This is a mixed-initiative Dialogue System designed to guide the user when accessing the web. It has been designed to access different types of web services and information. The main goal in the system design have been its adaptability to the different types of web contents and different types of users and different languages. The Dialogue System consists of four main modules: the Natural Language Analyzer, the Dialogue Manager, the Language Generator and the Services’ Access. The Dialogue System is focused in the development of a platform for accessing Web Services. The communication channel can be either textual or by voice, using natural language. The aim of our Dialogue System is to facilitate the integration of new applications into the platform. Then, the development of application specific resources should be easy and able to do by non skilleds.

Textual or spontaneous speech access to Web services (currently, Spanish on-line access to cultural agenda and large objects collection services).

The Dialogue System is developed in J2EE, nevertheless some modules additionally uses other technologies. The Natural Language Analyzer has been developed in Prolog, more suitable for this task. All the resources are written in XML files. Some of them use standards, such as VoiceXML, SRGSXML and OWL. Standards are used when possible. Additionally, we have developed specific XML Schemas for writing the rest of the resources, for example the communication between the Prolog subsytem and the Natural Language Analyzer module, the linguistic resources for the Language Generator, and the Dialogue Plans.

Basic System: The basic Dialogue System uses Textual Web Chat Client, runnin on Tomcat Server. The rest of the System runs JVM plus Swi-Prolog Server. More specific technical requirements depends on the Application integrated in the System. Our current Services are also Java Applications using XML files containing the database. Voice-added System: The Voice subsystem is telephone based, and needs specific hardware and software. The HOPS System uses Loquendo(TM) VoiceXML platform plus a telephone card. Our current system doesn’t allow voice access to the platform, nevertheless the linguistic resources are developed taking into account both channels.

For dialogue management we developed an independent module, the Dialogue Manager (DM). It enables mixed-initiative conversations and it is mode, language and domain independent. It follows the issue-based dialogue management model described in (Larsson, 2002), which describes dialogues in terms of issues that are raised and that have to be solved. The Dialogue Manager uses plans for guiding the conversation through the Operations of the Service. We have developed general recipes for creating plans and resources for new applications. Those recipes are classified into three types: SEARCH, QUERY and TRANSACTION.

The Dialogue System incorporates a Natural Language Generator (NLG) component to generate the system’s messages. In order to obtain the most appropriate system prompts for a specific service, the generator component uses a syntactic-semantic taxonomy which relates the specific domain concepts to the linguistic structures needed for their expression. The system messages are not generated at run-time but when adapting the DS to a new web service. Several of the messages generated contain variables that would be instantiated at run-time, considering the dialogue context. When the DS is adapted to a new sevice, the service task parameters have to be classified according to the syntactic-semantic taxonomy and linked to the corresponding lexical entries. Then, the system prompts are generated automatically. However, the resulting sentences can be supervised and selected manually.At run-time the NLG would complete these messages considering dialogue context. Additionally, when system intervention includes several communication acts, the messages expressing this acts are combined dynamically (following a fixed order): greet moves first, then confirmation moves, then answer moves, then asking moves, and finally quit moves.

Textual inputs are processed by a left-corner parser performing syntactic and semantic analysis in parallel has been adapted for a practical dialogue system for multiple applications. The parser uses domain-restricted grammars and lexicons obtained from ontologies representing application specific knowledge. It also uses the knowledge of the dialogue context to select the grammar rules related to the dialogue focus.

The current working System uses XML based data sources for the simulating the Services Access. The ongoing work includes the use of OWL and RDFS for enabling the online access to Web Services as well as the Standarization of the resources.

An ontology-based representation is used for sharing data among modules, as well as for specific modules resources. This representation facilitates the integration of new applications into the system.

One of the main goals of this project has been the improvement of existing commercial Dialogue Systems by combining practical results from various research areas: Speech, Natural Language, Dialogue Management and Ontologies.

Most recent works on spoken DSs in the commercial area are based on exploiting the VoiceXML language. But, it also presents several limitations, for example, it presents some dialogue management drawbacks, as the limited support to the user-initiative interactions.

