AmI-Blocks’08, European Conference on Ambient Intelligence 2008 manuscript No.(will be inserted by the editor)

CoRA - Interactive Communication with Smart Products

Sabine Janzen · Wolfgang Maass

Received: date / Accepted: date

Abstract The concept of interactive natural languagecommunication between users and products within aphysical environment holds the capability for a redesignand extension of communication at the point of sale.After considering specific requirements of ambient en-vironments, we present the model of a ConversationalRecommendation Agent (CoRA), a domain-specific web-based dialogue system based on semantic knowledgerepresentations (OWL-DL), which enables a product-centered communication between customers and smartproducts within ambient environments.

Keywords Smart product · dialogue system ·Question-Answer · Natural Language Processing ·semantic representations · OWL-DL

1 Introduction

By increasing informational complexity of consumerproducts, communication needs of customers regard-ing products at the point of sale rise intensely. Prod-ucts possess a lot of different information, such as userguides, service and support features or user generatedor professional reviews. Furthermore, consumers lookfor information about products that are alternatives orcomplements. For satisfying these consumers’s commu-nication needs, we investigate how Natural Languageservices can be embedded into products that supportconsumer-product communication. A Natural Language

Sabine Janzen · Wolfgang Maass

Furtwangen UniversityRobert-Gerwig-Platz 1

78120 Furtwangen

Tel.: +49 (0)7723 920-2924E-mail: {sabine.janzen, wolfgang.maass}@hs-furtwangen.de

based communication within a physical shopping envi-ronment enables an improved filtering and an intuitiverepresenting of product information [1]. The direct em-bedding of communication services into the product,resuming the idea of Ubiquitous Computing [2], formsthe basis for communication between users and physicalproducts in ambient environments. The linkage of phys-ical products with communicative functions requiresthe integration of value-added mobile services and dig-ital product descriptions [3] and represents the conceptof a smart product [4]. In order to realize Natural Lan-guage communication within physical environments, di-alogue systems and appropriate knowledge representa-tions are necessary that meet the specific requirementsof ambient environments. In this article, we introducethe concept of a web-based dialogue system for NaturalLanguage consumer-product communication in ambi-ent retail environments. Therefore, we adopt a designscience methodology [5]. Our results will be presented inthe sense of a ”proof by construction” [6]. In Section 2we will outline the main research issues and works in thecontext of this article. Following, Section 3 describes therequirements on a dialogue system in ambient environ-ments. After that, we introduce the architecture of ourdialogue system that meets the identified requirements(Section 4). In Section 5 we exemplify our approachof Natural Language Processing through an example.Finally we discuss our current results and future work(Section 6).

2 Related Work

Interactive communication with smart products con-tacts two fields of research - Ambient Intelligence andNatural Language Processing (NLP). Through the in-

2

tegration of concepts of the NLP into ambient envi-ronments, the two scientific fields as yet separated aremerged and exciting research issues are opened [1].Ambient Intelligence (AmI) is close to Ubiquitous Com-puting that describes the disappearing of informationtechnology within physical things of the daily use andthe opportunity to communicate with them [2]. In con-tinuation, Ambient Intelligence ”implies a seamless en-vironment of computing, advanced networking technol-ogy and specific interfaces” [1]. An AmI system gen-erally cope with large amounts of unstructured infor-mation from heterogeneous sources. These have to beselected and prepared to satisfy user information needs[1]. At this point, NLP technologies become relevant,because they enable interpretation, transformation andfiltering of information and therefore enhance the qual-ity of AmI systems. The more natural a communicationwith a physical entity the higher is the acceptance ofservices and information [7], [8], [9]. Natural Languageis a means for increased adaptation of communicationto the needs of consumers which is important for moreeffective marketing models [4].We assume that the appliance of Natural Languageinterfaces is attractive especially for creating humancomputer interfaces in tangible shopping environments.Such ambient environments hold specific characteris-tics, for instance, the prevalent embedding of an AmIsystem into mobile devices or their distribution amongmultiple components [1]. This means that Ambient In-telligence places high demands on a NLP approach forthe realization of an interaction in form of discourseswithin physical environments.Within the NLP, there are four dominant discoursetheories of Hobbs [10], Grosz & Sidner [11], Mann &Thompson [12] and McKeown [13]. What they havein common is the aspect that discourses are purpose-driven, i.e. communication is driven by user intentionsto achieve a communicative goal. Furthermore, thereexist two perspectives to consider the planning of dis-courses: functionalist and formalist perspective [14].Whereas the former focuses on communicative intentsof a speaker and their reflection within the discoursestructure, the latter capitalizes on formalisms of the lin-guistic structure of discourse segments. We assume thata combination of a functionalist and a viable formal-ist perspective constitutes an approach for interactivecommunication with smart products in ambient envi-ronments [14], [15].

