A Best-Fit Approach for Productive Analysis of Omitted Arguments Eva Mok & John Bryant University of...

A Best-Fit Approach for Productive Analysis of Omitted Arguments

Eva Mok & John BryantUniversity of California, Berkeley

International Computer Science Institute

Simplify grammar by exploiting the language understanding process

Omission of arguments in Mandarin Chinese

Construction grammar framework

Model of language understanding

Our best-fit approach

Mother (I) give you this (a toy).

CHILDES Beijing Corpus (Tardiff, 1993; Tardiff, 1996)

ma1+ma

gei3 ni3 zhei4+ge

mother give 2PS this+CLS You give auntie [the peach].

Oh (go on)! You give [auntie] [that].

Productive Argument Omission (in Mandarin)

1

2

3

ni3 gei3 yi2

2PS giveaunti

e

ao ni3 gei3 ya

EMP 2PS give EMP

4 gei3

give

[I] give [you] [some peach].

Arguments are omitted with different probabilities

All arguments omitted: 30.6% No arguments omitted: 6.1%

% elided (98 total utterances)

Giver

Recipient

Theme

0.00%

10.00%

20.00%

30.00%

40.00%

50.00%

60.00%

70.00%

80.00%

90.00%

100.00%

Construction grammar approach

Kay & Fillmore 1999; Goldberg 1995

Grammaticality: form and function

Basic unit of analysis: construction, i.e. a pairing of form and meaning constraints

Not purely lexically compositional

Implies early use of semantics in processing Embodied Construction Grammar (ECG) (Bergen & Chang,

2005)

Problem: Proliferation of constructions

Subj Verb Obj1 Obj2↓ ↓ ↓ ↓

Giver Transfer Recipient

Theme

Verb Obj1 Obj2↓ ↓ ↓

Transfer Recipient

Theme

…

Subj Verb Obj2↓ ↓ ↓

Giver Transfer Theme

Subj Verb Obj1↓ ↓ ↓


If the analysis process is smart, then...

The grammar needs only state one construction

Omission of constituents is flexibly allowed

The analysis process figures out what was omitted

Subj Verb Obj1 Obj2↓ ↓ ↓ ↓


Theme

Best-fit analysis process takes burden off the grammar representation

Constructions

Simulation

Utterance Discourse & Situational Context

Semantic Specification:

image schemas, frames, action schemas

Analyzer:

incremental,competition-based, psycholinguistically

plausible

Competition-based analyzer finds the best analysis

An analysis is made up of: A constructional tree A set of resolutions A semantic specification

The best fit has the highest combined score

Combined score that determines best-fit

Syntactic Fit: Constituency relations Combine with preferences on non-local

elements Conditioned on syntactic context

Antecedent Fit: Ability to find referents in the context Conditioned on syntactic information, feature

agreement

Semantic Fit: Semantic bindings for frame roles Frame roles’ fillers are scored

Analyzing ni3 gei3 yi2 (You give auntie)

Syntactic Fit: P(Theme omitted | ditransitive cxn) = 0.65

P(Recipient omitted | ditransitive cxn) = 0.42

Two of the competing analyses:

ni3 gei3 yi2 omitted↓ ↓ ↓ ↓

Giver Transfer Recipient Theme

ni3 gei3 omitted yi2↓ ↓ ↓ ↓


(1-0.78)*(1-0.42)*0.65 = 0.08 (1-0.78)*(1-0.65)*0.42 = 0.03

Using frame and lexical information to restrict type of reference

Lexical Unit gei3

Giver (DNI)

Recipient (DNI)

Theme (DNI)

The Transfer Frame

Giver

Recipient

Theme

Manner

Means

Place

Purpose

Reason

Time

Can the omitted argument be recovered from context?

Antecedent Fit:





Discourse & Situational Context

child motherpeach auntietable

?

How good of a theme is a peach? How about an aunt?

The Transfer Frame

Giver (usually animate)

Recipient (usually animate)

Theme (usually inanimate)





Semantic Fit:

The argument omission patterns shown earlier can be covered with just ONE construction

Each cxn is annotated with probabilities of omission

Language-specific default probability can be set

Subj Verb Obj1 Obj2

↓ ↓ ↓ ↓


0.78 0.42 0.65P(omitted|cxn):

% elided (98 total utterances)

Giver

Recipient

Theme

0.00%

10.00%

20.00%

30.00%

40.00%

50.00%

60.00%

70.00%

80.00%

90.00%

16

Research goal

A computationally-precise modeling framework for learning early constructions

Linguistic Knowledge

LearningLearning

17

Frequent argument omission in pro-drop languages

Mandarin example:

ni3 gei3 yi2 (“you give auntie”)

