scieee Open visual document viewer

Dialogue Management in a Home Machine Environment: Linguistic Components over an Agent Architecture

Quesada Moreno, José Francisco; García, Federico; Sena, Esther; Bernal, José Ángel; Amores Carredano, José Gabriel De

Abstract

This paper presents the main characteristics of an Agent{based Architecture for the design and implementation of a Spoken Dialogue System. From a theoretical point of view, the system is based on the Information State Update approach, in particular, the system aims at the management of Natural Command Language Dialogue Moves in a Home Machine Environment. Speci cally, the paper is focused on the Natural Language Understanding and Dialogue Management Agents, and discusses their integration over a global agent architecture (which includes Action and Knowledge Managers, Speech Input/Output components and HomeSetup controllers).

Full text

Dialogue Managemen in a Home Machine En i onmen : Linguis ic Comp onen s o e an Agen A chi ec u e Jose F. Quesada, Fede ico Ga ca, Es he Sena, Jose Angel Be nal, Gab iel Amo es G up o de In es igacion Julie a Uni e sidad de Se illa Abs ac This pap e p esen s he main cha ac e is ics o an Agen {based A chi- ec u e o he design and implemen a- ion o a Sp oken Dialogue Sys em. F om a heo e ical p oin o iew, he sys em is based on he In o ma ion S a e Up da e app oach, in pa icula , he sys em aims a he managemen o Na u al Command Language Dialogue Mo es in a Home Machine En i onmen . Sp ecically, he pap e is o cused on he Na u al Lan- guage Unde s anding and Dialogue Man- agemen Agen s, and discusses hei in e- g a ion o e a global agen a chi ec u e (which includes Ac ion and Knowledge Manage s, Sp eech Inpu /Ou pu comp o- nen s and HomeSe up con olle s). 1 1 In oduc ion Sp oken Dialogue Sys ems ha e b ecome one o he mos p ominen esea ch a eas in Compu- a ional Linguis ics. O e he las ew yea s, se e al p o jec s all o e he wo ld ha e o- cused on his opic. This pap e desc ib es he main cha ac e is ics o a sp oken dialogue sys- em aimed a he con ol o a ligh ning sce- na io in a home machine en i onmen . F om a unc ional p oin o iew, he sys em may b e desc ib ed as a:  Sp oken Dialogue Sys em, which ac s as a na u al language in e ace b e ween use s and  Home De ices ins alled in a Home Ma- chine En i onmen  designed and implemen ed o e an Agen A chi ec u e (sp ecically, 1 This wo k has b een supp o ed by D'Homme (Di- alogues in he Home Machine En i onmen ), EC P o jec IST-2000-26280. The au ho s would like o hanks he es o he p o jec pa ne s o hei com- men s and sugges ions on he wo k. OAA - Op en Agen A chi ec u e [Ma in e al 99]). Sec ion 2 desc ib es he o e all a chi ec- u e and agen s, mainly he HomeSe up, KnowledgeManage , Ac ionManage , Tex - To-Sp eech and Dialogue Manage Agen s. One o he main goals o he Na u al Lan- guage Unde s anding comp onen o he sys- em is b o h o de ec he command and he de ice ha he use wan s o con ol and o se he le el o p e cen age ha he use sp ec- ies a a p ecise momen . Fo ins ance, he sen ence "Se he ki chen ligh a le el 5" , mus be ans o med in o a seman ic ep e- sen a ion, whe e he main ea u es will b e:  Command :Se  De ice : Ligh  Lo ca ion : Ki chen  Le el :5 Taking his in o ma ion as i s inpu , he Dialogue Manage mus be able o de ec i he seman ic s uc u e is:  Func ionally comple e (Which ligh do you wan o u n on i you ha e jus e- cei ed "Swi ch on he ligh " ?)  