+ All Categories
Home > Documents > Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue...

Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue...

Date post: 16-Oct-2020
Category:
Upload: others
View: 3 times
Download: 0 times
Share this document with a friend
40
Dialogue Systems Group Task-oriented Neural Dialogue Systems Apple, 08/03/2017 Tsung-Hsien (Shawn) Wen
Transcript
Page 1: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

DialogueSystemsGroup

Task-orientedNeuralDialogueSystems

Apple,08/03/2017Tsung-Hsien(Shawn)Wen

Page 2: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

Outline

¤  Intro

¤  NeuralDialogueSystems

¤  Wizard-of-OzDataCollecLon

¤  Experiments

¤  Conclusion

2

Page 3: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

Outline

¤  Intro

¤  NeuralDialogueSystems

¤  Wizard-of-OzDataCollecLon

¤  Experiments

¤  Conclusion

3

Page 4: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

Thetwoparadigms 4

M:Hello,welcometotheCambridgedialoguesystem,Whatkindoffoodwouldyoulike?H:YeahIwanttofindarestaurantthatservesEuropeanfood.M:HotelduVinandBistroisaniceplace.ItservesEuropeanfood.H:Uhwhatistheaddressandphonenumber?M:HotelduVinandBistroisaniceplace.TheiraddressisXXXandtheirphoneisYYY.H:Thankyougoodbye.M:Thankyouforusingthissystem.

A:Howoldareyou?B:I’m16.Whyareyouasking?A:Ithoughtyouwere12.B:Whatmadeyouthinkso?A:Idon’tknowwhatyouaretalkingabout.B:Youdon’tknowwhatyouaresaying.

[Suetal,2016]

[Lietal,2016]

Page 5: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

Thetwoparadigms 5

¤  Task-OrientedDialogueSystems¤  Goal-oriented¤  Requirepreciseunderstanding,hardtocollectdata.¤  Modular,highlyhandcraced,restrictedability,butmeaningful/

usefulsystems.¤  Chat-basedConversaLonalAgents

¤  Chit-chat(non-goal).¤  Vastamountofdata(butprobablynothelpful).¤  End-to-end,highlydata-driven,butmeaningless/inappropriate

responses,unreliablesystems.

¤  Canwetrainauseful(completetasks)dialoguesystemdirectlyfromdata?

¤  Howcanwecollectthedatatotrainthismodel?

Page 6: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

Outline

¤  Intro

¤  NeuralDialogueSystems

¤  Wizard-of-OzDataCollecLon

¤  Experiments

¤  Conclusion&Discussion

6

Page 7: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

TradiLonalDialogueSystems

SpeechRecogniLon

LanguageUnderstanding

SpeechSynthesis

DialogueManager

KB

Web

DialogueSystem

LanguageGeneraLon

7

text

text

Page 8: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

NeuralDialogueSystems

SpeechRecogniLon

SpeechSynthesis

KB

Web

NeuralDialogueSystem

8

text

text

Page 9: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

CanIhaveKorean LifleSeoulservesgreatKorean.

ANetwork-basedEnd-to-EndTrainableTask-OrientedDialogueSystem,Wenetal,2016

Page 10: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

CanIhave<v.food> <v.name>servesgreat<v.food>.

ANetwork-basedEnd-to-EndTrainableTask-OrientedDialogueSystem,Wenetal,2016

DelexicalisaLon

Page 11: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

IntentNetwork

CanIhave<v.food>

GeneraLonNetwork <v.name>servesgreat<v.food>.

zt

Seq2Seq

ANetwork-basedEnd-to-EndTrainableTask-OrientedDialogueSystem,Wenetal,2016

Page 12: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

CanIhavekorean

Korean0.7BriLsh0.2French0.1

BeliefTracker

IntentNetwork

CanIhave<v.food>

GeneraLonNetwork <v.name>servesgreat<v.food>.

zt

pt

LanguageGrounding

ANetwork-basedEnd-to-EndTrainableTask-OrientedDialogueSystem,Wenetal,2016

Page 13: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

<nil>

I

want

Korean

food

<nil>

JordanRNN-CNNbelieftrackers

1stconv. 2ndconv. 3rdconv. max-pool avg-pool

TurntInputlayer

Outputlayer

Hiddenlayer

DelexicalisedCNN

<nil>

I

want

v.food

s.food

<nil>

sentencerepresentaLon

BriLshFrenchKorean…Chinese 1.3 2.3 9.7 1.2 .01.02.85.01

13

[Hendersonetal,2014]