3 Interactive Communication betweenCustomers and Smart Products

The concept of interactive communication with smartproducts encompasses the capability for a redesign andextension of communication interfaces between cust-omers, retailers and manufacturers at the point of sale.Imagine, a customer is situated within a fashion shop.She is interested in a brown jacket and wants to knowwether there is a blue variant of the product. In linewith an interactive communication with the smart prod-uct, she is able to ask: ”Is this jacket available in blue?”.The jacket could answer the question: ”Yes, you canfind it on the second floor.” In order to realize the em-bedding of natural language communication functionsinto physical products, dialogue systems and seman-tic product representations are required [3]. The formerpresents information in an optimal, goal-driven and in-tuitive way. We assume that natural language outputenlarges the competitive advantage of products withinseveral domains. Furthermore, the relationship betweenthe manufacturer and the consumer is deepened. ”Greatsalespeople understand that customer relationships arebuilt, maintained, and expanded through dialogues,which take place one after another over time.” [16]

3.1 Requirements on Natural Language Processing inPhysical Environments

A dialogue system within the scenario of a communica-tion between physical products and customers in phys-ical environments has to fulfill specific requirements:

– Processing of applied dialogues in shopping situa-tions - a dialogue system processes applied dialoguesin shopping situations.

– Two-dimensional resolution of intentions - In or-der to realize product-centered communication be-tween manufacturer and customer, dialogue systemsneed to resolve several, in part oppositional commu-nicative intentions. On the one hand, the dialoguesystem adopts a supportive function concerning theinformational needs of the customer. On the otherhand, it has to pursue specific economic interestsaccording to concerns of manufacturers or retailers.

– Differentiated output of answers - Subject to thecourse of a dialogue and changing economic inten-tions of manufacturers, a dialogue system is obligedto generate messages in diverse form at differenttimes.

– Efficient Natural Language Processing (NLP) - Toassure an efficient Natural Language Processing ona mobile device within physical environments, the

3

input opportunities of the customer in line with theNatural Language Understanding have to be chan-neled and restricted, for instance, through the pro-posal of constituent options. Thereby the usabil-ity of the dialogue system is improved. ”We believethat, for many purposes, a suitable natural languagesubset will be much preferred, on grounds of con-ciseness and ease of typing alone [...]. It is fair tosay that our objective is to so constrain natural lan-guage that it becomes a formal, but user-friendly,query language” [17].

4 Architecture of the ConversationalRecommendation Agent

According to the requirements of Natural LanguageProcessing in physical environments, we present a modelof a domain-specific dialogue system that constitutes acommunicative interface between customers and smartproducts. We propose the model of a ConversationalRecommendation Agent (CoRA) that consists of threecomponents: discourse intentions, the knowledge baseand the processing layer (cf. Fig. 1).

4.1 Question Schemata

After analyzing a text corpus of sales conversations andconsulting talks within the product domain of consumerelectronics and fashion that was collected previously, weidentified 19 question schemata that appear frequentlywithin sales dialogues [13].

