《Towards AI-Complete Question Answering: A Set of Prerequisite Toy Tasks》

arxiv
Facebook AI
2015.02

主要工作

  • 构建数据集 bAbI-tasks 使用 QA 的方式评估阅读理解能力

dataset:babi
source:bAbI-tasks

Tasks

  • answer 都是单词或者单词列表
  • 无噪声,人类水准可以达到 100% 准确率
  • 每个任务有 1000 个 question 用于训练,1000 个 question 用于测试
# Task Class name
1 Basic factoid QA with single supporting fact WhereIsActor
2 Factoid QA with two supporting facts WhereIsObject
3 Factoid QA with three supporting facts WhereWasObject
4 Two argument relations: subject vs. object IsDir
5 Three argument relations WhoWhatGave
6 Yes/No questions IsActorThere
7 Counting Counting
8 Lists/Sets Listing
9 Simple Negation Negation
10 Indefinite Knowledge Indefinite
11 Basic coreference BasicCoreference
12 Conjunction Conjunction
13 Compound coreference CompoundCoreference
14 Time manipulation Time
15 Basic deduction Deduction
16 Basic induction Induction
17 Positional reasoning PositionalReasoning
18 Reasoning about size Size
19 Path finding PathFinding
20 Reasoning about agent’s motivation Motivations
  • Single Supporting Fact (1):用 1 个 supporting fact 就可以回答的 question
  • Two or Three Supporting Facts (2, 3):用 2 个或者 3 个 supporting facts 可以回答的 question
  • Two or Three Argument Relations (4, 5):回答 question 的关键是能分清楚 subjects 和 objects
  • Yes/No Questions (6):回答 Yes/No 类型的问题的能力
  • Counting and Lists/Sets (7, 8):计数、回答列表物件的能力
  • Simple Negation and Indefinite Knowledge (9, 10):对否定句以及可能性的理解能力
  • Basic Coreference, Conjunctions and Compound Coreference (11, 12, 13):对代指、多 subject 的理解能力
  • Time Reasoning (14):时间推理能力
  • Basic Deduction and Induction (15, 16):推理归纳能力
  • Positional and Size Reasoning (17, 18):空间推理
  • Path Finding (19):根据地理位置进行路径推理
  • Agent’s Motivations (20):主体动机推理

提供了 3 种版本

  • 正常英文
  • Hindi 文
  • 顺序打乱的英文

Example Alt text Alt text

训练

各种 baseline Alt text

  • AM:adaptive memory
  • NG:N-grams
  • NL:nonlinear matching function

读后感

就是提出了 babi-tasks 数据集,20 个 tasks 构造的思路很有意思,涵盖了语言理解能力的各种方向