P agma ically cohe en and echnically p ossible (you canno swi ch on a do o , o ins ance). Besides, he Dialogue Manage mus cla - i y ambiguous exp essions, such as "Swi ch on he li ing oom ligh " , when he e a e se - e al ligh s in he li ing o om. In o de o sol e hese p oblems, he Di- alogue Manage is closely connec ed o b o h he Na u al Language Unde s anding and he Knowledge Manage mo dules. Nex , we p esen he linguis ic comp onen s o he a chi ec u e, ha is, he Na u al Lan- guage Unde s anding and he Dialogue Man- age Agen s. A his le el, i is wo h no ing ha he sys em allows he ollowing unc ion- ali y:  Na u al Command Languages  Exp ec a ions  Mul iple Commands  De aul In e p e a ion A de ailed desc ip ion o he ke nel o he Dialogue Manage may be ound in [Quesada e al 00, Quesada, To e & Amo es 2000 ], including he sp ecica ion language o he desc ip ion o dialogue sys ems and he s a egies o he implemen a ion o he p e ious unc ions, mainly he manipula ion o exp ec a ions and he use o de aul in e p e a ions. Sec ion 3 desc ib es he Na u al Language Unde s anding mo dule, paying sp ecial a en- ion o he g amma ical le el: a unica ion{ based seman ic{o ien ed g amma sp ecica- ion. The in e ace b e ween he Na u al Lan- guage Unde anding and he Dialogue Man- age agen s is based on he DTAC p o o col ([Amo es & Quesada 2000 ]). Sec ion 4 in- cludes an in e es ing se o examples inco - p o a ing o each example he DTAC ep- esen a ion o he sen ence ob ained by he g amma p esen ed in sec ion 3. Finally, sec ion 5 highligh s one o he main challenges o he sys em: o ob ain a na u- al language in e ace o he Home Machine En i onmen . This means ha a use mus be able o ins all, plug o unplug, and con- gu e new o p e ious de ices and o wo k and alk wi h hem on he y. This sec ion includes some examples which illus a es he main unc ions supp o ed by he sys em so a . 2 O e al l A chi ec u e and Agen s The sys em has b een implemen ed using he Op en Agen A chi ec u e [Ma in e al 99]. Figu e 1 illus a es he main agen s apa om he Facili a o , he cen al agen in cha ge o he communica ion be ween he es o agen s. Figu e 1: O e all agen a chi ec u e.  HomeSe up : This agen isincha ge o he simula ion o a Home Machine En i- onmen . This agen mainly allows he ins alla ion o new de ices and hei con- gu a ion. As pa o he congu a ion he use may ins an ia e he se o de- sc ip o s ela ed o he de ice.  KnowledgeManage : This agen con- ains b o h he s a ic (s uc u e o he house, gene al on ology, ...) and dy- namic (ins alla ion, congu a ion and s a us o he de ices) knowledge in- ol ed in he ask.  Ac ionManage : The cu en de ices will supp o he ac ual unc ionali y o he sys em. This unc ionali y ( u n on/o he de ice, ask he cu en s a us, change he le el o a dimme able de ice, ...) mus b e accessed h ough he Home- Se up agen , in cha ge o he simula ion and moni o ing o he scena io. Ne e - heless he de ices inco p o a e p imi i e unc ions o he sys em. The Ac ion- Manage inco p o a es new highe le el unc ionali y ha will b e ansla ed in o sequences o p imi i e ac ions.  Tex - o-Sp eech : In cha ge o he gen- e a ion o he sp eech ou pu . In o de o do so we ha e de elop ed an OAA- w app e o Fes i al, allowing in his way he use o his sys em as pa o he gen- e al a chi ec u e.  