1. Definition Pure Definition, e.g. What meansLycra R©? What means TMC?

2. Definition Functionality, e.g. How does Bluetoothwork?

3. Equation Difference, e.g. Whereby trouser A dif-fers from trouser B?

4. Equation Equity, e.g. Are the materials of the jack-ets equal? Are jacket A and jacket B equal?

5. Description Survey Average, e.g. What is the av-erage price of navigation systems?

6. Description Survey NumberOfProducts, e.g.Which navigation systems are available in the colorred? Which skirts are priced less than e 80?

7. Description Survey NumberOfProperties, e.g.Which colors are available? Which materials are avail-able?

8. Description Decision Comparison, e.g. Is there asmaller navigation sytem?

9. Description Decision Existence Property, e.g.Does product A possess Bluetooth?

10. Description Decision Existence Product, e.g. Isthis jacket available in blue? Is there a navigationsystem in red?

11. Description Information Price, e.g. How muchdoes product A cost?

12. Description Information PProperty, e.g. Whichsoftware is used? Which special features does theproduct have?

13. Description Information Usage, e.g. How can prod-uct a be used?

14. Description Information Person, e.g. Who is themanufacturer?

15. Description Information Modality, e.g. How tallis product A? How large is the memory? How longis the guarantee issued?

16. AdditiveProduct Survey Matching, e.g. Which prod-uct matches with product A? Which colors matchwith product A?

17. AdditiveProduct Bundle Price, e.g. How much doesthe bundle of product A and product B costs?

18. AdditiveProduct Decision Matching, e.g. Does prod-uct A matches with product B?

19. AdditiveProduct Discount, e.g. Is there a discounton the purchase of product A and product B?

Fig. 1 Model of the Conversational Recommendation Agent

(CoRA)

4.2 Discourse Intentions

The discourse intention component covers linguistic in-tentions, user-centered and vendor-centered intents forparticipating in a sales conversation. We derived thecommunicative intentions of the customer from the ques-tion schemata mentioned above. The results were trans-ferred to the taxonomy of communicative top-level-intentions of the customer [13] and integrated into thephases of the Simplified Consumer Choice Model (SCCM),which represents the course of a shopping situation [18](cf. Fig. 2).

Furthermore, vendor’s and retailer’s intentions arepre-defined. Along with derived communicative inten-

4

tions of a consumer, vendor-retailer-centered intentscompose the model of a two-dimensional goal resolu-tion, i.e. the processing of differentiated partly oppo-sitional intentions of several participants within a dis-course. Consequently, there are three sorts of discourseintentions in CoRA:

1. Linguistic intentions - Basically, CoRA pursues lin-guistic intentions that represent the conversationalmaxims of Grice, e.g, the dialogue system has theintent to generate syntactical correct messages with-out redundant information [19].

2. User-centered intentions - User-centered intents com-promise the communicative intentions of a customerduring the discourse.

3. Vendor-centered intentions - For instance, satisfac-tion of a consumer’s information needs and increaseof profitability and revenue.

In our configuration vendor-centered intentions aresubordinated to linguistic and user-centered intentions[20].

Fig. 2 Taxonomy of communicative goals in line with the SCCM[13], [18]

4.3 Knowledge Base

The knowledge base of the dialogue system exclusivelyprocesses web-based semantic knowledgerepresentations. CoRA uses a non-linguistic and a lin-guistic knowledge base. The former is represented by anontology-based semantic description of products, calledSmart Product Description Object (SPDO) [21], [3], [4].The linguistic knowledge component is based on theSemantic QA Structure Model, which is divided intoa domain independent and a domain dependent part[22]. Furthermore, a lexicon offers a taxonomy of parts-of-speech items. A specific interface concept referenceslemmas stored in the lexicon for realizing the integra-tion of linguistic material of the lexicon into the struc-tures of the NLP.

