Chicago, May 25, 2005 1 Adventures with Camille, a Computational Simulation of a Minimalist Language Learner Peter W. Culicover (The Ohio State University)

Chicago, May 25, 2005Chicago, May 25, 2005 11

Adventures with Camille, Adventures with Camille, a Computational a Computational

Simulation of a Minimalist Simulation of a Minimalist Language LearnerLanguage Learner

Peter W. Culicover (The Ohio State Peter W. Culicover (The Ohio State University)University)

collaborating withcollaborating withAndrzej Nowak (Warsaw / FAU )Andrzej Nowak (Warsaw / FAU )Wojciech Borkowski (Warsaw)Wojciech Borkowski (Warsaw)

Piotr Kochanski (Warsaw)Piotr Kochanski (Warsaw)


Thanks to the James S. Mcdonnell Thanks to the James S. Mcdonnell Foundation, the Center for Cognitive Foundation, the Center for Cognitive Science at OSU & the Center for Science at OSU & the Center for Complex Systems at U. of Warsaw for Complex Systems at U. of Warsaw for their support.their support.


QuestionsQuestions

How should we think about what How should we think about what knowledge of language consists of?knowledge of language consists of?

What is the architecture of the What is the architecture of the learner that explains how we have learner that explains how we have this knowledge?this knowledge?

Can we simulate language Can we simulate language acquisition to explore this view?acquisition to explore this view?


OutlineOutline

Perspective: Simpler Syntax; the Perspective: Simpler Syntax; the Constructionalist ManifestoConstructionalist Manifesto

Facts: Syntactic Nuts and the Facts: Syntactic Nuts and the Architecture of the Language FacultyArchitecture of the Language Faculty

Learner: Learner: CAMiLLe CAMiLLe and Dynamical and Dynamical Grammar; computational simulationGrammar; computational simulation

Social network simulationSocial network simulation


PerspectivesPerspectives

On the standard view in linguistic On the standard view in linguistic theory, a generative grammar of a theory, a generative grammar of a language is an idealized language is an idealized characterization of the linguistic characterization of the linguistic knowledge, or linguistic competence, knowledge, or linguistic competence, of an idealized native speaker. of an idealized native speaker.

A theory of generative grammar in A theory of generative grammar in turn is a characterization of possible turn is a characterization of possible generative grammars. generative grammars.


The standard viewThe standard view

On this view, a generative grammar is a theory On this view, a generative grammar is a theory something that really exists, the linguistic something that really exists, the linguistic competence of the native speaker. competence of the native speaker.

And the theory of grammar is a theory about And the theory of grammar is a theory about the language faculty that explains why the language faculty that explains why linguistic competence takes the form that it linguistic competence takes the form that it does. does.

Both the grammar and the theory of grammar Both the grammar and the theory of grammar are in some sense "in the mind" and form a are in some sense "in the mind" and form a part of the explanation of human linguistic part of the explanation of human linguistic abilities.abilities.


An alternative An alternative viewview

What resides in the language faculty is What resides in the language faculty is actually something of a different sort actually something of a different sort that reflects in its architecture the that reflects in its architecture the dynamical character of language. dynamical character of language.

In particular, it directly reflects the fact In particular, it directly reflects the fact that the grammar is acquired over time, that the grammar is acquired over time, that it is a psychological mechanism that it is a psychological mechanism used for speaking and understanding, used for speaking and understanding, and that it undergoes change over time.and that it undergoes change over time.


The challenge... The challenge...

...for an approach to human ...for an approach to human language that does not incorporate a language that does not incorporate a standard version of generative standard version of generative grammar is to account for the grammar is to account for the linguistic phenomena themselves. linguistic phenomena themselves.

Such an alternative must be able to Such an alternative must be able to explain what generative grammar explain what generative grammar explains, and more. explains, and more.


The challenge, cont’dThe challenge, cont’d Minimally, the linguistic content of the Minimally, the linguistic content of the

alternative formulation must be at least alternative formulation must be at least equivalent to the content of the equivalent to the content of the generative grammar.generative grammar.

That is, both must provide the same (or That is, both must provide the same (or equivalent) empirically adequate equivalent) empirically adequate structural descriptions of words, phrases, structural descriptions of words, phrases, idioms, constructions and sentences.idioms, constructions and sentences.

Moreover, such an account should Moreover, such an account should explain the dynamical properties of explain the dynamical properties of language.language.


A Constructionist ManifestoA Constructionist Manifesto

The Neural Basis of Cognitive Development: A The Neural Basis of Cognitive Development: A Constructionist Manifesto. S. R. Quartz and T. J. Constructionist Manifesto. S. R. Quartz and T. J. Sejnowski. 1997.Sejnowski. 1997. Behavioural and Brain Behavioural and Brain Sciences 20(4): 537 -596. Sciences 20(4): 537 -596.

““In contrast to learning as selective induction, the central In contrast to learning as selective induction, the central component of the constructivist model is that it does not component of the constructivist model is that it does not involve a search through an involve a search through an a prioria priori defined hypothesis defined hypothesis space, and so is not an instance of model-based estimation, space, and so is not an instance of model-based estimation, or parametric regression. Instead, the constructivist learner or parametric regression. Instead, the constructivist learner builds this hypothesis as a process of activity-dependent builds this hypothesis as a process of activity-dependent construction of the representations that underlie mature construction of the representations that underlie mature skills.”skills.”


ConstructionalismConstructionalismWe adopt a Jackendoffian perspective on grammar (e.g. Culicover, We adopt a Jackendoffian perspective on grammar (e.g. Culicover,

Syntactic Nuts, Syntactic Nuts, Oxford, 1999; Jackendoff, Oxford, 1999; Jackendoff, Foundations of LanguageFoundations of Language, , Oxford, 2001; Culicover & Jackendoff, Oxford, 2001; Culicover & Jackendoff, Simpler Syntax Simpler Syntax Oxford, to Oxford, to appear), which is a constructionalist one:appear), which is a constructionalist one:a.a. The job of grammar is to describe the sound-meaning The job of grammar is to describe the sound-meaning correspondences.correspondences.b.b. Some of these correspondences are unanalyzable (words).Some of these correspondences are unanalyzable (words).c.c. Some have linguistic structure but are simple or not entirely Some have linguistic structure but are simple or not entirely transparent on the meaning side (idioms) (no nice structure/meaning transparent on the meaning side (idioms) (no nice structure/meaning matchups).matchups).d.d. Some have structure and are transparent on the meaning side Some have structure and are transparent on the meaning side (compositional semantics interpreting canonical phrase structure).(compositional semantics interpreting canonical phrase structure).e.e. Some are a combination of the above ('constructions'), Some are a combination of the above ('constructions'), ranging from quasi-idioms, double-objects, movement along a path ranging from quasi-idioms, double-objects, movement along a path expressions, syntactic nuts (see above), various operator-trace binding expressions, syntactic nuts (see above), various operator-trace binding constructions, etc. Each has some degree of predictability and constructions, etc. Each has some degree of predictability and generality, and some idiosyncrasies.generality, and some idiosyncrasies.