Dialogue Managemen : This agen includes he Use In e ace (which con- ols a a high le el he communica ion wi h he use ), he Pa se , he Seman- ic In e p e e and he Dialogue Man- age comp onen s. The Knowledge Manage is in cha ge o s o ing and o ganizing he in o ma ion e- la ed o he de ices ins alled in he house ega ding ubica ion, cha ac e is ics and e- la ions wi h o he de ices. The app oach adop ed o o ganizing his in o ma ion is based on he con o ma ion o a seman ic ne - wo k. The p op e ies o his ne wo k a e:  No des o ganized in a g aph op ology  Non- di ec ed connec ions be ween no des.  Non- conexed.  Non- cyclical.  Pa en -child ela ion dened in each conec ion b e ween no des. Each no de con ains a lis o comp onen s. A comp onen consis s o a pai o s ings: a lab el and a alue. We can desc ib e he eby he s uc u e o he ne wo k as ha ing wo ela ions:  A comp onen b elongs o a no de. componen  node  A no de includes o he no de. node 1  node 2 e.i. node 1 is pa en o node 2. Le see an example o seman ic ne wo k: g ound < indoo s g ound < ou doo s indoo s < ki chen indoo s < den indoo s < co ido indoo s < li ing_ oom ou doo s < on ou doo s < back whe e he symbol  is ep esen ed by < . The gu e 2 ep esen s he g aph con o ming he ne wo k sp ecied. Figu e 2: G aph ep esen a ion o a seman ic ne wo k example. 3 Na u al Language Unde s anding Dialogue sys ems enable use s o in e ac wi h compu e sys ems ia na u al and in- elligen dialogues, as hey would wi h hu- man agen s. De elopmen o such sys ems equi es a wide ange o sp eech and language echnologies, including na u al language and dialogue p o cessing, o de e mine he mean- ings and in en ions o he u e ances and o gene a e a ex - o- ex in e ac ion be ween he use and he sys em. As ega ds his do cumen , wha we ha e done is o de elop an in e ace o b o h En- glish and Spanish o he sp ecica ion o di - e en unc ions in an in elligen house sce- na io. In o de o do so, we ha e made use o Na u al Command Languages. Na u al Command Languages comp ise he se o in- pu and ou pu na u al language exp essions which a e accep able in a gi en applica ion domain. The mos ou s anding cha ac e - is ics o hese NCLs a e he use o na u al language o cabula y and syn ac ic s uc u e. Also, NCLs eec na u al syn ac ic o de and mo e p ecisely he pa e ns o sp oken language. In ac , hey a e a sub-language, wi h easonably clea b ounda ies bu also one in which new unc ionali y can b e added wi hou al e ing he s uc u al p op e ies o he language. Nex subsec ion p esen s he main ules o he unica ion- based g amma ha has b een implemen ed. A his le el i is wo h no ing ha he g amma inco p o a es Non{ e minal ypica ion, and includes ules o cons uc a Na u al Command Language Dialogue ep- esen a ion mo del, based on Dialogue Mo es whe e DMOVE ep esen s he dialogue mo e i sel [Amo es & Quesada 2000 ], TYPE he sp ecic yp e o dialogue mo e, ARGS he unc ional sub ca ego iza ion mo del o he yp e o dialogue mo e and CONT he ac ual con en s o e minal s uc u es. 3.1 Unica ion{based and Seman ic{o ien ed G amma Roo sO G amma : // Pa ame e s Value De ice // Commands Command This is a congu a ion pa ame e used by he pa se , in ac , i is he only one ha he pa se uses. On he o he hand, he unie is he one in cha ge o he es o he pa am- e e s. By means o hese Ro o so G amma , all g amma symb ols ha may b e ea ed as o o s a e sp