4.4 Processing Layer

On the Processing Layer all courses of the Natural Lan-guage Processing in CoRA are controlled and man-aged. The Interpretation Manager represents the Natu-ral Language Understanding, whereas the scope of theNatural Language Generation is constituted throughthe Generation Manager. The Behavioural Agent is thecentral controlling component that possesses a staticdiscourse model that maps the discourse intentions, andthe dynamic discourse management. The latter holdsthe process management, the discourse history and thefocus stack for storing the subjects of interest duringthe dialogue. The trichotomy enables a plain disjunc-tion of the components interpretation, generation andplanning. Based on this, the handling of more complexdomains and the portability of the dialogue system isimproved [23].

The approach of the NLP in CoRA covers the de-mands of both domains of the NLP in dialogues under-standing of questions and generation of answers. Nat-ural Language Understanding in CoRA uses a schema-based approach [13]. Each schema consists of a con-sumer’s communicative intention, a linguistic goal, con-straints, keywords (cue-words) [24] and a schematic com-position of a question.

The approach of the Natural Language Generationin CoRA is represented through a combination of textplanning technologies [25]; rhetorical relations [10], [12],[14]; text schemas [13] and focus of attention [11]. Thelinkage of goal-driven planning technologies with thelight-weight constitution of linguistic structures withinthe schemas enables the conjunction of the formalistand functionalist perspective of discourse planning [14].Furthermore, the application of planning technologiesallows an improved mapping of intentional structures[15] and provides a basis for realization of a two-dimensional goal resolution. Text plans are representedby plan operators [25], which are defined based on ananalysis of a text corpus and assigned to schemas forquestion understanding. A plan operator defines a textplan for achieving an effect that satisfies a conumer’scommunicative intention(s). It consists of a compul-sive part that provides information requested by a con-sumer, and optional parts, which conduce to the sat-isfaction of vendor-retailer-centered intentions. A fur-ther component, the library of rhetorical relations, in-terlinks the parts of a plan operator [10], [12], [14]. Themodel of the Natural Language Generation in CoRA iscompleted by the focus of attention that represents thesalient concepts of a discourse [13], [11].

5

Fig. 3 Course of a NLP cycle in CoRA

4.5 Course of a NLP cycle in CoRA

At the beginning of a NLP cycle in CoRA, a consumerapproaches a product in focus with a particular commu-nicative intention and selects the first segment of her de-sired question (cf. Fig. 3). In the course of incrementalselection of question segments by the consumer, CoRAanalyses the question and processes question schemasof the Semantic QA Structure Model for identifying aconsumers communicative goal. The question is deemedto be understood in CoRA, if the communicative inten-tion of the user and the appropriate question schemawas detected. In addition, CoRA derives a linguisticgoal from the detected question schema and translatesit into an effect that constitutes a key concept betweena domain of a question understanding and an answergeneration.

Meanwhile, the consumer can approve the completedquestion. Thus, the process of the answer generation isstarted. The Behavioural Agent checks the pool of planoperators based on the identified effect and selects theappropriate operator. Within the processing of the planoperator, the text plan of the answer is generated withregard to the vendor-retailer-centered intentions (calledMacroplanning [26]). Each part of the plan operator isexpanded; and resulting linguistic material is allocated(called Microplanning [26]). After checking the focusstack, CoRA decides which parts of an expanded planoperator are integrated into the final answer. Simulta-neously, the foci of processed fragments of a plan oper-ator are stored within the focus stack.

Finally, the generated answer is displayed and theBehavioural Agent saves the discourse act consistingof question, answer, effect and timestamp within thediscourse history of the SPDO for building up a productmemory.

4.6 Implementation of CoRA

On implementation level, CoRA constitutes a moduleof the Tip ’n Tell framework [4] and is implemented inform of a Java web service. The Tip ’n Tell client ona mobile device can communicate with the CoRA webservice based on SOAP1. The implementation of thedialogue system in form of a web service enables theplain enclosure of the NLP module from the remainingframework and the realization of a performant dialoguesystem on mobile devices within physical environments.