Simple SyntaxSimple Syntax

Simple(r) Syntax Hypothesis (SSH):Simple(r) Syntax Hypothesis (SSH):

The most explanatory syntactic theory is The most explanatory syntactic theory is one that imputes the minimum one that imputes the minimum structurestructure necessary to mediate between sound and necessary to mediate between sound and meaning.meaning.


The Correspondence SpectrumThe Correspondence Spectrum

I’ll go through these quickly just to give you a I’ll go through these quickly just to give you a feel for what I’m talking about...feel for what I’m talking about...


WordsWords

Many words are unanalyzable correspondences between sound Many words are unanalyzable correspondences between sound and meaning.and meaning.

– (although some (e.g. Hale & Keyser) have argued (although some (e.g. Hale & Keyser) have argued that apparently simple words are syntactically that apparently simple words are syntactically complex and are the product of derivations complex and are the product of derivations involving movement and deletion.)involving movement and deletion.)

– (but the relations captured by such derivations can (but the relations captured by such derivations can be captured in non-derivational (constructionist) be captured in non-derivational (constructionist) ways, and the latter are ways, and the latter are requiredrequired for certain for certain aspects of the correspondences.)aspects of the correspondences.)


IdiomsIdioms

Idioms have recognizable syntactic Idioms have recognizable syntactic structure but unpredictable meaning:structure but unpredictable meaning:

by and largeby and large lo and beholdlo and behold beat a dead horsebeat a dead horse make amendsmake amends cast aspersions on (*at / *to)cast aspersions on (*at / *to) a flash in the pana flash in the pan put up withput up with


VP constructional idiomsVP constructional idiomsa.a. WayWay-construction (Jackendoff 1990, Goldberg 1995):-construction (Jackendoff 1990, Goldberg 1995):

Elmer hobbled/laughed/joked his way to the bankElmer hobbled/laughed/joked his way to the bank. . (( ‘Elmer went/made his way to the bank hobbling/laughing ‘Elmer went/made his way to the bank hobbling/laughing

/joking’])/joking’])

b.b. Time-Time-awayaway construction (Jackendoff 1997b): construction (Jackendoff 1997b):Hermione slept/drank/sewed/programmed three whole evenings Hermione slept/drank/sewed/programmed three whole evenings

away.away.(( ‘Hermione spent three whole evenings sleeping/drinking/sewing ‘Hermione spent three whole evenings sleeping/drinking/sewing

/programming’)/programming’)

c.c. Sound+motion construction (Levin and Rappaport Hovav 1995):Sound+motion construction (Levin and Rappaport Hovav 1995):The car whizzed/rumbled/squealed past Harry.The car whizzed/rumbled/squealed past Harry.(( ‘the car went past Harry, making whizzing/rumbling/squealing ‘the car went past Harry, making whizzing/rumbling/squealing

noises’)noises’)

d.d. Resultative constructionResultative constructionThe chef cooked the pot blackThe chef cooked the pot black..(( ‘the chef made the pot black by cooking in/with it’) ‘the chef made the pot black by cooking in/with it’)


Syntax-semantics mismatchesSyntax-semantics mismatches

All of these constructions share the same All of these constructions share the same basic syntax (not surprisingly, since basic syntax (not surprisingly, since they are all English); what is they are all English); what is idiosyncratic is the way in which their idiosyncratic is the way in which their meanings are related to the meanings meanings are related to the meanings of the parts and to the structure in of the parts and to the structure in which they (the parts) appear.which they (the parts) appear.


Motto: Construction of language Motto: Construction of language produces constructions in languageproduces constructions in language

which means...which means... as knowledge of language is constructed as knowledge of language is constructed

dynamically by a learner,dynamically by a learner, what emerges are constructions that may what emerges are constructions that may

ultimately become ‘rules’, but only if given ultimately become ‘rules’, but only if given enough evidence and a suitable generalization enough evidence and a suitable generalization mechanism,mechanism,

otherwise, they remain constructions.otherwise, they remain constructions.


How does the learner know what How does the learner know what she/he is dealing with?she/he is dealing with?

Since there is no way for the learner Since there is no way for the learner to know where on the spectrum a to know where on the spectrum a correspondence really is, the correspondence really is, the conservative strategy is to start at conservative strategy is to start at the word/idiom end, and then move the word/idiom end, and then move away as the weight of the evidence away as the weight of the evidence warrants generalization. (Tomasello)warrants generalization. (Tomasello)


CAMiLLeCAMiLLe

CC onservative (or Concrete)onservative (or Concrete)(don’t generalize much beyond the (don’t generalize much beyond the

evidence)evidence)

AA ttentivettentive(all input is potentially relevant)(all input is potentially relevant)

MiMi nimalist nimalist LL anguage anguage LeLe arnerarner


Goals of Goals of CAMiLLeCAMiLLe

Pursuing the logic of Concrete Minimalism (and Pursuing the logic of Concrete Minimalism (and Simpler SyntaxSimpler Syntax), we constructed ), we constructed CAMiLLeCAMiLLe with with minimal prior knowledge of linguistic structure.minimal prior knowledge of linguistic structure.

Language acquisition by Language acquisition by CAMiLLe CAMiLLe is intended to is intended to simulate the formation of trajectories and flows, simulate the formation of trajectories and flows, and self-organization, in a dynamical system.and self-organization, in a dynamical system.

Our experiments with Our experiments with CAMiLLe CAMiLLe are intended to are intended to determine how much grammatical knowledge determine how much grammatical knowledge such a minimalist learner is capable of acquiring such a minimalist learner is capable of acquiring strictly from sound/meaning pairings.strictly from sound/meaning pairings.


The spatial metaphorThe spatial metaphorRegions = categories/featuresRegions = categories/features

(slice of a many dimensional space)(slice of a many dimensional space) Sound/meaning Sound/meaning

pairs are computed pairs are computed as the system as the system changes states, changes states, represented as represented as trajectories in a trajectories in a space.space.

Structure guides Structure guides movement from movement from one region of the one region of the space to another.space to another.


N -> A NN -> A N

N

A N


A

N

Chicago, May 25, 2005Chicago, May 25, 2005 2525A

N

N -> A NN -> A N

Recursion


FlowFlow

Multiple trajectories Multiple trajectories from one region to from one region to another create another create flows.flows.

If the flow is If the flow is restricted in the restricted in the space it is a space it is a construction.construction.


Flow -> RuleFlow -> Rule

The individual The individual trajectories may trajectories may carve out a broad carve out a broad region of the entire region of the entire trajectory space.trajectory space.

If there are enough If there are enough of them, we could of them, we could fill in the empty fill in the empty spaces & make a spaces & make a rule.rule.


Self-organizationSelf-organization

CAMiLLeCAMiLLe, as a dynamical system, should self-, as a dynamical system, should self-organize when it is possible to collapse organize when it is possible to collapse individual rules or representations. This individual rules or representations. This produces generalization and over-produces generalization and over-generalization.generalization.

(Self-organization is limited in the current (Self-organization is limited in the current implementation of implementation of CAMiLLeCAMiLLe, it should be , it should be noted.)noted.)


StrategyStrategy

Pursuing the logic of Concrete Pursuing the logic of Concrete Minimalism, we constructed Minimalism, we constructed CAMiLLeCAMiLLe with minimal prior with minimal prior knowledge of linguistic structure.knowledge of linguistic structure.

