S it’s like a foreign word with no English equivalent,
hence need a descriptive phrase
Beck et al.’s Conclusion
• “less skilled readers … receive little benefit from” CVA
• A9: CVA can only help in learning correct meanings.
• But:
– CVA uses same techniques as general reading comprehension:
•
•
•
•
•
careful, slow reading
careful analysis of text
directed search for information useful for computing a meaning
application of relevant prior knowledge
application of reasoning for purpose of extracting information from text
– CVA, if properly taught & practiced, can improve general
reading comprehension
II. Are Context Clues Unreliable
Predictors of Word Meanings?
• Schatz, Elinor; & Baldwin, R. Scott (1986):
– “Context Clues Are Unreliable Predictors of Word Meanings”
– Reading Research Quarterly 21(4): 439-453.
• “context does not usually provide clues to the
meanings of low-frequency words”
• “context clues inhibit the correct prediction of word
meanings just as often as they facilitate them”
S&B’s Argument
• A10: CVA is not an efficient mechanism for inferring word
meanings.
• Because:
– Co-text can’t help you figure out the correct meaning of an unfamiliar
word.
– (assumptions A2 & A3 again!)
• But:
– Wide context can help you figure out a meaning for an unfamiliar word.
– So, context (& CVA) are efficient mechanisms for inferring (better:
computing) word meanings.
Incidental vs. Deliberate CVA
• S&B:
– “context clues should help readers to infer meanings of words
without the need for readers to interrupt the reading act(*) with
diversions to external sources”
• (*) true for incidental CVA
• (*) not for deliberate CVA
• External sources are no solution anyway:
– Dictionary definitions are just more co-texts! (Schwartz 1988)
– CVA is base case of recursion, one of whose recursive clauses is:
“Look it up in a dictionary”
Why not use a dictionary?
Because:
• People are lazy (!)
• Dictionaries are not always available
• Dictionaries are always incomplete
• Dictionary definitions are not always useful
– ‘chaste’ =df clean, spotless / “new dishes are chaste”
– ‘college’ =df a body of clergy living together and
supported by a foundation
• Most words are learned via incidental CVA,
not via dictionaries
• Most importantly:
– Dictionary definitions are just more contexts!
Why not use a dictionary?
•
Merriam-Webster New Collegiate Dictionary:
– chaste.
1. innocent of unlawful sexual intercourse
– student: stay away from that one!
2. celibate
– student: huh?
3. pure in thought and act: modest
– student: I have to find a sentence for that?
4. a: severely simple in design or execution: austere
– student: huh? “severely”? “austere”?
b: clean, spotless
– student: all right!: “The plates were still chaste after much use.”
–
Deese 1967 / Miller 1985
Why not use a dictionary?
•
Merriam-Webster (continued):
– college.
1. a body of clergy living together and supported by a
foundation
2. a building used for an educational or religious
purpose
3. a: a self-governing constituent body of a
university offering living quarters and
instruction but not granting degrees…
b: a preparatory or high school
c: an independent institution of higher
learning offering a course of general
studies leading to a bachelor’s degree…
– Problem: ordering is historical!
Why not use a dictionary?
• Merriam-Webster (continued):
– infract:
– infringe:
– encroach:
infringe
encroach
• to enter by gradual steps or by stealth into the
possessions or rights of another
• to advance beyond the usual or proper limits;
trespass
Why not use a dictionary?
•
Collins COBUILD Dictionary
–
–
“Helping Learners with Real English”
chaste.
1. Someone who is chaste does not have sex with
anyone, or only has sex with their husband or
wife; an old-fashioned use, used showing
approval. EG She was a holy woman, innocent and
chaste.
2. Something that is chaste is very simple in style,
without much decoration. EG …chaste houses built
in 1732
Why not use a dictionary?
•
Collins COBUILD Dictionary
–
college.
1.
•
•
infract
infringe.
1.
•
A college is 1.1 an institution where students study for
qualifications or do training courses after they have left school. …
[not in dictionary]
If you infringe a law or an agreement, you break it.
encroach.
1.
To encroach on or upon something means to slowly take
possession or control of it, so that someone else loses it bit by bit.
S&B’s Experiments
• 25 natural passages from novels
• words chosen (the only cited examples):
– Adj/Adv
– N
– V
~67%
~27%
~ 6%
• But:
– what are actual %s?
– which lexical categories were hardest?
– how do facilitative/confounding co-texts correlate with
lexical category?
– should have had representative sample of
4 co-text categories X 3 or 4 lexical categories
S&B’s Experiments
CVA vs. Word-Sense Disambiguation
• 2 experiments:
– S’s chose “correct” meaning from list of 5 possible
meanings
– This is WSD, not CVA!
• WSD = multiple choice
• CVA = essay question
• 3rd experiment:
– real CVA, but “interested only in full denotative
meanings or accurate synonyms”
– cf. assumption A3 about “correct” meanings!
S&B’s Experiments
Space & Time Limits
• Space limits on size of co-text?
– S&B: 3 sentences
– CCVA: start small, work “outward”
• Time limits on size of co-text?