ecied. This in u n le us ha e g amma s wi h se e al o o symb ols. NonTe minalTypes: // Pa ame e s De ice = (DMOVE: speci yPa ame e , TYPE: De ice) Numbe = (DMOVE: speci yPa ame e , TYPE: Value, SUBTYPE: Numbe ) Le el = (DMOVE: speci yPa ame e , TYPE: Value, SUBTYPE: Le el) Pe cen age = (DMOVE: speci yPa ame e , TYPE: Value, SUBTYPE: Pe cen age) // Commands CommandOn = (DMOVE: speci yCommand, TYPE: CommandOn, ARGS: [De ice]) CommandO = (DMOVE: speci yCommand, TYPE: CommandO , ARGS: [De ice]) CommandIs = (DMOVE: speci yCommand, TYPE: CommandIS , ARGS: [De ice]) CommandSe = (DMOVE: speci yCommand, TYPE: CommandSe , ARGS: [De ice,Value]) CommandInc = (DMOVE: speci yCommand, TYPE: CommandInc, ARGS: [De ice,Value]) CommandDec = (DMOVE: speci yCommand, TYPE: CommandDec, ARGS: [De ice,Value]) AskHelp = (DMOVE: askHelp, TYPE: AskHelp) Qui = (DMOVE: qui , TYPE: Qui ) Unde he lab el NonTe minalTyp es, we ha e b o h pa ame e s and commands. Bo h yp es will b e he ones ha will app ea in he DTAC s uc u es ha he Dialogue Manage will ake as i s inpu . This can be clea ly no iced in sec ion 4. As ega ds pa ame- e s, he e a e ou , namely de ice, p e cen - age, numb e and le el. The lab el de ice may s and o wo ds such as ki chen", he lab el numbe s ands o a plain numbe such as 5", and he wo d le el along wi h p e cen - age, indica e ei he a le el such as a le el 5", o a p e cen age such as o 25 p e cen " e- sp ec i ely. These pa ame e s will b e ins an i- a ed as he a gumen s o he commands, bu hey will b e enclosed unde he lab el alue", ha is, his alue oge he wi h he de ice, will be wha will ac ually app ea as he a - gumen o he command. The e o e, p e cen - age, numb e and le el will b e alue sub yp es. Weha e c ea ed his cons uc o a oid ambi- gui ies in b o h he g amma and he seman ic ep esen a ion. Commands a e di ided in o eigh yp es. The  s g oup o commands (CommandOn, CommandO, and Comman- dIs ) akes a de ice as i s a gumen . A his p oin , he use may ask he sys em o swi ch on a ligh , o swi ch i o, o he may jus b e asking o he s a us o a ligh . The second se o commands a e he ones used wi h dim- me able de ices, he e o e hey ake a de ice plus a alue. They a e used o se a ligh a any le el, p e cen age o numb e , o inc ease i , o o dec ease i . The hi d se consis s o wo commands, one o hem o asking help, and he o he one o qui ing he sys em. As i can be no iced b elow, he do no equi e any pa ame e s. /////////// PARAMETERS (A100 : AuxDe ice -> LAuxDe ice) @up = @sel -1; g (A101 : AuxDe ice -> AuxDe ice LAuxDe ice) @up = @sel -2; g (P100 : De ice -> LKeyDe ice ) @up = @sel -1; g (P102 : De ice ->AuxDe ice LKeyDe ice ) @up = @sel -2; g (L100 : AuxPe c -> LAuxPe c ) @up = @sel -1; g (L101 : AuxLe el -> LAuxLe el) @up = @sel -1; g (L102 : Numbe -> LNumbe ) @up = @sel -1; g (L104 : Le el -> AuxLe el LNumbe ) @up = @sel -2; g (L105 : Le el -> LNumbe AuxLe el ) @up = @sel -1; g (L106 : Pe cen age -> LNumbe AuxPe c ) @up = @sel -1; g (L107 : Pe cen age -> AuxPe c LNumbe ) @up = @sel -2; g (L110 : Value -> Numbe ) @up = @sel -1; g (L111 : Value -> Le el ) @up = @sel -1; g (L112 : Value -> Pe cen age ) @up = @sel -1; g (L120 : A gs -> De ice ) @up.De ice = @sel -1; g (L121 : A gs -> Value ) @up.Value = @sel -1; g In his pa o he g amma , we can see he ules wi h he unc ional equa ions asso cia ed o hem. All o hem a e pa ame e ules. Wo