Our experiments with Our experiments with CAMiLLe CAMiLLe are are intended to determine how much intended to determine how much grammatical knowledge such a grammatical knowledge such a minimalist learner is capable of minimalist learner is capable of acquiring.acquiring.


Success would be nice, but some Success would be nice, but some failure can be quite informative.failure can be quite informative.


A Concrete Minimalist LearnerA Concrete Minimalist Learner

We assume that the learner has We assume that the learner has access toaccess to– sounds and the phonological system;sounds and the phonological system;– meanings and the Conceptual Structure meanings and the Conceptual Structure

system;system;– the minimal prior information about the minimal prior information about

grammar that is necessary to acquire a grammar that is necessary to acquire a descriptively adequate grammar given descriptively adequate grammar given the paired sounds and meanings of a the paired sounds and meanings of a language. language.

To begin, we assume, counterfactually, that this information is To begin, we assume, counterfactually, that this information is ZERO, and see what needs to be added. ZERO, and see what needs to be added.


Representations in Representations in CAMiLLeCAMiLLe

There are two representational systems in There are two representational systems in CAMiLLeCAMiLLe: : – the system that encodes meaning, the system that encodes meaning, – and the system that encodes form.and the system that encodes form.

Meaning is encoded as a structured list of arguments Meaning is encoded as a structured list of arguments and adjuncts, where thematic roles and modifiers are and adjuncts, where thematic roles and modifiers are explicitly specified.explicitly specified.

Syntax (form) is simply the linear arrangement of Syntax (form) is simply the linear arrangement of elements (words and morphemes) elements (words and morphemes)


MeaningMeaning A meaning in the CS presented to A meaning in the CS presented to CAMiLLeCAMiLLe is expressed in a is expressed in a

simple attribute-value language. E.g.,simple attribute-value language. E.g.,

TOUCH($AGENT:MAN,$THEME:ANIMAL)TOUCH($AGENT:MAN,$THEME:ANIMAL)

Relations typically expressed by verbs are represented as Relations typically expressed by verbs are represented as constants with an associated argument structure. constants with an associated argument structure.

Arguments are given as thematic roles with their values. (Like Arguments are given as thematic roles with their values. (Like $AGENT:MAN$AGENT:MAN))

We assume that the meaning that We assume that the meaning that CAMiLLeCAMiLLe is presented with is presented with contains only primitives that are cognitively accessible to contains only primitives that are cognitively accessible to CAMiLLeCAMiLLe at a given stage of development. at a given stage of development.


Cognitive DevelopmentCognitive Development

At the earliest stage of developmentAt the earliest stage of development CAMiLLe CAMiLLe (simulating an actual child) may only perceive (simulating an actual child) may only perceive that some man touches some animal. that some man touches some animal.

For example, For example, John touches the catJohn touches the cat could have could have the meaningthe meaning

TOUCH($AGENT:MAN,$THEME:ANIMAL)TOUCH($AGENT:MAN,$THEME:ANIMAL)

at an early stage.at an early stage.


Meanings become more sophisticated as a consequence of development of Meanings become more sophisticated as a consequence of development of cognition and perception.cognition and perception.

E.g., later, the learner may perceive that there is John, a distinct male E.g., later, the learner may perceive that there is John, a distinct male person, that there is a particular type of animal (a cat), that both are person, that there is a particular type of animal (a cat), that both are singular in this context, and that they participate in this relation.singular in this context, and that they participate in this relation.

TOUCH($AGENT:JOHN($TYPE:PERSON,TOUCH($AGENT:JOHN($TYPE:PERSON, $GENDER:MALE,$NUM:SG),$GENDER:MALE,$NUM:SG), $THEME:CAT($TYPE:ANIMAL,$NUM:SG))$THEME:CAT($TYPE:ANIMAL,$NUM:SG))


Capacities of Capacities of CAMiLLeCAMiLLe

CAMiLLeCAMiLLe is conservative, in that it does not is conservative, in that it does not form hypotheses for which it does not have form hypotheses for which it does not have some evidence.some evidence.

In this implementation, In this implementation, CAMiLLeCAMiLLe is assumed is assumed to have prior knowledge of what the words are to have prior knowledge of what the words are in a sentence. It does not perform word in a sentence. It does not perform word segmentation (although in principle it could).segmentation (although in principle it could).


Knowledge about categoriesKnowledge about categories CAMiLLeCAMiLLe must know that lexical categories exist (but not must know that lexical categories exist (but not

which ones) and tries to determine what categories there are which ones) and tries to determine what categories there are (that is, which words are similar in syntactic, semantic or (that is, which words are similar in syntactic, semantic or morphological characteristics).morphological characteristics).

CAMiLLeCAMiLLe will generalize elements into a category when it will generalize elements into a category when it appears that they share sufficiently many characteristics.appears that they share sufficiently many characteristics.

– Distributional characteristicsDistributional characteristics– Similar meaning (e.g. Similar meaning (e.g. catcat & & dog dog both refer to both refer to

similar animals)similar animals) CAMiLLeCAMiLLe does not know about the specific syntactic does not know about the specific syntactic

categories such as Noun, Verb, Adjective, etc.categories such as Noun, Verb, Adjective, etc.


Knowledge about structureKnowledge about structure

CAMiLLeCAMiLLe must know that there are heads and phrases and must know that there are heads and phrases and knows that a phrase consists of at least a head and possibly knows that a phrase consists of at least a head and possibly other material that bears some relationship to it.other material that bears some relationship to it.– there does not appear to be evidence in the raw data that would tell you there does not appear to be evidence in the raw data that would tell you

that there are such things if you weren’t looking for them.that there are such things if you weren’t looking for them.

Syntactic Nuts Syntactic Nuts (Culicover 1999): A lot of ‘core’ syntactic (Culicover 1999): A lot of ‘core’ syntactic structure is in CS. Unpredictable structure and particular linear structure is in CS. Unpredictable structure and particular linear relationships are in the correspondence rules specific to the relationships are in the correspondence rules specific to the language.language.


Ignorance is blissIgnorance is bliss CAMiLLeCAMiLLe is a concrete minimalist, in that it only is a concrete minimalist, in that it only

makes use of information about the linear order of makes use of information about the linear order of formatives and corresponding meanings presented to it formatives and corresponding meanings presented to it in the course of learning.in the course of learning.

CAMiLLeCAMiLLe does not know about functional heads. does not know about functional heads. There are no purely grammatical formatives in There are no purely grammatical formatives in CAMiLLe’sCAMiLLe’s implementation that implementation that CAMiLLeCAMiLLe tries to tries to match against linguistic input.match against linguistic input.

CAMiLLeCAMiLLe does not know about transformations does not know about transformations per seper se.. And lots more...And lots more...


More blissMore bliss

CAMiLLe CAMiLLe does not know about traces or other does not know about traces or other empty categories (but should, and will, in the empty categories (but should, and will, in the next implementation).next implementation).

CAMiLLeCAMiLLe does not have grammatical indices. does not have grammatical indices. CAMiLLeCAMiLLe does not know about constraints, does not know about constraints,

such as Subjacency and the ECP, and in fact such as Subjacency and the ECP, and in fact lacks all comparable prior knowledge of lacks all comparable prior knowledge of syntactic theory. syntactic theory. – Where do these come from?Where do these come from?


