Task Introduction

Expertise Style Transfer (EST) is a new task of text style transfer between expert language and layman language, which is to tackle the problem of discrepancies between an expert's advice and a layman's understanding of it. We contribute a manually annotated dataset, namely MSD, in the medical domain to promote research into this task.

Dataset Description

We provide a training set with non-parallel sentences in expert and layman styles, and a parallel test set for validation. Attached to each sentence is a UMLS concept list, which is to encourage knowledge-aware style transfer.

Data Format

Each line of train.txt and test.txt is a dictionary object in JSON with the following keys:

    text: the sentence text
    label: 0 for expert and 1 for layman style
    concepts: a list of concepts in the format {
        "range": the start and end word position of the concept
        "term": the term of the concept
        "cui": the CUI codes defined by UMLS
    }

In test.txt, the odd line 2n-1 and even line 2n are parallel.

Version

This is the 2nd version of dataset. We ask a local doctor to review and revise the test set according to the comments of reviewers. 18 pairs of parallel sentences are filtered.