ds b eginning wi h L a e e minal ca - ego ies. Fo ins ance: LAuxDe ice, LAux- Pe c,LNumb e ,LKeyDe ice. We ha e wo ds such as le el", p e cen age" and ligh ", which b elong o he co esp onding ca ego ies AuxLe el", AuxPe cen age" and AuxDe- ice". Also, as de ice is a non- e minal yp e, LkeyDe ice" is he lexical ca ego y s anding o wo ds such as ki chen". An A gs can b e ei he a alue o a de ice, and a alue can b e ei he a le el o a p e cen age. The c ea ion o hese cons uc s and he way in which he ules ha e b een implemen ed, allows us b o h o ha e any kind o wo d o de and o a oid g amma ical ambigui y. /////////// COMMANDS (C200 : CommandOn -> LCommandOn) @up = @sel -1; g (C201 : CommandO -> LCommandO ) @up = @sel -1; g (C213 : CommandIs -> LAuxIs B) @up.TYPE =a CommandIS ; g (C214 : CommandIs -> LAuxIs A LAuxIs B) @up.TYPE =a CommandIS ; g (C215 : CommandIs -> LAuxIs B LAuxIs C) @up.TYPE =a CommandIS ; g (C202 : CommandSe -> LCommandSe ) @up = @sel -1; g (C203 : CommandInc -> LCommandInc) @up = @sel -1; g (C204 : CommandDec -> LCommandDec) @up = @sel -1; g (C211 : XCommand -> CommandO ) @up = @sel -1; g (C212 : XCommand -> CommandIs ) @up = @sel -1; g (C213 : XCommand -> CommandOn ) @up = @sel -1; g (C220 : XCommand -> CommandSe ) @up = @sel -1; g (C221 : XCommand -> CommandInc ) @up = @sel -1; g (C222 : XCommand -> CommandDec ) @up = @sel -1; g (C223 : Command -> XCommand ) @up = @sel -1; g (C224 : Command -> XCommand A gs) @up = @sel -1; @up = @sel -2; g The way in which he command ules ha e b een a anged allows us o ha e any combi- na ion o wo ds, no ma e he o de in which hey a e ei he said o yp ed. This is a c u- cial p oin , sp ecially i we ha e in o accoun ha his is in ended o be sp oken, and sp o- ken language is by no means p edic able o xed as ega ds wo d o de . I we s a om he lexical ca ego ies, which a e he ones b e- ginning wi h L, we ge he Command*. The * s ands o On, O,Se , Inc, Dec. As we can no ice, all hese a e commands wi h a gu- men s and hey can b e o he same yp e as he ab o emen ioned pa ame e s.They can b e de- sc ib ed as sp ecic commands. We may ha e any combina ion o wo ds, sp ecically when he use wan s o ask o he s a us o a ligh . He may simply say Es ado de la co cina", o en que es ado se encuen a la co cina".In he la e , he wo keywo ds would be es- ado" and encuen a", and in he o me , he keywo d would b e would b e es ado". All he Command* a e XCommands. These a e commands wi hou a gumen s. Finally, he XCommand is inse ed wi hin a mo e gene al command which is jus a Command. /////////// ERROR REPAIR (N001 : Command -> Command No A gs) @up = @sel -1; @i (@sel -3.De ice) @ hen @up.Repai edDe ice = @sel -1.De ice; @up.De ice =a @sel -3.De ice; g @i (@sel -3.Value) @ hen @up.Repai edValue = @sel -1.Value; @up.Value =a @sel -3.Value; g g (N002 : Command -> Command No Command) @up = @sel -3; @i (!(@sel -3.De ice) && (@sel -1.De ice)) @ hen @up.De ice = @sel -1.De ice; g @i (!