Memorisethedelex.posiLon

Padzerostohavethe

samelength

Slot-specificdelex.ngram

feature

Value-specificdelex.ngramplaceholder

Value-specificdelex.ngram

feature

Page 14: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

JordanRNN-CNNbelieftrackers

1stconv. 2ndconv. 3rdconv. max-pool avg-pool

Userturnt Systemturnt-1

TurntInputlayer

Outputlayer

Hiddenlayer

DelexicalisedCNN

JordanRNN

<nil>

I

want

v.food

s.food

<nil>

sentencerepresentaLon

14

Page 15: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

CanIhavekorean

Korean0.7BriLsh0.2French0.1

BeliefTracker

IntentNetwork

CanIhave<v.food>

GeneraLonNetwork <v.name>servesgreat<v.food>.

zt

pt

LanguageGrounding

ANetwork-basedEnd-to-EndTrainableTask-OrientedDialogueSystem,Wenetal,2016

Page 16: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

CanIhavekorean

Korean0.7BriLsh0.2French0.1

BeliefTracker

000…01

MySQLquery:“Select*wherefood=Korean”

DatabaseOperator

IntentNetwork

CanIhave<v.food>

GeneraLonNetwork <v.name>servesgreat<v.food>.

Database

Sevendays CurryPrince

Nirala

RoyalStandard

LifleSeuol

DBpointer xt

zt

pt

qt

DatabaseAccessing

ANetwork-basedEnd-to-EndTrainableTask-OrientedDialogueSystem,Wenetal,2016

Page 17: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

CanIhavekorean

Korean0.7BriLsh0.2French0.1

BeliefTracker

000…01

MySQLquery:“Select*wherefood=Korean”

DatabaseOperator

IntentNetwork

CanIhave<v.food>

GeneraLonNetwork <v.name>servesgreat<v.food>.

PolicyNetwork

Database

Sevendays CurryPrince

Nirala

RoyalStandard

LifleSeuol

DBpointer xt

zt

pt

qt

DecisionMaking

ANetwork-basedEnd-to-EndTrainableTask-OrientedDialogueSystem,Wenetal,2016

Page 18: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

CanIhavekorean

Korean0.7BriLsh0.2French0.1

BeliefTracker

000…01

MySQLquery:“Select*wherefood=Korean”

DatabaseOperator

IntentNetwork

CanIhave<v.food>

GeneraLonNetwork <v.name>servesgreat<v.food>.

PolicyNetwork Copyfield

Database

Sevendays CurryPrince

Nirala

RoyalStandard

LifleSeuol

DBpointer xt

zt

pt

qt

ANetwork-basedEnd-to-EndTrainableTask-OrientedDialogueSystem,Wenetal,2016

Page 19: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

Outline

¤  Intro

¤  NeuralDialogueSystems

¤  Wizard-of-OzDataCollecCon

¤  Experiments

¤  Conclusion

19

Page 20: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

WizardofOzDataCollecLon 20

Hi,IwantacheapKoreanrestaurant.

Whatareaareyoulookingfor?

1

Task:Findarestaurant,cheap,Korean,NorthAskphonenumber

1

Whatuserwants?

Food Korean

Price Cheap

Area N/A

SearchTable

LifleSeoul …

BestKorea …

Page 21: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

WizardofOzDataCollecLon 21

Hi,IwantacheapKoreanrestaurant.

Whatareaareyoulookingfor?

Somewhereinthenorth.

Li>leSeoulisniceoneinthenorth. 2

Task:Findarestaurant,cheap,Korean,NorthAskphonenumber

2

Whatuserwants?

Food Korean

Price Cheap

Area North

SearchTable

LifleSeoul …

Page 22: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

WizardofOzDataCollecLon 22

Hi,IwantacheapKoreanrestaurant.

Whatareaareyoulookingfor?

Somewhereinthenorth.

Li>leSeoulisniceoneinthenorth.

Itsphonenumberis01223456789.

Whatisthephonenumber? 3

Task:Findarestaurant,cheap,Korean,NorthAskphonenumber

3

Whatuserwants?

Food` Korean

Price Cheap

Area North

SearchTable

LifleSeoul …

Page 23: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

WizardofOzDataCollecLon 23

Hi,IwantacheapKoreanrestaurant.

Whatareaareyoulookingfor?

Somewhereinthenorth.

Li>leSeoulisniceoneinthenorth.

Itsphonenumberis01223456789.

Whatisthephonenumber?

Thankyouverymuch,goodbye.

Thankyouforusingthesystem.

4

Task:Findarestaurant,cheap,Korean,NorthAskphonenumber

4

Whatuserwants?

Food Korean

Price Cheap

Area North

SearchTable

LifleSeoul …

Page 24: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

WizardofOzDataCollecLon 24

Hi,IwantacheapKoreanrestaurant.

Whatareaareyoulookingfor?

Somewhereinthenorth.

Li>leSeoulisniceoneinthenorth.

Itsphonenumberis01223456789.

Whatisthephonenumber?

Thankyouverymuch,goodbye.

Thankyouforusingthesystem.

Whatuserwants?

