Gespeichert in:
| Hauptverfasser: | , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | https://arxiv.org/abs/2412.07515 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866916516788699136 |
|---|---|
| author | Steindl, Sebastian Schäfer, Ulrich Ludwig, Bernd |
| author_facet | Steindl, Sebastian Schäfer, Ulrich Ludwig, Bernd |
| contents | Large-scale Wizard-Of-Oz dialogue datasets have enabled the training of deep learning-based dialogue systems. While they are successful as benchmark datasets, they lack certain types of utterances, which would make them more realistic. In this work, we investigate the creation of synthetic communication errors in an automatic pipeline. Based on linguistic theory, we propose and follow a simple error taxonomy. We focus on three types of miscommunications that could happen in real-world dialogues but are underrepresented in the benchmark dataset: misunderstandings, non-understandings and vaguely related questions. Our two-step approach uses a state-of-the-art Large Language Model (LLM) to first create the error and secondly the repairing utterance. We perform Language Model-based evaluation to ensure the quality of the generated utterances. We apply the method to the MultiWOZ dataset and evaluate it both qualitatively and empirically as well as with human judges. Our results indicate that current LLMs can aid in adding post-hoc miscommunications to benchmark datasets as a form of data augmentation. We publish the resulting dataset, in which nearly 1900 dialogues have been modified, as CoPrUS-MultiWOZ to facilitate future work on dialogue systems. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2412_07515 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | CoPrUS: Consistency Preserving Utterance Synthesis towards more realistic benchmark dialogues Steindl, Sebastian Schäfer, Ulrich Ludwig, Bernd Computation and Language Large-scale Wizard-Of-Oz dialogue datasets have enabled the training of deep learning-based dialogue systems. While they are successful as benchmark datasets, they lack certain types of utterances, which would make them more realistic. In this work, we investigate the creation of synthetic communication errors in an automatic pipeline. Based on linguistic theory, we propose and follow a simple error taxonomy. We focus on three types of miscommunications that could happen in real-world dialogues but are underrepresented in the benchmark dataset: misunderstandings, non-understandings and vaguely related questions. Our two-step approach uses a state-of-the-art Large Language Model (LLM) to first create the error and secondly the repairing utterance. We perform Language Model-based evaluation to ensure the quality of the generated utterances. We apply the method to the MultiWOZ dataset and evaluate it both qualitatively and empirically as well as with human judges. Our results indicate that current LLMs can aid in adding post-hoc miscommunications to benchmark datasets as a form of data augmentation. We publish the resulting dataset, in which nearly 1900 dialogues have been modified, as CoPrUS-MultiWOZ to facilitate future work on dialogue systems. |
| title | CoPrUS: Consistency Preserving Utterance Synthesis towards more realistic benchmark dialogues |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2412.07515 |