An Information State Approach to Aggregation in Multilingual Generation

Abstract:

The topic of the present dissertation is an information state approach to aggregation in multilingual generation (MLG). The aim is to define a formalism based on the information state approach and to apply it to aggregation in an MLG system. The information state refers to the current information available to the system, i.e. the text is generated from the information state. This information can change by means of a rule engine that operates on the information according to a control strategy defined by the system. In this system the process known as aggregation acts as the rule engine. Aggregation is the process by which more complex syntactic structures are built from simpler ones. It can also be defined as removing redundancies in text, which makes the text more coherent. The result is a program that generates text with respect to a specific domain in four different languages, viz. English, Swedish, German, and Spanish. The main challenges concern language independence, language accommodation, and addition of new languages. It is shown that language specific rules are needed, that the text is better accommodated to each language, and that it is relatively easy to add new languages.

Bio:

Anna Ljungberg is currently a Visiting Scholar at the Centre of Language Technology at Macquarie University, Sydney, where she is working with Rolf Schwitter on controlled languages. Previously she worked at the University of Skovde, Sweden. She has an MSc in Artificial Intelligence from Edinburgh University,Scotland and an MA in Linguistics from Lund University, Sweden. Her main research interests are language generation in multilingual environments and machine translation.

Enquiries: sals@mri.mq.edu.au