Even more blissEven more bliss

CAMiLLeCAMiLLe cannot compute morphological cannot compute morphological structure, and must have morphological structure, and must have morphological structure presented to it explicitly in order to structure presented to it explicitly in order to make use of it in forming grammatical make use of it in forming grammatical hypotheses.hypotheses.

CAMiLLeCAMiLLe does not know about classical X-bar does not know about classical X-bar theory. That is, theory. That is, CAMiLLeCAMiLLe does not know does not know about specifiers and complements per se, zero-about specifiers and complements per se, zero-level and maximal projections, and so on.level and maximal projections, and so on.

CAMiLLeCAMiLLe does not know about government. does not know about government.


CorrespondencesCorrespondences The primary task of The primary task of CAMiLLeCAMiLLe is to construct is to construct

correspondences, that is, mappings of strings of correspondences, that is, mappings of strings of words into meanings (... based purely on the pairing words into meanings (... based purely on the pairing of strings and meanings).of strings and meanings).dog dog DOG($TYPE:ANIMAL) DOG($TYPE:ANIMAL)

see Robin see Robin SEE($THEME:ROBIN) SEE($THEME:ROBIN)

see the big bird see the big bird SEE($THEME:BIRD($ATTRIB:BIG,$REF:$DEF))SEE($THEME:BIRD($ATTRIB:BIG,$REF:$DEF))

see [category] see [category] SEE($THEME:[CATEGORY]) SEE($THEME:[CATEGORY])


Rule formationRule formation

CAMiLLeCAMiLLe incorporates the information provided by incorporates the information provided by everyevery sentence into a new rule or into an update of existing rules. sentence into a new rule or into an update of existing rules.

The relevant information consists of the linear ordering of The relevant information consists of the linear ordering of words and other formatives in the string .words and other formatives in the string .

The meaning of each sentence is compared with all existing The meaning of each sentence is compared with all existing rules in terms of the meaning features mentioned in the rule. rules in terms of the meaning features mentioned in the rule.

CAMiLLeCAMiLLe extracts those features of meaning that are possibly extracts those features of meaning that are possibly

relevant to the correspondence.relevant to the correspondence.


CAMiLLeCAMiLLe constructs... constructs...

word/meaning correspondencesword/meaning correspondences string/meaning correspondences (idioms)string/meaning correspondences (idioms) category/category correspondences (based on category/category correspondences (based on

identical or very similar distribution)identical or very similar distribution) category/category correspondences in strings category/category correspondences in strings

(limited constructions or ‘templates’)(limited constructions or ‘templates’)


But to go beyond templates, But to go beyond templates, CAMiLLe CAMiLLe needs needs to generalize.to generalize.

And to learn the grammar of English, And to learn the grammar of English, CAMiLLe CAMiLLe needs a more sophisticated needs a more sophisticated understanding of what kinds of relations may understanding of what kinds of relations may hold across a string of words than it currently hold across a string of words than it currently has.has.

We believe that this is an achievable goal.We believe that this is an achievable goal.


So where do constraints/universals So where do constraints/universals come from?come from?

Competing formulations of the Competing formulations of the sound/meaning correspondences in a sound/meaning correspondences in a social networksocial network


The social network simulationThe social network simulation AgentsAgents Social impact functionSocial impact function Parameter: Interaction partners Parameter: Interaction partners Parameter: Interaction distanceParameter: Interaction distance Knowledge of languageKnowledge of language

– Features (in this case, 3)Features (in this case, 3)– and feature values (2, 4, 8, etc)and feature values (2, 4, 8, etc)

Noise (all other factors lumped Noise (all other factors lumped together)together)


Assumptions about learningAssumptions about learning

Each learner interacts with a number Each learner interacts with a number of individuals at each time of individuals at each time tt

Each learner is influenced by the Each learner is influenced by the individuals that it interacts with individuals that it interacts with dependent on their relative strength.dependent on their relative strength.

Majority ‘wins’Majority ‘wins’ There is no review or evaluation by There is no review or evaluation by

the learner of its own internal statethe learner of its own internal state


Display (start)Display (start)


Language distribution (start)Language distribution (start)

The Tower of Babel


Strengths of individualsStrengths of individuals


GapsGaps It is well-known that there are It is well-known that there are gapsgaps

in the possible languages, that is, in the possible languages, that is, languages that are logically possible languages that are logically possible but do not exist. Is this a deep fact but do not exist. Is this a deep fact about cognition (explained by UG), or about cognition (explained by UG), or not?not?

[maybe, maybe not.][maybe, maybe not.]


HypothesisHypothesis

Differential complexity of different Differential complexity of different ways of expressing a CS introduces a ways of expressing a CS introduces a bias into the network against some bias into the network against some alternativesalternatives

These will be disfavored in the These will be disfavored in the network, perhaps even disappear.network, perhaps even disappear.


Initial random distribution of feature valuesInitial random distribution of feature values


Initial population of the eight languagesInitial population of the eight languages

Chicago, May 25, 2005Chicago, May 25, 2005 5757Distribution of languages and features after 150 stepsDistribution of languages and features after 150 steps


Language distribution after 150 stepsLanguage distribution after 150 steps


Some examples of what Some examples of what CAMiLLe CAMiLLe doesdoes


Identifying and categorizing nounsIdentifying and categorizing nouns(Input: Nouns #1, Eve)(Input: Nouns #1, Eve)

you xxx more cookies ? = COOKIE($TYPE:FOOD)you xxx more cookies ? = COOKIE($TYPE:FOOD)how about another graham cracker ? = COOKIE($TYPE:FOOD)how about another graham cracker ? = COOKIE($TYPE:FOOD)would that do just as well ?would that do just as well ?here .here .here you go .here you go .you have another cookie right on the table .= COOKIE($TYPE:FOOD)you have another cookie right on the table .= COOKIE($TYPE:FOOD)more juice ? = JUICE($TYPE:DRINK)more juice ? = JUICE($TYPE:DRINK)would you like more grape juice ? = JUICE($TYPE:DRINK)would you like more grape juice ? = JUICE($TYPE:DRINK)where's your cup ? = CUP($TYPE:UTENSIL)where's your cup ? = CUP($TYPE:UTENSIL)oh I took it .oh I took it .I think that was Fraser . = FRASER($TYPE:PERSON)I think that was Fraser . = FRASER($TYPE:PERSON)I'm not sure .I'm not sure .what ?what ?are you saying Fraser ?= FRASER($TYPE:PERSON)are you saying Fraser ?= FRASER($TYPE:PERSON)Mr Fraser ?= FRASER($TYPE:PERSON)Mr Fraser ?= FRASER($TYPE:PERSON)yes that's much better .yes that's much better .Mr Fraser ?= FRASER($TYPE:PERSON)Mr Fraser ?= FRASER($TYPE:PERSON)what is that ?what is that ?huh ?huh ?what is that ?what is that ?yes a fly . = FLY($TYPE:INSECT)yes a fly . = FLY($TYPE:INSECT)why don't you go in the room and kill a fly ? = FLY($TYPE:INSECT)why don't you go in the room and kill a fly ? = FLY($TYPE:INSECT)you go in the room and kill a fly . = FLY($TYPE:INSECT)you go in the room and kill a fly . = FLY($TYPE:INSECT)yes you get a fly . = FLY($TYPE:INSECT)yes you get a fly . = FLY($TYPE:INSECT)you get one . = FLY($TYPE:INSECT)you get one . = FLY($TYPE:INSECT)


