|
Email:
|
|
| Description: |
TNO-ICT developed 'Novalink'; a system that searches news articles to identify named entities and tries to find relations between these entities. This information is displayed in a graph and users can interact with this graph so it shows the information they want.
Unfortunately, names have a lot of variances: usage of full names, abbreviations, titles etc. Currently the system does not identify these different names as referring to the same entity resulting in ambiguous or incomplete networks. For example 'C. Versloot' and 'Corne Versloot' refer to the same person but are spelled differently and are treated by the system as two different persons. This problem also has user-interface aspects since users must be able to fix or change mistakes made by the system.
To summarize, the task is to improve the existing novalink system using Named Entity disambiguation. This improvement haws two aspects:
- Designing and implementing NE disambiguation into the Novalink system.
- Investigating ways to adapt the current user interface to the new functionality and implement the changes.
|
|
|
| Description: |
The basic goal of the internship is to help improve the naturalness and social influence of the rapport system. This involves extending its behavioral repertoire to include appropriately triggered facial expressions and to identify and correct any peculiarities of the current system. Peculiarities arise from improper feature detection, improper behavior mapping, and improper behavior display. These can be better identified by analyzing data on experimental runs and comparing it with data of the human listener, of human coders and with previously validated feature detection algorithms (for example, one could run some other (offline) disfluency detection algorithm and compare its results with the dysfluencies identified by Rapport.
|
|
|
| Description: |
|
This study presents a new annotation scheme designed to annotate listeners non-verbal behavior. Four studies concerning listener and also speaker behavior form the basis for this schema. I tried to combine these different insights into one schema. The most important parts of this schema are dimensions; they represent different functions of listener behavior. Every dimension has a set of labels to annotate listener behavior. I used the schema to annotate listeners in various settings. The setup of the schema and the results of the tests are described in this paper. | |
|