Our work has been focused on the study of how Dialogue Systems, and in particular VoiceXML platforms, could be improved by using language, dialogue and ontology techniques. We also have work on techniques for adapting the interaction to the user expertise. We studied how it improves the communication process. Finally, we are currently working on improved plans and recipes representation. The aim of our current direction is to enable more sophisticated dialogue management, without losing the simplicity in the specific application resources.

The Dialogue System was developed in the context of the HOPS project (IST-2002-507967).

• Meritxell González. Dialogue Management for Multilingual communication through different channels. DEA and PT, LSI Department, UPC. Barcelona, July 2007.
• Marta Gatius and Meritxell González. Discourse Management in Voice Systems for Accessing Web Services. Workshop on the Semantics and Pragmatics of Dialogue. DECALOG 2007. Rovereto, May 2007.
• Marta Gatius, Meritxell González and Elisabet Comelles. An Information State-Based Dialogue Manager for Making Voice Web Smarter. 16th International Worl Wide Web Conference. WWW2007. Banff, May 2007.
• Marta Gatius, Meritxell González, Eli Comelles and Leonardo Lesmo. Natural Language Processing and Dialogue Management Development. European Project HOPS (IST-2002-507967). Deliverable D4.3. April 2007.
• Marta Gatius and Meritxell González. A multilingual Dialogue System for Accessing the WEB. 3rd International Conference on Web Information Systems and Technologies. WEBIST 2007. Barcelona, March 2007.
• Pablo Hernández, Jordi Sánchez, Ángel López, Sheyla Militello, Marta Gatius, Meritxell González, Eli Comelles, Leonardo Lesmo, Xavier Noria, Robert Salla, Carlos de la Morena, Jose Antonio Fernández, Alberto Deiro. HOPS Architecture Specifications. European Project HOPS (IST-2002-507967). Deliverable D4.1. May 2006.
• Marta Gatius and Meritxell González. Integrating Semantic Web and Language Technologies to Improve the Online Public Administrations Services. 15th International Worl Wide Web Conference. WWW2006. May 23-26, 2006, Edinburgh, Scotland. ACM 1-59593-323-9/06/0005.
• Marta Gatius and Meritxell González. Using Application-Specific Ontologies to Improve Performance in a Bottom-up Parser. Workshop Knowledge and Reasoning for Language Processing, KRAQ’06 . 11th Conference of the European Chapter of the Association for the Computacional Linguistics. EACL’06. Trento, Italy, April 2006. Association for the Computacional Linguistics, ISBN, 1-932432-59-0, pp. 12-19.
• Marta Gatius and Meritxell González. Obtaining Linguistic Resources for Dialogue Systems from Application Specifications and Domain Ontologies. 10th International Conference on Speech and Computer, SPECOM 2005. University of Patras, October 2005.
• Meritxell González, Marta Gatius. Un sistema de diálogo multicanal para acceder a la información y servicios de las administraciones públicas. XXI Congreso de la SEPLN, I Congreso Español de Informática, CEDI. Granada , Septiembre 2005.
• Marta Gatius, Meritxell González. The project HOPS: Enabling an Intelligent Natural Language Based Hub for the Deployment of Advanced Semantically Enriched Multi-channel Mass-scale Online Public Services. XXI Congreso de la SEPLN, I Congreso Español de Informática. Granada , Septiembre 2005
• Marta Gatius, Meritxell González. Un sistema de diálogo multilingüe dirigido por la semántica. Revista de la SEPLN, Vol.34. Junio 2005.
• Marta Gatius, Meritxell González, Leonardo Lesmo, Pietro Torasso. Natural Language Processing Technologies. European Project HOPS (IST-2002-507967). Deliverable D3.2. February 2005.
• Marta Gatius, Meritxell González. Using Ontologies for Improving the Communication Process in a Dialogue System. Proceedings of the Sixth International Workshop on Computational Semantics, IWCS-6. Tilburg, 2005.
• Marta Gatius, Meritxell González. Utilización de ontologías en el desarrollo de sistemas de diálogo. III Jornadas en Tecnología del Habla. Valencia, Noviembre 2004.
• Marta Gatius, Meritxell González. Ontology-driven VoiceXML Dialogues Generation. Berliner XML-Tage 2004. Humboldt University Berlin, October 2004.