Sample of noun correspondences identified by Sample of noun correspondences identified by CAMilLLeCAMilLLe in Nouns # 1in Nouns # 1

CHAIR CHAIR chair chair CHAIR($TYPE:FURNITURE) CHAIR($TYPE:FURNITURE) chair chair CLOTHES CLOTHES hat hat DIAPER DIAPER diaper diaper DIAPER($TYPE:CLOTHES) DIAPER($TYPE:CLOTHES) diaper diaper FRASER FRASER Fraser Fraser FRASER($TYPE:PERSON) FRASER($TYPE:PERSON) Fraser Fraser FURNITURE FURNITURE chair chair HAT HAT hat hat HAT($TYPE:CLOTHES) HAT($TYPE:CLOTHES) hat hat HEAD HEAD head head HEAD($TYPE:BODYPART) HEAD($TYPE:BODYPART) head head JUICE JUICE juice juice JUICE($TYPE:DRINK) JUICE($TYPE:DRINK) juice juice PENCIL PENCIL pencil pencil PENCIL($TYPE:UTENSIL) PENCIL($TYPE:UTENSIL) pencil pencil PUDDING PUDDING pudding pudding PUDDING($TYPE:FOOD) PUDDING($TYPE:FOOD) pudding pudding STOOL STOOL stool stool STOOL($TYPE:HOUSEHOLD) STOOL($TYPE:HOUSEHOLD) stool stool TELEPHONE($TYPE:TOY)$= TELEPHONE($TYPE:TOY)$= telephone . telephone+1->. .<-1->telephone telephone . telephone+1->. .<-1->telephoneTELEPHONE TELEPHONE telephone . telephone+1->. .<-1->telephone telephone . telephone+1->. .<-1->telephoneWATER WATER water water WATER($TYPE:DRINK) WATER($TYPE:DRINK) water water


Nouns exemplified in Nouns #1Nouns exemplified in Nouns #1

babybabybookbookbottlebottleboxboxchairchaircheesecheesecoffeecoffeecookiecookiecupcupdiaperdiaperduck duck EveEveeyeeyeFraserFraserhathat

headhead juicejuice manman milkmilk mommymommy paperpaper pencilpencil puddingpudding radioradio shoeshoe flyfly soldierssoldiers stoolstool telephonetelephone traintrain waterwater


Nouns learned in Nouns #1Nouns learned in Nouns #1

babybabybookbookbottlebottleboxboxchairchaircheesecheesecoffeecoffeecookiecookiecupcupdiaperdiaperduck duck EveEveeyeeyeFraserFraserhathat

headhead juicejuice manman milkmilk mommymommy paperpaper pencilpencil puddingpudding radioradio shoeshoe flyfly soldierssoldiers stoolstool telephonetelephone traintrain waterwater


Why Why CAMiLLeCAMiLLe thinks that thinks that telephonetelephone is a sentence final is a sentence final

elementelement

well go and get your well go and get your telephone .telephone . = = TELEPHONE($TYPE:TOY)TELEPHONE($TYPE:TOY)

yes he gave you your yes he gave you your telephone .telephone . = = TELEPHONE($TYPE:TOY)TELEPHONE($TYPE:TOY)

yes that's the yes that's the telephone .telephone . = TELEPHONE($TYPE:TOY) = TELEPHONE($TYPE:TOY)that was the that was the telephone .telephone . = TELEPHONE($TYPE:TOY) = TELEPHONE($TYPE:TOY)it was Papa on the it was Papa on the telephone .telephone . = =

TELEPHONE($TYPE:TOY)TELEPHONE($TYPE:TOY)yes the yes the telephone .telephone . = TELEPHONE($TYPE:TOY) = TELEPHONE($TYPE:TOY)


Verb #2Verb #21. a give b to c = GIVE($AGENT:A,$THEME:B,$RECIP:C)2. does y give x to z = GIVE($AGENT:Y,$THEME:X,$RECIP:Z)

GIVE give to give+2->to give<-2->to $P=1GIVE 1.a 2.give 3.b 4.to 5.c 1.does 2.y 3.give 4.x 5.to 6.z $P=0.5…3. tell s to give r to q please = GIVE($AGENT:S,$THEME:R,$RECIP:Q)4. give j to e now = GIVE($THEME:J,$RECIP:E)[ Same results, stronger rules]

5. c doesn't give k to a = $NEG(GIVE($AGENT:C,$THEME:K,$RECIP:A)) 6. can q give the w to s ? = $QU(GIVE($AGENT:Q,$THEME:W,$RECIP:S))

GIVE give to $P=1GIVE give to give+2->to give<-2->to $P=0.833333

(And with richer input we get:GIVE($RECIP:E) give to give+2->to give<-2->to e to+1->e e<-1->to)


Verbal inflection Verbal inflection

InputInputMary have ~s sleep ~en soundly = Mary have ~s sleep ~en soundly = SLEEP($AGENT:MARY,$ASPECT:SLEEP($AGENT:MARY,$ASPECT:

$COMPLETE,$TIME:$NOW)$COMPLETE,$TIME:$NOW)

John be ~s really snore ~ing = John be ~s really snore ~ing = SNORE($AGENT:JOHN,$ASPECT:SNORE($AGENT:JOHN,$ASPECT:$PROGRESSIVE,$TIME:$NOW)$PROGRESSIVE,$TIME:$NOW)

John probably have ~s fall ~en = John probably have ~s fall ~en = FALL($EXP:JOHN,$ASPECT:FALL($EXP:JOHN,$ASPECT:$COMPLETE,$TIME:$NOW)$COMPLETE,$TIME:$NOW)

John might have ~s see ~en = John might have ~s see ~en = SEE($EXP:JOHN,$ASPECT:SEE($EXP:JOHN,$ASPECT:$COMPLETE,$TIME:$NOW)$COMPLETE,$TIME:$NOW)

John suddenly see ~ed it = John suddenly see ~ed it = SEE($EXP:JOHN,$PAST)SEE($EXP:JOHN,$PAST)

and then Mary fall ~ed = and then Mary fall ~ed = FALL($THEME:MARY,$PAST)FALL($THEME:MARY,$PAST)


Output (ranked in order of strength)Output (ranked in order of strength)

