Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SOLOIST: Building Task Bots at Scale with Transfer Learning and Machine Teaching

Baolin Peng, Chunyuan Li, Jinchao Li, Shahin Shayandeh, Lars Liden, Jianfeng Gao

TL;DR

The paper introduces SOLOIST, a Transformer-based autoregressive framework that unifies the entire task-oriented dialog pipeline into a single model and leverages task-grounded pre-training on heterogeneous corpora. It then enables rapid domain adaptation to new tasks via few-shot fine-tuning and machine teaching, significantly reducing labeling costs. Empirical results on CamRest676 and MultiWOZ show state-of-the-art end-to-end performance and strong few-shot capabilities, with substantial gains from machine teaching. The work demonstrates a scalable path for building many task bots at scale by transferring grounding and dialog-management skills learned during pre-training to new domains.

Abstract

We present a new method SOLOIST that uses transfer learning and machine teaching to build task bots at scale. We parameterize classical modular task-oriented dialog systems using a Transformer-based auto-regressive language model, which subsumes different dialog modules into a single neural model. We pre-train, on heterogeneous dialog corpora, a task-grounded response generation model, which can generate dialog responses grounded in user goals and real-world knowledge for task completion. The pre-trained model can be efficiently adapted to accomplish new tasks with a handful of task-specific dialogs via machine teaching, where training samples are generated by human teachers interacting with the system. Experiments show that (i) SOLOIST creates new state-of-the-art on well-studied task-oriented dialog benchmarks, including CamRest676 and MultiWOZ; (ii) in the few-shot fine-tuning settings, SOLOIST significantly outperforms existing methods, and (iii) the use of machine teaching substantially reduces the labeling cost of fine-tuning. The pre-trained models and codes are available at https://aka.ms/soloist.

SOLOIST: Building Task Bots at Scale with Transfer Learning and Machine Teaching

TL;DR

The paper introduces SOLOIST, a Transformer-based autoregressive framework that unifies the entire task-oriented dialog pipeline into a single model and leverages task-grounded pre-training on heterogeneous corpora. It then enables rapid domain adaptation to new tasks via few-shot fine-tuning and machine teaching, significantly reducing labeling costs. Empirical results on CamRest676 and MultiWOZ show state-of-the-art end-to-end performance and strong few-shot capabilities, with substantial gains from machine teaching. The work demonstrates a scalable path for building many task bots at scale by transferring grounding and dialog-management skills learned during pre-training to new domains.

Abstract

We present a new method SOLOIST that uses transfer learning and machine teaching to build task bots at scale. We parameterize classical modular task-oriented dialog systems using a Transformer-based auto-regressive language model, which subsumes different dialog modules into a single neural model. We pre-train, on heterogeneous dialog corpora, a task-grounded response generation model, which can generate dialog responses grounded in user goals and real-world knowledge for task completion. The pre-trained model can be efficiently adapted to accomplish new tasks with a handful of task-specific dialogs via machine teaching, where training samples are generated by human teachers interacting with the system. Experiments show that (i) SOLOIST creates new state-of-the-art on well-studied task-oriented dialog benchmarks, including CamRest676 and MultiWOZ; (ii) in the few-shot fine-tuning settings, SOLOIST significantly outperforms existing methods, and (iii) the use of machine teaching substantially reduces the labeling cost of fine-tuning. The pre-trained models and codes are available at https://aka.ms/soloist.

Paper Structure

This paper contains 33 sections, 6 equations, 4 figures, 13 tables.

Table of Contents

  1. Introduction
  2. Soloist
  3. An Auto-Regressive Model for Dialog
  4. Task-Grounded Pre-Training
  5. Task 1: Belief Prediction.
  6. Task 2: Grounded Response Generation.
  7. Task 2: Grounded Response Generation.
  8. Task 3: Contrastive Objective.
  9. Full Pre-Training Objective.
  10. Implementation Details.
  11. Fine-Tuning and Machine Teaching
  12. Implementation Details.
  13. Experiments
  14. Experimental Setup
  15. Dialog Datasets for Fine-Tuning.
  16. ...and 18 more sections

Figures (4)

  • Figure 1: Illustration of a traditional modular task-oriented dialog system, an example for the model input, and the proposed model. The Soloist solution utilizes a single neural auto-regressive model in (c) to paramterize the sequential dialog pipeline in (a), with input sequence represented in (b). Different from GPT-2, the Soloist model learns to ground response generation in user goals and database/knowledge.
  • Figure 2: Illustration of the machine teaching process using conversion learner. The human-bot conversion log in (a) can be edited via correcting its belief state in (b), and selecting/inserting a more appropriate response in (c).
  • Figure 3: Machine teaching performance of different iterations in Restaurant domain. Machine teaching with CL achieves near 1.5X efficiency gain (i.e., the 1st iteration used 15 dialogs while the 3rd iteration has 25 dialogs) and boosts performance by 10 points compared with that without teaching.
  • Figure 4: An interactive example.