Research map / Large language models / Language, safety and interpretability
Filling Conversation Ellipsis for Better Social Dialog Understanding
Xiyuan Zhang, Chengxi Li, Dian Yu, Samuel Davidson, Zhou Yu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Combine predictions from original and automatically completed utterances | Dialog act prediction with BERT · Gunrock | F1 (%): 79.65 -> 80.90 | Improved |
| Use automatically completed utterances only instead of original utterances | Dialog act prediction with BERT · Gunrock | F1 (%): 79.65 -> 78.04 | Hurt |
| Add a copy mechanism to Seq2Seq (ablation) | Utterance completion · Gunrock | BLEU (%): 42.36 -> 71.85 | Improved |
| Combine original and completed utterance SRL predictions | Semantic role labeling with stacked Bi-LSTMs · Gunrock | F1 (%): 88.39 -> 90.87 | Improved |
| Use completed utterances only instead of original utterances | Semantic role labeling with stacked Bi-LSTMs · Gunrock | F1 (%): 88.39 -> 87.50 | Hurt |
| Add expert knowledge to logit addition during selection (ablation) | Dialog act prediction with BERT · Gunrock | F1 (%): 80.85 -> 80.90 | Improved |
| Use maximum rather than addition to combine logits (ablation) | Dialog act prediction with BERT · Gunrock | F1 (%): 79.85 vs 80.85 | Hurt |
| Combine hidden states by addition rather than using logit addition (ablation) | Dialog act prediction with BERT · Gunrock | F1 (%): 80.08 vs 80.85 | Hurt |
| Combine hidden states by concatenation rather than using logit addition (ablation) | Dialog act prediction with BERT · Gunrock | F1 (%): 79.65 vs 80.85 | Hurt |
| Combine hidden states by maximum rather than using logit addition (ablation) | Dialog act prediction with BERT · Gunrock | F1 (%): 79.63 vs 80.85 | Hurt |
| Use probability-based rather than rule-based expert knowledge for SRL selection (ablation) | Hybrid-EL-CMP · Gunrock | F1 (%): 90.62 -> 90.87 | Improved |
| Train two models on original utterances instead of using a single original-utterance model (ablation) | Dialog act prediction with BERT · Gunrock | F1 (%): 79.65 -> 79.70 | Improved |
Closest papers by meaning
- DialogLM: Pre-trained Model for Long Dialogue Understanding and Summarization (AAAI 2022 · similarity 0.86)
- DialectLLM: A Dialect-Aware Dialog[ue] Generation Framework Beyond Standard American English (NeurIPS 2026 · similarity 0.84)
- MIntRec2.0: A Large-scale Benchmark Dataset for Multimodal Intent Recognition and Out-of-scope Detection in Conversations (ICLR 2024 · similarity 0.84)
- SayNext-Bench: Why Do LLMs Struggle with Next-Utterance Anticipation? (NeurIPS 2026 · similarity 0.84)
- Aligning Spoken Dialogue Models from User Interactions (ICML 2025 · similarity 0.84)
- SD-Eval: A Benchmark Dataset for Spoken Dialogue Understanding Beyond Words (NeurIPS 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