[JOHN; MARY;] [JOHN; MARY;] [john; mary;][john; mary;]$COMPLETE $COMPLETE have ~en have ~en$NOW $NOW ~s ~s$PAST $PAST ~ed ~ed$PROG $PROG be ~ing be ~ingDIE($ASPECT:$COMPLETE) DIE($ASPECT:$COMPLETE) die ~en die+1->~en die<-1->~en die ~en die+1->~en die<-1->~enEAT($ASPECT:$COMPLETE) EAT($ASPECT:$COMPLETE) 5.eat ~en eat+1->~en eat<-1->~en 5.eat ~en eat+1->~en eat<-1->~enFALL($ASPECT:$PROG) FALL($ASPECT:$PROG) fall ~ing fall+1->~ing fall<-1->~ing fall ~ing fall+1->~ing fall<-1->~ingFALL($ASPECT:$COMPLETE) FALL($ASPECT:$COMPLETE) fall ~en fall+1->~en fall<-1->~en fall ~en fall+1->~en fall<-1->~enFIND FIND find ~s find<-1->~s find ~s find<-1->~sFIND($ASPECT:$COMPLETE) FIND($ASPECT:$COMPLETE) have find have+2->find find<-2->have ~s ~s+1->find find<-1->~s have find have+2->find find<-2->have ~s ~s+1->find find<-1->~s

~en find+1->~en find<-1->~en~en find+1->~en find<-1->~enFIND($TIME:$NOW) FIND($TIME:$NOW) find ~s find<-1->~s find ~s find<-1->~sJUMP($ASPECT:$PROG) JUMP($ASPECT:$PROG) jump ~ing jump+1->~ing jump<-1->~ing jump ~ing jump+1->~ing jump<-1->~ingJUMP($ASPECT:$COMPLETE) JUMP($ASPECT:$COMPLETE) have jump have+2->jump have<-2->jump ~s ~s+1->jump jump<- have jump have+2->jump have<-2->jump ~s ~s+1->jump jump<-

1->~s ~en jump+1->~en jump<-1->~en1->~s ~en jump+1->~en jump<-1->~enJUMP($TIME:$NOW) JUMP($TIME:$NOW) jump jumpJUMP($TIME:$PAST) JUMP($TIME:$PAST) jump ~ed jump<-1->~ed jump ~ed jump<-1->~ed LOOK($ASPECT:$COMPLETE) LOOK($ASPECT:$COMPLETE) look ~en look+1->~en look<-1->~en look ~en look+1->~en look<-1->~enLOOK($TIME:$NOW) LOOK($TIME:$NOW) look lookRUN($ASPECT:$COMPLETE) RUN($ASPECT:$COMPLETE) run ~en run+1->~en run<-1->~en run ~en run+1->~en run<-1->~en


SEE($TIME:$NOW) SEE($TIME:$NOW) see ~s see<-1->~s see ~s see<-1->~sSLEEP($ASPECT:$COMPLETE) SLEEP($ASPECT:$COMPLETE) sleep ~en sleep+1->~en sleep<-1- sleep ~en sleep+1->~en sleep<-1-

>~en>~enSLEEP($ASPECT:[$COMPLETE; $PROG;] SLEEP($ASPECT:[$COMPLETE; $PROG;] sleep [~en; ~ing;] sleep [~en; ~ing;]

sleep+1->[~en; ~ing;] sleep<-1->[~en; ~ing;]sleep+1->[~en; ~ing;] sleep<-1->[~en; ~ing;]SLEEP($ASPECT:$PROG) SLEEP($ASPECT:$PROG) sleep ~ing sleep+1->~ing sleep<-1- sleep ~ing sleep+1->~ing sleep<-1-

>~ing>~ingSNORE($ASPECT:$PROG) SNORE($ASPECT:$PROG) snore ~ing snore+1->~ing snore<-1- snore ~ing snore+1->~ing snore<-1-

>~ing>~ingSNORE($ASPECT:$COMPLETE) SNORE($ASPECT:$COMPLETE) snore ~en snore+1->~en snore<-1- snore ~en snore+1->~en snore<-1-

>~en>~en[DIE; LOOK;]($TIME:$NOW) [DIE; LOOK;]($TIME:$NOW) [die; look;] ~en [die; look;]+1->~en [die; look;] ~en [die; look;]+1->~en

[die; look;]<-1->~en[die; look;]<-1->~en[DIE; EAT; LOOK;] [DIE; EAT; LOOK;] [die; eat; look;] ~en [die; eat; look;] +1->~en [die; eat; look;] ~en [die; eat; look;] +1->~en

[die; eat; look;] <-1->~en[die; eat; look;] <-1->~en[DIE; FALL; LOOK; RUN; SLEEP; SNORE;] ($ASPECT:$COMPLETE) [DIE; FALL; LOOK; RUN; SLEEP; SNORE;] ($ASPECT:$COMPLETE)

[die; fall; look; run; sleep; snore;] ~en [die; fall; look; run; sleep; [die; fall; look; run; sleep; snore;] ~en [die; fall; look; run; sleep; snore;] +1->~en [die; fall; look; run; sleep; snore;] <-1->~ensnore;] +1->~en [die; fall; look; run; sleep; snore;] <-1->~en


[FALL; SLEEP; SNORE;]($ASPECT:[$COMPLETE; $PROG;][FALL; SLEEP; SNORE;]($ASPECT:[$COMPLETE; $PROG;] [fall; [fall; sleep; snore;] [~en; ~ing;] [fall; sleep; snore;]+1->[~en; ~ing;] sleep; snore;] [~en; ~ing;] [fall; sleep; snore;]+1->[~en; ~ing;]

[FALL; JUMP; SLEEP; SNORE;] ($ASPECT:$PROG)[FALL; JUMP; SLEEP; SNORE;] ($ASPECT:$PROG) [fall; jump; [fall; jump; sleep; snore;] ~ing [fall; jump; sleep; snore;] +1->~ingsleep; snore;] ~ing [fall; jump; sleep; snore;] +1->~ing

[FALL; RUN;] ($TIME:$NOW)[FALL; RUN;] ($TIME:$NOW) 5.[fall; run;] 5.[fall; run;][FIND; JUMP;][FIND; JUMP;] have [find; jump;] have+2->[find; jump;] ~s have [find; jump;] have+2->[find; jump;] ~s

~s+1->[find; jump;] ~en [find; jump;] +1->~en~s+1->[find; jump;] ~en [find; jump;] +1->~en[FIND; JUMP;]($TIME:$NOW)[FIND; JUMP;]($TIME:$NOW) have [find; jump;] have+2->[find; have [find; jump;] have+2->[find;

jump;] ~s+1->[find; jump;] [find; jump;]+1->~en jump;] ~s+1->[find; jump;] [find; jump;]+1->~en [FIND; JUMP; SLEEP;] ($ASPECT:$COMPLETE)[FIND; JUMP; SLEEP;] ($ASPECT:$COMPLETE) have [find; jump; have [find; jump;

sleep;] have+2->[find; jump; sleep;] ~s+1->[find; jump; sleep;] sleep;] have+2->[find; jump; sleep;] ~s+1->[find; jump; sleep;] [find; jump; sleep;] +1->~en[find; jump; sleep;] +1->~en

[FIND; SEE;][FIND; SEE;] ($TIME:$NOW)($TIME:$NOW) [find; see;] ~s [find; see;] ~s [JUMP; RUN;] ($TIME:$PAST)[JUMP; RUN;] ($TIME:$PAST) [jump; run;] ~ed [jump; run;] <-1- [jump; run;] ~ed [jump; run;] <-1-