Even in English, there are often no spoken antecedents to pronouns in conversations

Learner must integrate cues from

intentions, gestures, prior discourse, etc

18

A short dialogue

bie2 mo3 wai4+tou2 a: #1_3 ! ( 別抹外頭啊 ) NEG-IMP apply forehead

Don’t apply [lotion to your] forehead

mo3 wai4+tou2 ke3 jiu4 bu4 hao3+kan4 le a: . ( 抹外頭可就不好看了啊 ) apply forehead LINKER LINKER NEG good looking CRS SFP

[If you] apply [lotion to your] forehead then [you will] not be pretty

…

ze ya a: # bie2 gei3 ma1+ma wang3 lian3 shang4 mo:3 e: ! (嘖呀啊 # 別給媽媽往臉上抹呃 )

INTERJ # NEG-IMP BEN mother CV-DIR face on apply

INTERJ # Don’t apply [the lotion] on [your mom’s] face (for mom)

[- low pitch motherese] ma1+ma bu4 mo:3 you:2 . ( 媽媽不抹油 )

mother NEG apply lotion

Mom doesn’t apply (use) lotion

Goals, refined

Demonstrate learning given embodied meaning representation

structured representation of context

Based on Usage-based learning

Domain-general statistical learning mechanism

Generalization / linguistic category formation

19

Towards a precise computational model

Modeling early grammar learning

Context model & Simulation

Data annotation

Finding the best analysis for learning

Hypothesizing and reorganizing constructions

Pilot results

20

Embodied Construction Grammar

21

construction yi2-Nsubcase of Morpheme form

constraintsself.f.orth <-- "yi2"

meaning : @Auntevokes RD as rdconstraints

self.m <--> rd.referentself.m <--> rd.ontological_category

“you” specifies discourse role

22

construction ni3-Nsubcase of Morpheme form

constraintsself.f.orth <-- "ni3"

meaning : @Humanevokes RD as rdconstraints

self.m <--> rd.referentself.m <--> rd.ontological_categoryrd.discourse_participant_role <-- @Addresseerd.set_size <-- @Singleton

The meaning of “give” is a schema with roles

23

construction gei3-V2subcase of Morpheme form

constraintsself.f.orth <-- "gei3"

meaning : Give

schema Givesubcase of Transferconstraints

inherent_aspect <-- @Inherent_Achievementgiver <-- @Animaterecipient <-- @Animatetheme <-- @Manipulable_Inanimate_Object

schema Transfersubcase of Actionroles

giver : @Entityrecipient : @Entitytheme : @Entity

constraintsgiver <--> protagonist

Finally, you-give-aunt links up the roles

24

construction ni3-gei3-yi2subcase of Finite_Clause constructional

constituentsn : ni3-Ng : gei3-V2y : yi2-N

formconstraints

n.f meets g.fg.f meets y.f

meaning : Giveconstraints

self.m <--> g.mself.m.giver <--> n.mself.m.recipient <--> y.m

The learning loop: Hypothesize & Reorganize

25

Linguistic Knowledge

Discourse & Situational

Context

Discourse & Situational

Context

AnalysisAnalysis

World Knowledge

Context FittingContext Fitting

26

XIXI

INV

Discourse Segment

addressee speaker

If the learner has a ditransitive cxn

meets

meets

ni3ni3 AddresseeAddressee

giver

GiveGivegei3gei3

recipient

AuntAuntyi2yi2

omittedomitted

theme

Peach

MOT

27

XIXI

giver

Give

INV

recipient

Discourse Segment

addressee speaker

Context fitting recovers more relations

meets

meets


giver

GiveGivegei3gei3

recipient

AuntAuntyi2yi2

omittedomitted

theme

Peach

MOT

theme

attentional-focus

28

giver

Discourse Segment

addressee speaker

Peach

MOT

theme

attentional-focus

recipient

XIXI

Give

INV

But the learner does not yet have phrasal cxns


GiveGivegei3gei3

AuntAuntyi2yi2

giver

recipientmeets

meets

29

Context bootstraps learning


GiveGivegei3gei3

AuntAuntyi2yi2

meets

meets

construction ni3-gei3-yi2subcase of Finite_Clause constructional

constituentsn : ni3g : gei3y : yi2

formconstraints

n.f meets g.fg.f meets y.f

meaning : Giveconstraints

self.m <--> g.mself.m.giver <--> n.mself.m.recipient <--> y.m

giver

recipient

30

A model of context is key to learning

The context model makes it possible for the learning model to:

learn new constructions using contextually available information

learn argument-structure constructions in pro-drop languages

31

Understanding an utterance in context

Schemas + Constructions

Simulation

Simulation

Transcripts

Events + UtterancesEvents + Utterances

Semantic Specification

Analysis + ResolutionAnalysis + Resolution


Context Model

Recency ModelRecency Model

32

Context model: Events + Utterances

Setting

participants,entities, & relations

Start Event Event DS

Sub-Event Sub-Event

33

Entities and Relations are instantiated

Setting

CHI, MOT (incl. body parts)livingroom (incl. ground, ceiling, chair, etc), lotion