To realize the handling of the semantic representa-tions, more precisely the SPDO and the Semantic QAStructure Model, we use the Jena Semantic Web Frame-work2. This allows the use of SPARQL-based queriesfor retrieving product information from distributed se-mantic knowledge bases [27] and the processing of datawithin the application. Furthermore, each part of theknowledge base is formalized by semantic representa-tions (currently in OWL-DL3).

5 Exemplary NLP cycle on CoRA

The customer is situated in a physical shopping envi-ronment and poses a question to a product in focus. Af-ter identification of this product via RFID-based com-munication [3], the mobile client displays the dialogueinterface to the consumer. By means of the client, theconsumer is able to compose a question step-by-step.

In the course of question composition, the CoRAweb service returns potential question segments of thenext step according to consumer selections. At this point,

1 http://www.w3.org/TR/2007/REC-soap12-part1-20070427/

2 http://jena.sourceforge.net/3 http://www.w3.org/TR/owl-features/

6

Fig. 4 Exemplary question understanding in CoRA

the CoRA web service processes the information re-trieved from the knowledge base - the schematic lin-guistic structures as well as the non-linguistic productinformation of the SPDO (cf. Fig. 4). After comple-tion of a question by a consumer, CoRA identifies theassigned question schema; the linguistic goal and theresulting effect (cf. Fig. 4).

Then, CoRA starts to generate an answer to thequestion (in this example ”Wieviel kostet das Etuik-leid?”4) by means of the effect. Based on the effect, anadequate plan operator is retrieved from the Seman-tic QA Structure Model. The plan operator is used forcontrolling further requests of the linguistic and non-linguistic knowledge base (cf. Fig. 5).

Fig. 5 Exemplary generation of an answer in CoRA

Product-comprehensive information, for instance,product matching or product bundle results, are gen-erated via reasoning based on standardized web-basedrules that are applied to SPDO instantiations of prod-ucts [28], [29]. When linguistic and informational struc-tures of an answer are completed, the CoRA web ser-vice sends a response to the mobile client: ”Das Etuik-leid kostet e 89,95. Das Etuikleid passt zu den Balleri-nas Sunshine. Die Produkte kosten im Bndel nur e 119,95.”5 (cf. Fig. 5)

4 ”How much does the dress cost?”5 ”The dress costs e 89,95. The dress fits to the ballerinas ”sun-

shine”. The products as a bundle cost only e 119,95.”

6 Conclusion and Future Work

An interactive communication between customers andproducts in physical environments implies the integra-tion of technologies of the Natural Language Processing(NLP) into ambient environments. Such an embeddingof interactive communication functions into physicalproducts requires the assignment of dialogue systemsand semantic product representations [3] (cf. Section 1& 2).

In Section 3, we identified specific requirements forsuch a dialogue system within an ambient environment.Then, we presented with the Conversational Recom-mendation Agent (CoRA) a model of a dialogue systemthat realizes the communication between consumers andsmart products based on questions and answers [22] (cf.Section 4). It was shown how CoRA fulfills the require-ments of an ambient environment. It enables a distri-bution of the dialogue system among multiple compo-nents, embedding into mobile devices and processingof web-based semantic knowledge representations. Fur-thermore, CoRA affords two-dimensional goal resolu-tion according to communicative intentions of a con-sumer and economic intents of manufacturers and re-tailers. We exemplified our approach within an examplein Section 5.

Our current version of CoRA uses a rough set ofrestrictions within the expansion of plan operators. Toachieve more improved results, the integration of fur-ther specific restrictions is targeted. Furthermore, theprocessing of economic intentions of manufacturer andretailers mainly proceeds on the basis of the focus stackin a streamlined way. In addition to that, NLP-orientedSPARQL requests are rather complex and therefore af-fect flexibility of operating on the semantic data.

