Are Large Language Models General-Purpose Solvers for Dialogue Breakdown Detection? An Empirical Investigation
Bibliographic record
Abstract
This study addresses the challenge of dialogue breakdown—characterized as incoherence, irrelevance, or any disruption that significantly hampers the flow of the conversation. The impact of dialogue breakdowns has become critical with the adoption of large language models in various industries for companies with dialogue-based systems, such as Salesforce, Amazon, and Microsoft. Leveraging the Dialogue Breakdown Detection Challenge Dataset, we investigate the performance of generalist large language models, including ChatGPT, GPT-4, and Mistral-Medium, in identifying instances of dialogue breakdown without domain-specific fine-tuning. Through a series of experiments employing zero-shot and few-shot prompting techniques combined with chain-of-thought reasoning, this research sets a new benchmark in the field. Our findings reveal that GPT-4 outperforms both specialized models previously considered state-of-the-art and other generalist models in detecting dialogue breakdowns by over a 3% margin, achieving an accuracy of 82.0%. To our knowledge, this study is the first to demonstrate the enhanced capability of generalist large language models in this domain. Our experiments found that when detecting dialogue breakdowns, larger models like GPT-4 are less sensitive to how they are prompted. In contrast, smaller models like ChatGPT and Mistral-Medium can improve their performance by using prompting techniques that combine few-shot learning with the chain-of-thought method. This work proposes future research directions, including enhanced error analysis and developing an Ensemble RAG approach for improved generalization in dialogue breakdown detection.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".