– S&B: “all students finished in allotted time”
– CCVA: no time limits
S&B’s Experiments
Teaching CVA
• S&B: “did not control for S’s knowledge of
how to use context clues”
• CCVA:
– deliberate CVA is a skill
• needs to be taught, modeled, & practiced
– there is other (later) evidence that such training
works
• including “critical thinking” education
S&B’s 3 Questions
1.
(answered in the negative)
“Do context clues occur with sufficient frequency to
justify them as a major element of reading instruction?”
a)
b)
2.
“Does context usually provide accurate clues to
denotations & connotations of low-frequency words?”
a)
b)
3.
Irrelevant if CVA fosters good reading comprehension & critical
thinking skills
Context clues do occur & teaching them is justified, if augmented by
reader’s prior knowledge & knowledge of CVA skills.
Irrelevant under our conception of CVA: accuracy not needed
CVA can provide clues to revisable hypotheses about unfamiliar
word’s meaning
Are “difficult words in natural [co-texts] usually
amenable to such analysis?”
a)
Such words are always amenable to yielding at least some
information about their meaning.
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
State of the Art: Computational Linguistics
• Information extraction systems
• Autonomous intelligent agents
• There can be no complete lexicon
• Such systems/agents shouldn’t have
to stop to ask questions
94
QuickTime™ and a
GIF decompressor
are needed to see this picture.
State of the Art: Computational Linguistics
•
•
•
•
Granger 1977: “Foul-Up”
– Based on Schank’s theory of “scripts” (schema theory)
– Our system not restricted to scripts
Zernik 1987: self-extending phrasal lexicon
– Uses human informant
– Ours system is really “self-extending”
Hastings 1994: “Camille”
– Maps unknown word to known concept in ontology
– Our system can learn new concepts
Word-Sense Disambiguation:
– Given ambiguous word & list of all meanings, determine the
“correct” meaning
• Multiple-choice test 
– Our system: given new word, compute its meaning
• Essay question 
95
QuickTime™ and a
GIF decompressor
are needed to see this picture.
State of the Art: Vocabulary Learning (I)
• Elshout-Mohr/van Daalen-Kapteijns 1981,1987:
– Application of Winston’s AI “arch” learning theory
– (Good) reader’s model of new word = frame
• Attribute slots, default values
• Revision by updating slots & values
– Poor readers update by replacing entire frames
– But EM & vDK used:
• Made-up words
• Carefully constructed contexts
– Presented in a specific order
96
QuickTime™ and a
GIF decompressor
are needed to see this picture.
Elshout-Mohr & van Daalen-Kapteijns
Experiments with neologisms in 5 artificial contexts
•
•
•
•
•
When you are used to a view it is depressing when you live in a room with
kolpers.
– Superordinate information
At home he had to work by artificial light because of those kolpers.
During a heat wave, people want kolpers, so sun-blind sales increase.
– Contexts showing 2 differences from the superordinate
I was afraid the room might have kolpers, but plenty of sunlight came into
it.
This house has kolpers all summer until the leaves fall out.
– Contexts showing 2 counterexamples due to the 2 differences
97
QuickTime™ and a
GIF decompressor
are needed to see this picture.
State of the Art: Psychology
• Johnson-Laird 1987:
– Word understanding  definition
– Definitions aren’t stored
– “During the Renaissance, Bernini
cast a bronze of a mastiff eating
truffles.”
98
QuickTime™ and a
GIF decompressor
are needed to see this picture.
State of the Art: Psychology
• Sternberg et al. 1983,1987:
– Cues to look for (= slots for frame):
•
•
•
•
•
•
•
Spatiotemporal cues
Value cues
Properties
Functions
Cause/enablement information
Class memberships
Synonyms/antonyms
– To acquire new words from context:
• Distinguish relevant/irrelevant information
• Selectively combine relevant information
• Compare this information with previous beliefs
99
QuickTime™ and a
GIF decompressor
are needed to see this picture.
Sternberg
• The couple there on the blind date was
not enjoying the festivities in the least.
An acapnotic, he disliked her smoking;
and when he removed his hat, she, who
preferred “ageless” men, eyed his
increasing phalacrosis and grimaced.
100
QuickTime™ and a
GIF decompressor
are needed to see this picture.
From Algorithm to Curriculum
•
State of the art in vocabulary learning from
context:
–
–
Mauser 1984: “context” = definition!
Clarke & Nation 1980: a “strategy” (algorithm?):
1. Determine part of speech of word
2. Look at grammatical context
•
Who does what to whom?
3. Look at surrounding textual context
•
Search for clues (as we do)
4. Guess the word; check your guess
CVA: From Algorithm to Curriculum
•
“guess the word”
=
“then a miracle occurs”
•
Surely, computer scientists
can “be more explicit”!
•
And so should teachers!
Terminology: “Guessing”?
• Does reader …
– “guess” a meaning?!
• not computational!
– “deduce” a meaning?
• too restrictive; ignores other kinds of inference
– “infer” a meaning?
• too vague; ignores other kinds of reasoning (cf. Herbert Simon)
– “figure out” a meaning?