In our future work, we will focus on three issues: (1)A more differentiated processing of dynamically chang-ing intentions of different stakeholders within the gen-eration of answers in real-time, (2) representation ofcourses of conversations and those integration into thediscourse model of CoRA based on scripts [30] or tasks[31], and (3) improving the flexibility of answer genera-tion through integration of user models [32], [33], [34].

References

1. Gurevych, I., Muhlhauser, M.: Natural language processingfor ambient intelligence. Kunstliche Intelligenz/Special Issue:

Ambient Intelligence und Kunstliche Intelligenz (2), 10–16(2007)

2. Weiser, M.: The computer of the 21st century. ACM SIG-

MOBILE Mobile Computing and Communications Review

3(3), 3–11 (1999)3. Maass, W., Filler, A.: Towards an infrastructure for semanti-

cally annotated physical products. In: G. fur Informatik e.V.

7

(ed.) Informatik 2006 - Informatik fur Menschen: ConferenceProceedings (2006)

4. Maass, W., Janzen, S.: Dynamic product interfaces: A key el-ement for ambient shopping environments. In: M.L. Markus,

J.F. Hampe, J. Gricar, A. Pucihar, G. Lenart (eds.) 20th Bled

eConference - Conference Proceedings: ”eMergence: Merg-ing and Emerging Technologies, Processes, and Institutions”

(June 3- 6, 2007). Bled, Slovenia (2007)

5. Hevner, A.R., March, S.T., Park, J., Ram, S.: Design science

in information systems research. MIS Quarterly 28(1), 75–

105 (2004)

6. Nunamaker Jr., J.F., Chen, M., Purdin, T.D.M.: Systems de-

velopment in information systems research. Journal of Man-agement Information Systems 7(3), 89–106 (1990)

7. Bo, X., Benbasat, I.: E-commerce product recommendation

agents: Use, characteristics, and impact. MIS Quarterly31(1), 137–209 (2007)

8. Cho, Y.H., Kim, J.K., Kim, S.H.: A personalized recom-mender system based on web usage mining and decision tree

induction. Expert Systems with Applications 23(3), 329–342

(2002)

9. Haubl, G., Trifts, V.: Consumer decision making in online

shopping environments: The effects of interactive decisionaids. Marketing Science 19(1), 4–21 (2000)

10. Hobbs, J.R.: Why is Discourse Coherent?: Technical Note176. Menlo Park, California (1978)

11. Grosz, B.J., Sidner, C.L.: Attention, intentions, and the

structure of discourse. Computational Linguistics 12(3),175–204 (1986)

12. Mann, W.C., Thompson, S.A.: Assertions from discoursestructure. In: H.L.T. Conference (ed.) Proceedings of the

workshop on Strategic computing natural language, pp. 257–

270. Marina del Rey, California (1986)

13. McKeown, K.R.: Discourse strategies for generating natural-

language text. Artificial Intelligence (27) (1985)

14. Hovy, E.H.: Automated discourse generation using discourse

structure relations. Artificial Intelligence (63), 341–385

(1993)

15. Moore, J.D., Paris, C.L.: Planning text for advisory di-

alogues: Capturing intentional and rhetorical information.Computational Linguistics pp. 651–694 (1994)

16. Kalyanam, K., Zweben, M.: The perfect message at the per-fect moment. Harvard Business Review pp. 112–120 (2005)

17. Warren, D.H.D., Pereira, F.C.N.: An efficient easily adapt-

able system for interpreting natural language queries. Com-putational Linguistics (8), 110–122 (1982)

18. Maass, W., Kowatsch, T.: Adoption of dynamic product in-formation: An empirical investigation of supporting purchase

decisions on product bundles. In: Proceedings of the 16th Eu-

ropean Conference on Information Systems (ECIS). Galway,Ireland (2008)