>~ed>~ed


DP StructureDP Structure Input, DP#3Input, DP#3

See the man = See the man = SEE($THEME:MAN($TYPE:PERSON))SEE($THEME:MAN($TYPE:PERSON))See the woman = See the woman = SEE($THEME:WOMAN($TYPE:PERSON))SEE($THEME:WOMAN($TYPE:PERSON))look at the baby = look at the baby = LOOK($THEME:BABY($TYPE:PERSON))LOOK($THEME:BABY($TYPE:PERSON))that 's a cat = that 's a cat = CAT($TYPE:ANIMAL)CAT($TYPE:ANIMAL)and this is a dog = and this is a dog = DOG($TYPE:ANIMAL)DOG($TYPE:ANIMAL)show me the horse = show me the horse = SHOW($THEME:HORSE($TYPE:ANIMAL))SHOW($THEME:HORSE($TYPE:ANIMAL))what a tall man = what a tall man = MAN($TYPE:PERSON,$ATTR:TALL)MAN($TYPE:PERSON,$ATTR:TALL)he is a nice man = he is a nice man = MAN($TYPE:PERSON,$ATTR:NICE)MAN($TYPE:PERSON,$ATTR:NICE)and this is a tall woman = and this is a tall woman = WOMAN($TYPE:PERSON,$ATTR:TALL)WOMAN($TYPE:PERSON,$ATTR:TALL)look at the little baby = look at the little baby = LOOK($THEME:BABY($TYPE:PERSON,LOOK($THEME:BABY($TYPE:PERSON,

$ATTR:LITTLE))$ATTR:LITTLE))see the nice cat = see the nice cat = SEE($THEME:CAT($TYPE:ANIMAL,$ATTR:NICE))SEE($THEME:CAT($TYPE:ANIMAL,$ATTR:NICE))see the big dog = see the big dog = SEE($THEME:DOG($TYPE:ANIMAL,$ATTR:BIG))SEE($THEME:DOG($TYPE:ANIMAL,$ATTR:BIG))and the big horse = and the big horse = HORSE($TYPE:ANIMAL,$ATTR:BIG)HORSE($TYPE:ANIMAL,$ATTR:BIG)that 's a little dog = that 's a little dog = DOG($TYPE:ANIMAL,$ATTR:LITTLE)DOG($TYPE:ANIMAL,$ATTR:LITTLE)


Rule formed in DP#3Rule formed in DP#3

[BABY; DOG; HORSE;] [BABY; DOG; HORSE;] [big; little;] [baby; dog; horse;][big;little;]+1->[baby; dog; horse;][big; little;] [baby; dog; horse;][big;little;]+1->[baby; dog; horse;]


ArgumentArgument structurestructure

1. [7704] SEE see7. [392] SEE($EXP:$ME) see . see[+1->; +2->;]. i[+1->;

+2->;] see 9. [250] SEE($THEME:[$IT; $YOU;] see [it; you;] see+1-

>[it; you;]10. [216] SEE($THEME:[BECKY; EVE;] see see+1->[becky;

eve;]

Rules for see based on 115 sentences with see spoken to one actual child


Word order & argument Word order & argument structurestructure

InputInput18. I see mary = SEE($EXP:$ME,18. I see mary = SEE($EXP:$ME,

$THEME:MARY($TYPE:PERSON))$THEME:MARY($TYPE:PERSON))19. do you see john ? = SEE($EXP:$YOU,19. do you see john ? = SEE($EXP:$YOU,

$THEME:JOHN($TYPE:PERSON))$THEME:JOHN($TYPE:PERSON))20. I see john = SEE($EXP:$ME,$THEME:JOHN($TYPE:PERSON))20. I see john = SEE($EXP:$ME,$THEME:JOHN($TYPE:PERSON))21. and I see a boy too = SEE($EXP:$ME,21. and I see a boy too = SEE($EXP:$ME,

$THEME:BOY($TYPE:PERSON))$THEME:BOY($TYPE:PERSON))

RulesRulesSEE($EXP:$ME) SEE($EXP:$ME) i see i+1->see i see i+1->seeSEE($THEME:[JOHN; MARY;] SEE($THEME:[JOHN; MARY;] see [john; mary;] see+1- see [john; mary;] see+1-

>[john; mary;]>[john; mary;]SEE($THEME:BOY) SEE($THEME:BOY) see boy see+2->boy see boy see+2->boy


Discontinuous dependenciesDiscontinuous dependenciesInputInputyou are say-ing something = SAY($AGENT:$YOU,$THEME:THING($REF:$INDEF))you are say-ing something = SAY($AGENT:$YOU,$THEME:THING($REF:$INDEF))you will say something = SAY($AGENT:$YOU,$THEME:THING($REF:$INDEF))you will say something = SAY($AGENT:$YOU,$THEME:THING($REF:$INDEF))you did say something = SAY($AGENT:$YOU,$THEME:THING($REF:$INDEF))you did say something = SAY($AGENT:$YOU,$THEME:THING($REF:$INDEF))you are say-ing nonsense = SAY($AGENT:$YOU,$THEME:NONSENSE)you are say-ing nonsense = SAY($AGENT:$YOU,$THEME:NONSENSE)you will say nonsense= SAY($AGENT:$YOU,$THEME:NONSENSE)you will say nonsense= SAY($AGENT:$YOU,$THEME:NONSENSE)you did say nonsense= SAY($AGENT:$YOU,$THEME:NONSENSE)you did say nonsense= SAY($AGENT:$YOU,$THEME:NONSENSE)what are you say-ing = $QU(SAY($AGENT:$YOU,$THEME:$WHTHING))what are you say-ing = $QU(SAY($AGENT:$YOU,$THEME:$WHTHING))what will you say = $QU(SAY($AGENT:$YOU,$THEME:$WHTHING))what will you say = $QU(SAY($AGENT:$YOU,$THEME:$WHTHING))what did you say = $QU(SAY($AGENT:$YOU,$THEME:$WHTHING))what did you say = $QU(SAY($AGENT:$YOU,$THEME:$WHTHING))[ and similarly with [ and similarly with do,do, eat eat ]]

RulesRules[DO; EAT; SAY;]($THEME:$WHTHING) [DO; EAT; SAY;]($THEME:$WHTHING) 1.what 4.[do; eat; say;] 1.what 4.[do; eat; say;]


Mock-Japanese wh-questionsMock-Japanese wh-questions

InputInput– 400 sentences of the general form400 sentences of the general form

this ACC do IMP . = $IMP(DO($THEME:THING($REF:$DEF)))this ACC do IMP . = $IMP(DO($THEME:THING($REF:$DEF)))sing IMP . = $IMP(SING($AGENT:$PERSON))sing IMP . = $IMP(SING($AGENT:$PERSON))sleep PRES Q . = $QU(SLEEP($AGENT:$PERSON)) sleep PRES Q . = $QU(SLEEP($AGENT:$PERSON))

what ACC do PAST Q . = $QU(DO($AGENT:$PERSON,what ACC do PAST Q . = $QU(DO($AGENT:$PERSON,$THEME:WHTHING))$THEME:WHTHING))

where TOP men NOM carrots ACC eat FUT Q . = where TOP men NOM carrots ACC eat FUT Q . = $QU(EAT($AGENT:MAN,$THEME:CARROT,$QU(EAT($AGENT:MAN,$THEME:CARROT,$PLACE:WHPLACE))$PLACE:WHPLACE))