(@sel -3.Value) && (@sel -1.Value)) @ hen @up.Value = @sel -1.Value; g @up.Repai edCommand = @sel -1; g These ules a e used when he use makes an e o and he inmedia ely co ec s himsel . He may b e epai ing an a gumen whichmay b e a le el, p e cen age, o a de ice, he maybe changing he command, ha is o say,hemay be asking he sys em o u n o he ki chen ligh when he has jus ealized ha ha ligh is al eady o, and he may be changing b o h he command and he a gumen . This is clea ly exemplied in sec ion 4, in examples 4, 5 and 6. In numb e 4, he use is changing he command, in numbe 5, he is changing b o h he de ice and he alue, and in num- be 6, he is epai ing he command and he alue. 4 Seman ic ep esen a ion o dialogue mo es Nex , we p esen some examples which illus a e he use o he p e ious g amma . These esul s ha e b een ob ained wi h he Epis eme NLP o ol ([Quesada & Amo es 2000 ]). The goal o his sec ion is o illus a e he gene ali y and simplici y o he ep esen a ion o malism wi h a wide ange o phenomena: incomple e commands, commands wi h die en a gumen s, die en yp es o a gumen s (numb e s, desc ip o s, ...) and e o epai s, b o h o commands and pa ame e s: 1. Tu n on 2 4 DM OV E : speci y C ommand TY PE : S w dC ommandO n ARGS : [ S w dD e ice ] 3 5 2. Tu n on he ki chen ligh 2 6 6 6 6 6 6 6 6 4 DM OV E : : speci y C ommand TY PE : : S w dC ommandO n ARGS : [ S w dD e ice ] CON T : S w dD e ice : 2 4 DM OV E : speci y P a ame e TY PE : S w dD e ice CON T : k i chen 3 5 3 7 7 7 7 7 7 7 7 5 3. Dim he ki chen ligh down o 25 p e cen 2 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4 DM OV E : speci y C ommand TY PE : S w dC ommandD ec ARGS : [ S w dD e ice ] ; [ S w dV al ue ] CON T : S w dD e ice : 2 4 DM OV E : speci y P a ame e TY PE : S w dD e ice CON T : k i chen 3 5 S w dV alue : 2 6 6 4 DM OV E : speci y P a ame e TY PE : S w dV alue CON T : 25 SU BT Y P E : S w dP e cen ag e 3 7 7 5 3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 5 4. Inc ease he ki chen lamp no no dec ease i 2 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4 DM OV E : speci y C ommand TY PE : S w dC ommandD ec ARGS : [ S w dD e ice ] ; [ S w dV al ue ] CON T : S w dD e ice : 2 4 DM OV E : speci y P a ame e TY PE : S w dD e ice CON T : K i chen 3 5 E o C ommand : 2 6 6 6 6 6 6 6 6 4 DM OV E : speci y C ommand TY PE : S w dC ommandI nc ARGS : [ S w dD e ice ] ; [ S w dV al ue ] CON T : S w dD e ice : 2 4 DM OV E : speci y P a ame e TY PE : S w dD e ice CON T : K i chen 3 5 3 7 7 7 7 7 7 7 7 5 3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 5 5. Inc ease he lamp in he ki chen o le el 10 no no no he ligh in he dinning o om o 25 p e cen 2 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4 DM OV E : speci y C ommand TY PE : S w dC ommandI nc ARGS : [ S w dD e ice ] ; [ S w dV al ue ] CON T : S w dD e ice : 2 4 DM OV E : speci y P a ame e TY PE : S w dD e ice CON T : l i ing oom 3 5 S w dV alue : 2 6 6 4 DM OV E : speci y P a ame e TY PE : S w dV alue CON T : 25 SU BT Y P E : S w dP e cen ag e 3 7 7 5 E o S wdDe ice : 2 4 DM OV E : speci y P a ame e TY PE : S w dD e ice CON T : K i chen 3 5 E o S wdV alue : 2 6 6 4 DM OV E : speci y P a ame e TY PE : S w dV alue CON T : 10 SU BT Y P E : S w dLe el 3 7 7 5 3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 5 6. Se he ligh in he ki chen a le el 5 no no no inc ease i o 25 p e cen . 