Sta

rt

apply02

applier = CHIsubstance = lotionsurface = face(CHI)

ds04

admonishing05speaker = MOTaddressee = CHIforcefulness = normal

caused_motion01

forceful_motion motion

translational_motion03

mover = lotionspg = SPG

The context model is updated dynamically

Extended transcript annotation: speech acts & events

Simulator inserts events into context model & updates it with the effects

Some relations persists over time; some don’t.

34

Simulation

Simulation

EventsEvents

Context Model


35

Competition-based analyzer finds the best analysis

An analysis is made up of: A constructional tree

A semantic specification

A set of resolutions

Bill gave Mary the book

MaryBill

Ref-Exp Ref-Exp Ref-ExpGive

A-GIVE-B-X

subj v obj1 obj2

book01

@Man @WomanGive-Action @Book

giver

recipient

theme

36

Combined score that determines best-fit

Syntactic Fit: Constituency relations

Combine with preferences on non-local elements

Conditioned on syntactic context

Antecedent Fit: Ability to find referents in the context

Conditioned on syntactic information, feature agreement

Semantic Fit: Semantic bindings for frame roles

Frame roles’ fillers are scored

37

XIXI

giver

Give

INV

recipient

Discourse Segment

addressee speaker

Context Fitting goes beyond resolution

meets

meets


giver

GiveGivegei3gei3

recipient

AuntAuntyi2yi2

omittedomitted

theme

Peach

MOT

theme

attentional-focus

38

Context Fitting, a.k.a. intention reading

Context Fitting takes resolution a step further considers entire context model, ranked by

recency

considers relations amongst entities

heuristically fits from top down, e.g.

• discourse-related entities

• complex processes

• simple processes

• other structured and unstructured entities

more heuristics for future events (e.g. in cases of commands or suggestions)

39

Adult grammar size

~615 constructions total

~100 abstract cxns (26 to capture lexical variants)

~70 phrasal/clausal cxns

~440 lexical cxns (~260 open class)

~195 schemas (~120 open class, ~75 closed class)

40

Starter learner grammar size

No grammatical categories (except interjections)

Lexical items only

~440 lexical constructions ~260 open class: schema / ontology meanings

~40 closed class: pronouns, negation markers, etc

~60 function words: no meanings

~195 schemas (~120 open class, ~75 closed class)

41

The process hierarchy defined in schemas

ProcessProcess

StateState ActionActionState_

ChangeState_

Change

Complex_ProcessComplex_Process

Proto_TransitiveProto_Transitive

Intransitive_State

Intransitive_State

Two_Participant_State

Two_Participant_State

Mental_StateMental_State

Joint_MotionJoint_Motion Caused_MotionCaused_Motion

Concurrent_Processes

Concurrent_Processes Cause_EffectCause_Effect

Serial_ProcessesSerial_Processes

42


ActionAction

Translational_Motion

Translational_Motion

Translational_Self_Motion

Translational_Self_Motion

MotionMotionIntransitive_ActionIntransitive_Action

ExpressionExpression Self_MotionSelf_Motion

Force_ApplicationForce_Application

Continuous_Force_Application

Continuous_Force_Application

Agentive_ImpactAgentive_Impact

Forceful_MotionForceful_Motion

43


ActionAction

PerceptionPerception

IngestionIngestion

CommunicationCommunication

TransferTransfer

Cause_ChangeCause_Change

Other_Transitive_Action

Other_Transitive_Action

ObtainmentObtainment

44

Understanding an utterance in context

Schemas + Constructions

Simulation

Simulation

Transcripts

Events + UtterancesEvents + Utterances

Semantic Specification

Analysis + ResolutionAnalysis + Resolution


Context Model


Hypothesize & Reorganize

Hypothesize: utterance-driven;

relies on the analysis (SemSpec & context)

operations: compose

Reorganize: grammar-driven;

can be triggered by usage (to be determined)

operations: generalize

45

Composing new constructions

Compose operation:If roles from different constructions

point to the same context element, propose a new construction and set up a meaning binding.