Food Korean

Price Cheap

Area North

Page 25: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

WizardofOzDataCollecLon 25

¤  OnlineparallelversionofWOZonMTurk¤  Randomlyhireaworkertobeuser/wizard.¤  Task:Enteranappropriateresponseforoneturn.¤  RepeattheprocessunLlalldialoguesarefinished.

¤  Exampleuserpage

Page 26: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

WizardofOzDataCollecLon 26

¤  Examplewizardpage

Page 27: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

CamRest676dataset 27

¤  Ontology:¤  Cambridgerestaurantdomain,99venues.¤  3informableslots: area,pricerange,foodtype¤  3requestableslots: address,phone,postcode

¤  Dataset¤  676dialogues,~2750turns¤  3000HITS,takes3days,costs~400USD¤  Datacleaningtakes2-3daysforoneperson

Link:hfps://www.repository.cam.ac.uk/handle/1810/260970

Page 28: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

Outline

¤  Intro

¤  NeuralDialogueSystems

¤  Wizard-of-OzDataCollecLon

¤  Experiments

¤  Conclusion

28

Page 29: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

Experiments 29

¤  Experimentaldetails¤  Train/valid/test:3/1/1¤  SGD,l2regularisaLon,earlystopping,gradientclip=1¤  Hiddensize=50,Vocabsize:~500

¤  Twostagetraining:¤  Trainingtrackerswithlabelcrossentropy¤  Trainingotherpartswithresponsecrossentropy

¤  Decoding¤  Beamsearchw/beamwidth10¤  Decodewithaveragewordlikelihood

Page 30: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

ResponseGeneraLonTask 30

Model Match(%) Success(%) BLEU

Seq2Seq[Sutskeveretal,2014] - - 0.1718

HRED[Serbanetal,2015] - - 0.1861

Ourmodelw/oreq.trackers 89.70 30.60 0.1799 Ourfullmodel 86.34 75.16 0.2313 Ourfullmodel+afenLon 90.88 80.02 0.2388

Page 31: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

HumanevaluaLon 31

Qualityassessment SystemComparison

Page 32: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

Exampledialogues 32

Page 33: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

Exampledialogues 33

Page 34: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

VisualisingacLonembedding 34

Page 35: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

Outline

¤  Intro

¤  NeuralDialogueSystems

¤  Wizard-of-OzDataCollecLon

¤  Experiments

¤  Conclusion

35

Page 36: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

Conclusion

¤  Anend-to-endtrainabletask-orienteddialoguesystemarchitectureisintroduced.

¤  AcomplementaryWOZdatacollecLonisusedtocollectthetrainingdata(nolatency,parallel,cheap).

¤  Resultsshowthatitcanlearnfromhuman-humanconversaLonsandhelpuserstocompletetasks.

¤  Explicitlanguagegroundingiscrucial,butwhatisthebestwaytorepresentsemanLcs?

36

Page 37: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

FutureWork 37

¤  LatentIntenLonDialogueModels(underreview)¤  Learnanembeddedlatentpolicyfromasupervised

corpus.¤  Fine-tunepolicyusingreinforcementlearning.

¤  MulL-domainNeuralDialogueSystems¤  CollectWOZdataacrossseveraldomains.¤  Trainaneuralcontrollertoread/writememorytapes

(trackers)andemitresponses.

Page 38: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

Thepaper

¤  Tsung-HsienWen,DavidVandyke,NikolaMrksic,MilicaGasic,LinaM.R.Barahona,Pei-HaoSu,StefanUltes,andSteveYoung.ANetwork-basedEnd-to-End Trainable Task-orientedDialogue System. To appear EACL2017.

¤  Tsung-HsienWen,MilicaGasic,NikolaMrksic, LinaM.Rojas-Barahona,Pei-Hao Su, Stefan Ultes, David Vandyke, Steve Young. CondiConalGeneraConandSnapshotLearninginNeuralDialogueSystems.EMNLP2016.

38

Page 39: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

References

¤  P-H.Su,M.Gasic,N.Mrksic,L.Rojas-Barahona,S.Ultes,D.Vandyke,T-H. Wen, and S. Young. On-line AcCve Reward Learning for PolicyOpCmisaConinSpokenDialogueSystems,ACL2016.

¤  M. Henderson, B. Thomson and S. Young.Word-Based Dialog StateTrackingwithRecurrentNeuralNetworks,SigDial2014.

¤  J.Li,W.Monroe,A.Rifer,D.Jurafsky.DeepReinforcementLearningforDialogueGeneraCon,EMNLP2016.

39

Page 40: Task-oriented Neural Dialogue Systems - GitHub Pages · End-to-End Trainable Task-oriented Dialogue System. To appear EACL 2017. ¤ Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Lina

DialogueSystemsGroup

Thankyou!QuesLons?

Tsung-HsienWenissupportedbyastudentshipfundedbyToshibaResearchEuropeLtd,CambridgeResearchLaboratory


Recommended