2 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4 DM OV E : speci y C ommand TY PE : S w dC ommandI nc ARGS : [ S w dD e ice ] ; [ S w dV al ue ] CON T : S w dV alue : 2 6 6 4 DM OV E : speci y P a ame e TY PE : S w dV alue CON T : 25 SU BT Y P E : S w dP e cen ag e 3 7 7 5 S w dD e ice : 2 4 DM OV E : speci y P a ame e ; TY PE : S w dD e ice CON T : k i chen 3 5 E o C ommand : 2 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4 DM OV E : speci y C ommand; TY PE : S w dC ommandS e ARGS : [ S w dD e ice; S w dV al ue ] CON T : S w dD e ice : 2 4 DM OV E : speci y P a ame e TY PE : S w dD e ice CON T : k i chen 3 5 S w dV alue : 2 6 6 4 DM OV E : speci y P a ame e TY PE : S w dV alue CON T : 5 SU BT Y P E : S w dLe el 3 7 7 5 3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 5 3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 5 5 Na u al Language Dialogues Finally, we p esen some dialogue examples which illus a e he main unc ions o he sys- em om a linguis ic p oin o iew: 5.1 Na u al Command Languages and Exp ec a ions Scena io: Two ligh s: one in he ki chen and he o he in he co ido . Bo h a e u nedo a he beginning Dialogue 1  Use : Tu n on.  Sys em: Wha de ice would you like o u n on?  U: The ki chen ligh .  S: The ligh in he ki chen has jus b een u ned on. In his example, he sys em de ec s ha in o de o u n on a de ice, he use mus supply he desc ip o o he ac ual de ice. This unc ionali y is implemen ed as a exp ec a ion o he dialogue manage ([Quesada, To e & Amo es 2000 ]). 5.2 Mul iple Commands Dialogue 2  U: Tu n o he ki chen and swi ch he co ido on.  S: The ligh in he ki chen has jus b een u ned o.  S: The ligh in he co ido has jus b een u ned on. Al hough he ou pu gene a ed by he sys- em mus be imp o ed in o de o ob ain jus one sen ence, his example illus a es one o he mos complex phenomena in dialogue managemen : he use asks he sys em o ex- ecu e se e al commands in one u e ance. 5.3 Exp ec a ions, p e ious commands and de aul in e p e a ion Dialogue 3  U: Tu n on he co ido .  S: The ligh in he co ido is al eady on.  U: So y, he ki chen ligh .  S: The ligh in he ki chen has jus b een u ned on. This example shows he in eg a ion o he dialogue manage wi h he knowledge man- age , which has de ec ed ha he co ido ligh was al eady on. Besides, he example il- lus a es an ambi ious dialogue s a egy: he de aul in e p e a ion o a pa ame e as pa o a p e ious command. In his scena io, he sp ecica ion o he pa ame e is analised as an exp ec a ion o he p e ious command. 5.4 Plug & Play Dialogue 4  U: Wha is he cu en s a e o he li ing o om ligh ?  S: The ligh in he li ing o om is no a ailable a he momen .  Using he HomeSe up Agen , he use in- s al ls, congu es and plugs a new ligh a he li ing oom.  U: Now, u n he li ing o om on.  S: The ligh in he li ing o om has jus b een u ned on. Taking in o accoun ha he 4 p e ious sub dialogues b elong in ac o jus one di- alogue ob ained om he eal sys em, i is wo h no ing he abili y o supp o plug & play capabili ies wi hou es a ing he dia- logue manage no he es o agen s. 6 Conclusion A dialogue sys em designed o supp o na u- al and exible dialogues in ended o con ol a ligh ning sys em in a home machine en i- onmen has b een p esen ed. The pap e has concen a ed on he linguis ic comp onen s o he agen {based a chi ec u e: he g amma and he seman ic ep esen a ion o complex sen ences. A se o dialogue examples illus- a ing he linguis ic capabili ies o he sys- em ha e b een p esen ed and desc ib ed. Re e ences [Amo es and Quesada 1997] Amo es, J.G. and J.F. Quesada. 1997. Epis eme. P ocesamien o del Lenguaje Na u al 21 . pp. 1-16. [Amo es & Quesada 2000] Amo es, J. G., Quesada, J. F. 2000. Dialogue Mo es in Na u al Command Languages. Deli e able 1.1. Si idus p o jec .