|
Email:
|
|
| Description: |
The automatic generation of objects descriptions in natural language (such as Dutch or English) is an important research topic in the field of natural language generation (NLG). Given a number of objects in a visual scene, the task of the NLG system is to create a description that allows the human user to identify the intended target object. The task is usually carried out in two steps:
(1) Attribute selection: select a set of properties that uniquely characterize the target object (and none of the other objects in the scene)
(2) Realisation: create a noun phrase in natural language that expresses these properties, for example "the dark-haired man with the glasses".
So far, most research has focused on step (1), but of course step (2) is equally important. HMI student Ivo Brugman created a realisation component for referring expressions in English, and coupled it to an existing attribute selection algorithm. The combination performed very well in the last TUNA Evaluation Challenge for referring expressions.
The goal of the current research assignment is to create another realisation component, this time for Dutch referring expressions, and evaluate it. A corpus of Dutch referring expressions, produced by human speakers, is already available. This corpus was collected at the University of Tilburg. The assignment will involve, among other things, analysis of the corpus to inform the implementation of the new realiser.
NB: this Capita Selecta (or Research Topic) could be done as preparation for a graduation project involving a more complex domain; see this Final Msc project.
References
Ivo Brugman's Capita Selecta research paper |
|
|
| Description: |
|
The project is a preparation research for an eventual Master assignment (final project) on generating story reports in natural language for the Virtual Storyteller. The project has two main goals: to formulate recommendations for the internal story model of the new Virtual Storyteller, and to come up with ideas on how to improve language generation for this goal. | |
|
