Development of Generic Resources for Language Technology

Software developed within both these areas are to be included in the first code resource library. At present we have the following running research prototypes that we wish to make more flexible and robust and place in the library:

• JavaChart: A chart parser for interpretation of natural language. Currently we are rewriting and extending a lisp-based chart parser prototype called Flexchart to Java. The new system JavaChart will be more compliant with other new network-based systems than its predecessor and have a more flexible interface, easily integrated with a user interface as well as with a dialogue engine.
• DG: A compiler written in Java for so-called dialogue grammars that handles high-level control of the dialogue interface of a dialogue system. Future work include, for instance, extending the user support and making DG more clearly separated from its main use today as a help-tool for the dialogue manager module in the MALIN framework.
• DialogueHead: A Java-module for high-level control concerning dialogue aspects of a visually speaking head. Currently the module supervises the dialogue behavior of an animated head (a system called Orator) and synthesis-module from Telia Research (a system called Telia Text-Till-Tal, or T4).
• MALIN: an object-oriented framework in Java that supports construction of complete speech-based dialogue systems. The MALIN framework has evolved over a number of research projects and today consists of the following components:
  • Interpretation Manager: a Java-module that supervises the interpretation and transformation of incoming speech into a syntactic interpretation-structure. The software needs more work in various ways for the module to become a robust application-independent system.
  • Dialogue Manager (DM): The kernel module of the MALIN framework kernel is the Dialogue Manager which interprets utterances in context and directs the dialogue. The Dialogue Manager is now in a third version, written in Java. Major refinements involves separating generic parts from more domain specific and also techniques to make it more robust, to handle for instance unknown situations better and more uniformly.
  • Domain Knowledge Manager: This is a new module developed for dialogue interfaces, aimed at handling various types of domain knowledge utilised in a dialogue interface. The Domain Knowledge Manager is based on the Open Agent Architecture and is, thus, suitable for network based software development.
• Linköping Word Aligner(LWA): A system, that can be used for processing bilingual texts, the one a translation of the other. It links (i.e. align) words, phrases and tokens, that are the translations of each other in the two texts, aligning more than 50% of the texts.
• Frasse: A module for identifying phrases and multi-word terms from a raw text. Currently, it can handle Swedish and English texts. The module can for instance be used for extracting multi-word units for bi-lingual word alignment and terminology management.

We are also currently involved in a dialogue systems project (SCIN) together with Telia Research AB where we are utilising a synthetic talking head as output modality. In this project we have further developed the software that controls the talking head (Orator) and the speech synthesis module (T4) by introduction of a new level of control in the new module DialogueHead. The new module also makes the system network-based, i.e. other modules in the Dialogue System utilise the synthesis resources remotely as services running via the Internet.