AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

MOSS: End-to-End Dialog System Framework with Modular Supervision

Weixin Liang, Youzhi Tian, Chengcai Chen, Zhou Yu

AAAI 2020 · Large language models › Language, safety and interpretability · approach: Faithfulness of LLM explanations · 44 citations (OpenAlex)

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 changedWhereMeasuredEffect
add supervision from all four dialog modulesMOSS · CamRest676 testMat: 0.927 -> 0.951Improved
add supervision from dialog policy learning (ablation)MOSS · CamRest676 testBLEU: 0.251 -> 0.259Improved
add supervision from natural language understanding (ablation)MOSS · CamRest676 testMat: 0.932 -> 0.951Improved
add 40% raw dialogs to 60% fully annotated training data (ablation)MOSS-all · CamRest676 testBLEU: 0.202 -> 0.221Improved
reduce training data to 60% (ablation)MOSS-all · CamRest676 testBLEU: 0.259 -> 0.202Hurt
reduce training data to 60% (ablation)MOSS-all · CamRest676 testMat: 0.951 -> 0.947No clear effect
add supervision from dialog policy learning (ablation)MOSS · LaptopNetwork testSucc.accImproved
add supervision from natural language understanding (ablation)MOSS · LaptopNetwork testSucc.acc: 0.50 -> 0.61Improved
use modular supervision from all four modules rather than TSCP supervisionMOSS-all · LaptopNetwork testSucc.acc: 0.18 -> 0.61Improved
reduce training data to 40% (ablation)MOSS-all · LaptopNetwork testSucc.accImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy