D · i· a· l· o· g· R· E

A Dialogue-Based Relation Extraction Dataset

What is DialogRE

DialogRE is the first human-annotated dialogue-based relation extraction dataset, containing 1,788 dialogues originating from the complete transcripts of a famous American television situation comedy Friends. We annotate all occurrences of 36 possible relation types that exist between an argument pair in a dialogue. DialogRE is available in English and Chinese.

Sample Instances

S1:        Hey Pheebs.
S2:        Hey!
S1:        Any sign of your brother?
S2:        No, but he's always late.
S1:        I thought you only met him once?
S2:        Yeah, I did. I think it sounds y'know big sistery, y'know, `Frank's always late.'
S1:        Well relax, he'll be here.
Argument pair Trigger Relation type
(Frank, S2) brother per:siblings
(S2, Frank) brother per:siblings
(S2. Pheebs) none per:alternate_names
(S1, Pheebs) none unanswerable

Relation Schema

ID Subject Relation Type Object Inverse Relation
1 PER per:positive_impression NAME
2 PER per:negative_impression NAME
3 PER per:acquaintance NAME per:acquaintance
4 PER per:alumni NAME per:alumni
5 PER per:boss NAME per:subordinate
6 PER per:subordinate NAME per:boss
7 PER per:client NAME
8 PER per:dates NAME per:dates
9 PER per:friends NAME per:friends
10 PER per:girl/boyfriend NAME per:girl/boyfriend
11 PER per:neighbor NAME per:neighbor
12 PER per:roommate NAME per:roommate
13 PER per:children NAME per:parents
14 PER per:other_family NAME per:other_family
15 PER per:parents NAME per:children
16 PER per:siblings NAME per:siblings
17 PER per:spouse NAME per:spouse
18 PER per:place_of_residence NAME gpe:residents_of_place
19 PER per:place_of_birth NAME gpe:births_in_place
20 PER per:visited_place NAME gpe:visitors_of_place
21 PER per:origin NAME
22 PER per:employee_or_member_of NAME org:employees_or_members
23 PER per:schools_attended NAME org:students
24 PER per:works NAME
25 PER per:age VALUE
26 PER per:date_of_birth VALUE
27 PER per:major STRING
28 PER per:place_of_work STRING
29 PER per:title STRING
30 PER per:alternate_names NAME/STRING
31 PER per:pet NAME/STRING
32 GPE gpe:residents_of_place NAME per:place_of_residence
33 GPE gpe:births_in_place NAME per:place_of_birth
34 GPE gpe:visitors_of_place NAME per:visited_place
35 ORG org:employees_or_members NAME per:employee_or_member_of
36 ORG org:students NAME per:schools_attended
37 NAME unanswerable NAME/STRING/VALUE

Baseline Results

V1: The original English version (used in the paper)

V2-EN: The updated English version with a few annotation errors fixed (updated on August 2020)

V2-CN: The Chinese version (updated on August 2020)

V1 V2-EN V2-CN
Dev Test Dev Test Dev Test
F1 (σ) F1c (σ) F1 (σ) F1c (σ) F1 (σ) F1c (σ) F1 (σ) F1c (σ) F1 (σ) F1c (σ) F1 (σ) F1c (σ)
BERT 60.6 (1.2) 55.4 (0.9) 58.5 (2.0) 53.2 (1.6) 59.4 (0.7) 54.7 (0.8) 57.9 (1.0) 53.1 (0.7) 63.7 (0.9) 59.5 (0.7) 63.2 (0.9) 58.4 (0.8)
BERTS 63.0 (1.5) 57.3 (1.2) 61.2 (0.9) 55.4 (0.9) 62.2 (1.3) 57.0 (1.0) 59.5 (2.1) 54.2 (1.4) 65.5 (0.9) 61.0 (0.4) 63.5 (0.9) 58.7 (0.9)


Please contact dialogre@dataset.org.

Sibling Datasets

Check DREAM (https://dataset.org/dream) or C3 (https://dataset.org/c3) if you are interested in dialogue-based multiple-choice machine reading comprehension for English or Chinese.