Sample of rules formedSample of rules formed$QU$QU q . q .SAYSAY say X . say X .DODO do X . do X .IMPIMP imp . imp . WHTHING WHTHING what acc X X q . what acc X X q . WOMANWOMAN woman+1->nom woman+1->nomSAYSAY say say EATEAT eat eat [$IMP; $QU;] [$IMP; $QU;] [imp; q;] . [imp; q;] .DODO do do $QU([DO; EAT; SLEEP;] $QU([DO; EAT; SLEEP;] [do; eat; sleep] X q . [do; eat; sleep] X q .DO($THEME:WHTHING) DO($THEME:WHTHING) what+1->acc do+2->q . what+1->acc do+2->q .CARROTCARROT carrots+1->acc carrots+1->accSAY($THEME:WHTHING) SAY($THEME:WHTHING) what+1->acc q say say+2->q what+1->acc q say say+2->q EAT EAT eat eat NONSENSE NONSENSE nonsense+1->acc nonsense+1->acc[DO; SAY;]($THEME:WHTHING) [DO; SAY;]($THEME:WHTHING) what+1->acc q [do; say;]+2->q what+1->acc q [do; say;]+2->q[DO; EAT; SAY] [DO; EAT; SAY] .[do; eat; say] .[do; eat; say] EAT($AGENT:[CHILD; MAN; WOMAN]) EAT($AGENT:[CHILD; MAN; WOMAN]) [child; men; woman] nom [child; men; woman] nomEAT($AGENT:[CHILD; WOMAN] EAT($AGENT:[CHILD; WOMAN] [child; woman] nom [...] eat X . [child; woman] nom [...] eat X .

[NOTE: Rules have been highly edited and simplified for readability][NOTE: Rules have been highly edited and simplified for readability]


ScramblingScramblinghit bill-acc = HIT($AGENT:$HIM,$THEME:BILL)

‘(He) hits Bill.’

bob-nom see = SEE($EXP:BOB)

‘Bob sees.’

See jim-acc = SEES($EXP:$HIM,$THEME:JIM)

‘(He) sees Jim.’

...

CAMiLLe correctly determines that the $EXP role of SEE is assigned to the (interpretation of the) DP marked -nom, that the $THEME role of SEE is assigned to the (interpretation of the) DP marked -acc, and that the $AGENT role of HIT is assigned to the (interpretation of the) DP marked -nom.:

SEE($EXP:[ARTHUR; BOB]) 1. [arthur; bob] 2.~nom 3.seeSEE($THEME:[ARTHUR; BOB]) [arthur; bob] ~acc [arthur; bob]+1-

>~acc [arthur; bob]<-1->~acc

HIT($AGENT:[ARTHUR; BOB]) 1.hit 2. [arthur; bob] 3.~nom


Inversion (English type)Inversion (English type)

InputInputJohn = JOHN($TYPE:PERSON)John = JOHN($TYPE:PERSON)Bird = BIRD($TYPE:ANIMAL)Bird = BIRD($TYPE:ANIMAL)John see ~s the bird = SEE($EXP:JOHN,$THEME:BIRD)John see ~s the bird = SEE($EXP:JOHN,$THEME:BIRD)do ~s John see a bird = $QU(SEE($EXP:JOHN,do ~s John see a bird = $QU(SEE($EXP:JOHN,

$THEME:BIRD))$THEME:BIRD))the bird see ~s John = SEE($EXP:BIRD,$THEME:JOHN)the bird see ~s John = SEE($EXP:BIRD,$THEME:JOHN)do ~s the bird see John = $QU(SEE($EXP:BIRD,do ~s the bird see John = $QU(SEE($EXP:BIRD,

$THEME:JOHN))$THEME:JOHN))a bird see ~s John = SEE($EXP:BIRD,$THEME:JOHN)a bird see ~s John = SEE($EXP:BIRD,$THEME:JOHN)......

RulesRules$QU $QU 1.do 2.~s 1.do 2.~s


Inversion (non-English type)Inversion (non-English type)

InputInputSandy eat ~s the cake = EAT($AGENT:SANDY,Sandy eat ~s the cake = EAT($AGENT:SANDY,

$THEME:CAKE)$THEME:CAKE)eat ~s Sandy the cake = $QU(EAT($AGENT:SANDY,eat ~s Sandy the cake = $QU(EAT($AGENT:SANDY,

$THEME:CAKE))$THEME:CAKE))Sandy come ~s = COME($AGENT:SANDY)Sandy come ~s = COME($AGENT:SANDY)come ~s Sandy = $QU(COME($AGENT:SANDY))come ~s Sandy = $QU(COME($AGENT:SANDY))Sandy go ~s = GO($AGENT:SANDY)Sandy go ~s = GO($AGENT:SANDY)go ~s Sandy = $QU(GO($AGENT:SANDY))go ~s Sandy = $QU(GO($AGENT:SANDY))......

RulesRules$QU $QU 2.~s 2.~s


ImperativesImperatives Sample InputSample Inputeat = $IMP(EAT($AGENT:$YOU))eat = $IMP(EAT($AGENT:$YOU))eat the cereal = $IMP(EAT($AGENT:$YOU,eat the cereal = $IMP(EAT($AGENT:$YOU,

$THEME:CEREAL))$THEME:CEREAL))drink = $IMP(DRINK($AGENT:$YOU))drink = $IMP(DRINK($AGENT:$YOU))drink the milk = $IMP(DRINK($AGENT:$YOU,drink the milk = $IMP(DRINK($AGENT:$YOU,

$THEME:MILK))$THEME:MILK))

RulesRules[DRINK; EAT;]($AGENT:$YOU) [DRINK; EAT;]($AGENT:$YOU) 1.[drink; eat;] 1.[drink; eat;][HEAR; SEE;]($EXP:$YOU) [HEAR; SEE;]($EXP:$YOU) 1.[hear; see;] 1.[hear; see;]$IMP([DRINK; EAT; HEAR; LOOK; SEE;]) $IMP([DRINK; EAT; HEAR; LOOK; SEE;]) 1.[drink; eat; 1.[drink; eat;

hear; look; see;]hear; look; see;]


TemplatesTemplates

These are all templates – they are not These are all templates – they are not correspondence rules that take into account correspondence rules that take into account grammatical structure, and hence cannot be grammatical structure, and hence cannot be fully general.fully general.

They are good for local constructions (like They are good for local constructions (like inversion), but not for those that hold over inversion), but not for those that hold over unbounded strings (like wh-questions and unbounded strings (like wh-questions and many others).many others).


CAMiLLeCAMiLLe needs to know about grammatical needs to know about grammatical structure in order to find grammatical structure structure in order to find grammatical structure in the input.in the input.

CAMiLLeCAMiLLe needs to know about relations that needs to know about relations that hold across syntactic structures in order to find hold across syntactic structures in order to find such relations in the input.such relations in the input.

Chicago, May 25, 2005 1 Adventures with Camille, a Computational Simulation of a Minimalist Language Learner Peter W. Culicover (The Ohio State University)

Documents

theory of grammar

theory of generative

linguistic theory

knowledge of language

human language

linguistic competence

language facultylearner

linguistic knowledge