ICASSP2023 MUG Challenge Track3: Topic Title Generation (TTG)

Who this competition fits

This suits NLP practitioners who can fine-tune text-generation models for Mandarin meeting transcripts and handle long, conversational input. It is open to academia and industry, but Alibaba employees and Zhejiang University faculty/students cannot participate.

From registration to a first submission

1. 01

   Mandarin NLP

2. 02

   abstractive title generation

3. 03

   meeting-transcript processing

4. 04

   Transformer fine-tuning

5. 05

   ROUGE evaluation

6. 06

   ModelScope

Before you commit: The central challenge is producing concise, informative titles from long spoken-meeting segments that contain disfluencies, redundancies, grammar errors, coreference, and fragmented sentences. Participants must also work within a constrained-track setup: only specified public pre-trained models, the provided corpus, and permitted public corpora may be used.