Kontrolovatelné generování jazyka

• Thesis title in Czech: Kontrolovatelné generování jazyka
• Thesis title in English: Controllable natural language generation
• Academic year of topic announcement: 2019/2020
• Thesis type: dissertation
• Thesis language: čeština
• Department: Institute of Formal and Applied Linguistics (32-UFAL)
• Supervisor: Mgr. et Mgr. Ondřej Dušek, Ph.D.
• Author: hidden - assigned and confirmed by the Study Dept.
• Date of registration: 17.07.2019
• Date of assignment: 17.07.2019
• Confirmed by Study dept. on: 28.11.2019

Guidelines

Current neural NLG methods are quite hard to control, the generation typically proceeds in a language model style. The outputs often do not contain all content required by the input specification, or contain additional, irrelevant (hallucinated) content. This in effect excludes neural NLG from deployment in production systems. This project will address the reliability problem and explore novel, controllable NLG architectures. A possible extension can focus on the style of generated texts, such as generating outputs with specific personality traits.

References

TBD