19. Grice, H.P.: Logic and conversation. In: P. Cole, J. Morgan(eds.) Syntax and Semantics Volume 3. Speech Acts., vol. 3,

pp. 41–58. Academic Press, New York (1975)

20. Beveridge, M., Milward, D.: Combining task descriptions and

ontological knowlegde for adaptive dialogue. In: V. Ma-

tousek, P. Mautner (eds.) Proceedings International Confer-ence on Text Speech and Dialogue (TSD), pp. 341–348. CeskeBudejovice (2003)

21. Janzen, S., Maass, W.: Smart product description object(spdo). In: Poster Proceedings of the 5th International

Conference on Formal Ontology in Information Systems(FOIS2008). Saarbrucken, Germany (2008)

22. Janzen, S.: Product-centered communication in natural lan-

guage within shopping situations: master’s thesis. Ph.D. the-sis, Furtwangen University, Furtwangen (2008)

23. Allen, J., Ferguson, G., Stent, Amanda: An architecture formore realistic conversational systems. In: J.C. Lester (ed.)

IUI 2001 - 2001 International Conference on Intelligent User

Interfaces, Santa Fe, New Mexico, USA, January 14 - 17,2001, pp. 1–8. ACM Press, New York, NY (2001)

24. Cohen, R.: A computational theory of the function of clue

words in argument understanding. In: A. for Computa-tional Linguistics (ACL) (ed.) Proceedings of the 22nd an-

nual meeting on Association for Computational Linguistics,

pp. 251–258. Stanford, California (1984)25. Mann, W.C.: Discourse structures for text generation. In:

A. for Computational Linguistics (ACL) (ed.) Proceedings ofthe 22nd annual meeting on Association for Computational

Linguistics, pp. 367–375. Stanford, California (1984)

26. Bateman, J., Zock, M.: Natural language generation. In:R. Mitkov (ed.) The Oxford Handbook of Computational

Linguistics. Oxford University Press, Oxford, New York

(2003)27. Prud’Hommeaux, E., Seaborne, A.: SPARQL Query Lan-

guage for RDF (2007)

28. Maass, W., Filler, A.: Tip ’n tell: Product-centered mobilereasoning support for tangible shopping. In: Proceedings of

MSWFB 2007: Making Semantics Work For Business, part of

1st European Semantic Technology Conference. Wien (2007)29. Maass, W., Filler, A., Janzen, S.: Reasoning on smart prod-

ucts in consumer good domains. In: Workshop AmI Blocks at

the European Conference on Ambient Intelligence (AmI-07):Proceedings. Darmstadt (2007)

30. Schank, R.C.: Using knowledge to understand. In: A. forComputational Linguistics (ACL) (ed.) Proceedings ”Work-

shop on Theoretical issues in natural language processing”:

SESSION: Representing Knowledge, pp. 117–121. Cambridge(1975)

31. Core, M.G., Allen, J.: Coding dialogues with the damsl anno-

tation scheme. In: A. Magazine (ed.) AAAI Symposium: Pro-ceedings ”Communicative Action in Humans and Machines”.

Cambridge (1997)

32. Paris, C.L.: Description strategies for naive and expert users.In: A. for Computational Linguistics (ACL) (ed.) Proceed-

ings of the 23rd annual meeting on Association for Compu-

tational Linguistics, pp. 238–245. Chicago (1985)33. Paris, C.L.: Tailoring object descriptions to a user’s level of

expertise. Computational Linguistics 14(3), 64–78 (1988)34. Sieg, A., Mobasher, B., Burke, R.: Web search personaliza-

tion with ontological user profiles. In: M.J. Silva, A.A.F.

Laender, R. Baeza-Yates, D.L. McGuinness, B. Olstad, Ø.H.Olsen, A.O. Falcao (eds.) Proceedings of the 2007 ACM Con-

ference on Information and Knowledge Management: Lisboa,

Portugal, Nov. 6 - 10, 2007, pp. 525–534. ACM, New York,NY (2007)