46

ni3ni3 Addressee Addressee

Give Givegei3gei3

giverrecipienttheme

XIXI

INV

Peach

MOT

Creating pivot constructions

Pivot generalization:Given a phrasal cxn, look for another cxn

that shares 1+ constituents. Line up roles and bindings. Create new cxn category for the slot.

47


Give Givegei3gei3

@Aunt @Auntyi2yi2

giver

recipientmeets

meets


Give Givegei3gei3

@Human @Humanwo3wo3

giver

recipientmeets

meets

48

Resulting constructions

construction ni3-gei3-cat01constituents

ni3, gei3, cat01meaning : Give

constraintsself.m.recipient <--> g.m

construction wo3subcase of cat01meaning: @Human

construction yi2subcase of cat01meaning: @Aunt

general construction cat01subcase of Morphememeaning: @Human

Pilot Results: Sample constructions learned

Composed:

Pivot Cxns:

49

chi1_fan4ni3_chuan1_xie2ni3_shuo1bu4_na2wo3_qu4ni3_ping2zi_gei3_wo3ni3_gei3_yi2wo3_bu4_chi1

eat riceyou wear shoeyou sayNEG takeI goyou bottle give meyou give auntI NEG eat

ni3 {shuo1, chuan1}ni3 {shuo1, hua4}wo3 {zhao3, qu4}bu4 {na2, he1}{wo3, ma1} cheng2

you {say, wear}you {say, draw}I {find, go}NEG {take, drink}{I, mom} scoop

Challenge #1: Non-compositional meaning

Non-compositional meaning:Search for additional meaning schemas

(in context or in general) that relate the meanings of the individual constructions

50

youyou Addressee Addressee

Bake Bakebakebake

bakerbaked

Bake-Event

CHI

Cake

MOT

a cake

a cake

@Cake @Cake

Give-Event

Challenge #2: Function words

Function words tend to indicate relations rather than events or entities

51

youyou Addressee Addressee

Bake Bakebakebake

bakerbaked

Bake-Event

CHI

Cake

MOT

a cake

a cake

@Cake @Cake

forfor

Benefaction

Challenge #3: How far up to generalize

Eat rice

Eat apple

Eat watermelon

Want rice

Want apple

Want chair

52

Inanimate ObjectInanimate Object

ManipulableObjects

ManipulableObjects

Unmovable Objects

Unmovable Objects

FoodFood FurnitureFurniture

FruitFruit SavorySavory ChairChair SofaSofa

appleapple watermelon

watermelon

ricerice

Challenge #4: Beyond pivot constructions

Pivot constructions: indexing on particular constituent type

Eat rice; Eat apple; Eat watermelon

Abstract constructions: indexing on role-filler relations between constituents

53

Schema Eatroles

eater <--> agent

food <--> patient

food

Eat catX

Schema Wantroles

wanter <--> agentwanted <-->

patient

wanted

Want catY

Challenge #5: Omissible constituents

Intuition:

Same context, two expressions that differ by one constituent a general construction with the constituent being omissible

May require verbatim memory traces of utterances + “relevant” context

54

When does the learning stop?

Most likely grammar given utterances and context

The grammar prior is a preference for the “kind” of grammar

In practice, take the log and minimize cost Minimum Description Length (MDL)

55

)(),|(argmax

),|(argmaxˆ

GPZGUP

ZUGPG

G

G

Bayesian Learning FrameworkSchemas +

Constructions

SemSpec

Analysis +

Resolution

Analysis +

Resolution

Context Fitting

Context Fitting

Intuition for MDL

S -> Give me NP

NP -> the book

NP -> a book

S -> Give me NP

NP -> DET book

DET -> the

DET -> a

56

Suppose that the prior is inversely proportional to the size of the grammar (e.g. number of rules)

It’s not worthwhile to make this generalization

Intuition for MDL

S -> Give me NP

NP -> the book

NP -> a book

NP -> the pen

NP -> a pen

NP -> the pencil

NP -> a pencil

NP -> the marker

NP -> a marker

S -> Give me NP

NP -> DET N

DET -> the

DET -> a

N -> book

N -> pen

N -> pencil

N -> marker

57

How to calculate the prior of this grammar

(Yet to be determined)

There is evidence that the lexicalized constructions do not completely go away

If the more lexicalized constructions are retained, the size of grammar is a bad indication of degree of generality

58

A Best-Fit Approach for Productive Analysis of Omitted Arguments Eva Mok & John Bryant University of...

Documents

Transcript of A Best-Fit Approach for Productive Analysis of Omitted Arguments Eva Mok & John Bryant University of...