Research map / Large language models / Retrieval and knowledge
A Controllable Model of Grounded Response Generation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Add dialogue grounding to the dialogue-context input (ablation) | GPT-2 · Grounded Reddit, user-controlled | NIST: 0.90 -> 0.98 | Helped |
| Replace the CMR model with GPT-2 | Grounded response generation with X+G · Grounded Reddit, user-controlled | NIST: 0.34 -> 0.98 | Helped |
| Add control phrases to the dialogue context (ablation) | GPT-2 · Grounded Reddit, user-controlled | NIST: 0.90 -> 1.67 | Helped |
| Use control-relevant grounding instead of the full grounding (ablation) | GPT-2 · Grounded Reddit, user-controlled | NIST: 0.98 -> 1.34 | Helped |
| Add explicit control phrases alongside control-relevant grounding (ablation) | GPT-2 · Grounded Reddit, user-controlled | NIST: 1.34 -> 1.77 | Helped |
| Apply control phrases in the model input during training rather than only at decoding with GBS (ablation) | GPT-2 grounded generation · Grounded Reddit, user-controlled | NIST: 1.60 -> 1.77 | Helped |
| Add inductive attention to the model using control phrases and control-relevant grounding (ablation) | GPT-2 · Grounded Reddit, user-controlled | NIST: 1.77 -> 1.80 | Helped |
| Add retrieval-based control-relevant grounding to the context-only input (ablation) | GPT-2 · Grounded Reddit, predicted-control multi-reference | NIST: 1.42 -> 1.61 | Helped |
| Add retrieval-based predicted control phrases to control-relevant grounding (ablation) | GPT2IA · Grounded Reddit, predicted-control multi-reference | NIST: 1.61 -> 1.67 | Helped |
| Use BERT QA rather than retrieval to predict control phrases (ablation) | GPT2IA with predicted controls and control-relevant grounding · Grounded Reddit, predicted-control multi-reference | NIST: 1.67 -> 1.67 | No clear effect |
| Use GPT2IA instead of GPT-2 without inductive attention (ablation) | X+C+G_C · Grounded Reddit, user-controlled | BLEU: 3.22% -> 3.26% | Helped |
| Use GPT2IA rather than GPT-2 for responses with control and control-relevant grounding (ablation) | X+C+G_C · Grounded Reddit, human evaluation | grounding consistency preference: 28.1% vs. 27.6% | No clear effect |
Closest papers by meaning
- Zero-Resource Knowledge-Grounded Dialogue Generation (NeurIPS 2020 · similarity 0.90)
- DialogBERT: Discourse-Aware Response Generation via Learning to Recover and Rank Utterances (AAAI 2021 · similarity 0.89)
- Low-Resource Knowledge-Grounded Dialogue Generation (ICLR 2020 · similarity 0.89)
- A Knowledge-Grounded Neural Conversation Model (AAAI 2018 · similarity 0.89)
- KPT: Keyword-Guided Pre-training for Grounded Dialog Generation (AAAI 2023 · similarity 0.88)
- Improving Open-Domain Dialogue Response Generation with Multi-Source Multilingual Commonsense Knowledge (AAAI 2024 · similarity 0.86)
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.