• just vague enough?
• My preference:
– The reader computes a meaning!
A More Precise, Teachable Algorithm
• Treat “guess” as a procedure call
– Fill in the details with our algorithm
• Convert the algorithm into a curriculum
– To enhance students’ abilities to use deliberate
CVA strategies
From Algorithm to Curriculum (cont’d)
• We have explicit, GOF (symbolic) AI theory of how to do CVA
 Teachable!
• Goal:
– Not:
teach people to “think like computers”
– But:
explicate computable & teachable methods
to hypothesize word meanings from context
• AI as computational psychology:
– Devise computer programs that faithfully simulate
(human) cognition
– Can tell us something about (human) mind
• Joint work with Michael Kibby (UB Reading Clinic)
– We are teaching a machine, to see if what we learn in
teaching it can help us teach students better
CVA: From algorithm to curriculum …
• Treat “guess” as a procedure call (“subroutine”)
– Fill in the details with our algorithm
– Convert the algorithm into a curriculum
• To enhance students’ abilities to use deliberate CVA strategies
– To improve reading comprehension
… and back again!
• Use knowledge gained from CVA case studies to
improve the algorithm
• I.e., use Cassie to learn how to teach humans
& use humans to learn how to teach Cassie
106
QuickTime™ and a
GIF decompressor
are needed to see this picture.
Problem in Converting
Algorithm into Curriculum
• “A knight picks up a brachet and carries it away …”
• Cassie:
– Has “perfect memory”
– Is “perfect reasoner”
– Automatically infers that brachet is small
• People don’t always realize this:
– May need prompting: How big is the brachet?
– May need relevant background knowledge
– May need help in drawing inferences
• Teaching CVA =? teaching general reading comprehension
– Vocabulary knowledge correlates with reading comprehension
107
QuickTime™ and a
GIF decompressor
are needed to see this picture.
Ongoing Research:
From Algorithm to Curriculum
• more robust algorithms
–
–
–
–
–
–
better N coverage needed
much better V coverage needed
no general adjective/adverb coverage yet
need “internal” context (morphology, etc.)
need NL interface
need acting component
• need curriculum
– CVA taught, but not well (emphasis on “guessing”)
– we have explicit, teachable theory of how to do CVA
– joint work w/ Michael Kibby, UB/LAI/Reading Clinic
Outline of a Curriculum
• http://www.cse.buffalo.edu/~rapaport/CVA/curriculum-outline.html
Future Work
• Is there any general prior knowledge that is especially
useful?
• Are there any limitations (besides PK) to our definition
algorithms?
– Need to test more words to find out answers to both
• Convert Lisp algorithms to SNeRE
• Refine and test curriculum
• Applications?
– web searches, text processing, computational lexicography…
Computation & Philosophy
• Computational philosophy =
– Application of computational (i.e., algorithmic) solutions
to philosophical problems
• Use of SNePS KRR & belief-revision system
to solve problems in representation of fictional entities
– Rapaport 1991; Rapaport & Shapiro 1995, 1999
• CVA
• Philosophical computation =
– Application of philosophy to CS problems
• Use of Castañeda’s theory of quasi-indexicals
to solve problems in knowledge representation
– Maida & Shapiro 1982; Rapaport 1986; Rapaport, Shapiro, & Wiebe 1997
• CVA
• (more later)
CVA as Computational Philosophy & Philosophical Computation
1.
CVA & holistic semantic theories:
–
Semantic networks:
•
–
Holism:
•
–
Meaning of a word is its relationships to all other words in the language
Problems (Fodor & Lepore):
•
•
•
•
•
•
–
“Meaning” of a node is its location in the entire network
No 2 people ever share a belief
No 2 people ever mean the same thing
No 1 person ever means the same thing at different times
No one can ever change his/her mind
Nothing can be contradicted
Nothing can be translated
CVA offers principled way to restrict “entire network”
to a useful subnetwork
•
•
That subnetwork can be shared across people, individuals, languages,…
Can also account for language/concept change
–
Via “dynamic”/“incremental” semantics
CVA as Computational Philosophy & Philosophical Computation (cont’d)
2.
CVA and the Chinese Room
–
How would Searle-in-the-Room figure out the meaning of an unknown
squiggle?
•
–
Searle’s CR argument from semantics:
1.
2.
3.

–
By CVA techniques!
Computer programs are purely syntactic
Cognition is semantic
Syntax alone does not suffice for semantics
No purely syntactic computer program can exhibit semantic cognition
“Syntactic Semantics”
•
(Rapaport 1985ff)
Syntax does suffice for the kind of semantics needed for NLU in the CR
–
–
All input—linguistic, perceptual, etc.—is encoded in a single network
(or: in a single, real neural network: the brain!)
Relations—including semantic ones—among nodes of such a network
are manipulated syntactically
» Hence computationally (CVA helps make this precise)
Summary
• Contextual Vocabulary Acquisition
project is:
– Computational philosophy
• And computational psychology!
– Philosophical computation
– With applications to:
• Computational linguistics
• Reading education