2018 Fiscal Year Final Research Report
Building acceptability rating database that enables authentication of linguists' intuitions
| Project/Area Number |
16K13223
|
|---|
| Research Category |
Grant-in-Aid for Challenging Exploratory Research
|
| Allocation Type | Multi-year Fund |
|---|
| Research Field |
Linguistics
|
|---|
| Research Institution | Kyorin University |
|---|
Principal Investigator |
Kuroda Kow 杏林大学, 医学部, 准教授 (30425764)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
阿部 慶賀 岐阜聖徳学園大学, 教育学部, 准教授 (70467041)
|
|---|
| Research Collaborator |
ASAO Yoshihiko
KANAMARU Toshiyuki
KOBAYASHI Yuuichiro
TAGAWA Takumi
TSUCHIYA Tomoyuki
YOKONO Hikaru
|
|---|
| Project Period (FY) |
2016-04-01 – 2019-03-31
|
| Keywords | acceptability judgment / multivariate analysis / computational modeling / large scale data / evidence-based research |
|---|
| Outline of Final Research Achievements |
In 2016, we ran a survey of acceptability judgments by professional linguists based on samples taken from textbooks of Japanese linguistics. The results were reported at 23rd Annual Meeting of NLP. In 2017, we developed a script for semi-automated generation of potentially deviant sentences, called Japanese sentence mutators (JSM). We made this public. We then ran a pilot study targeting roughly 300 raters which consists of university students. Used stimuli were 200 and each of the participants rated randomized group of 20 sentences. The results were reported in 24th Annual Meeting of NLP. In 2018, we carried out our main study in two phases. In phase 1, roughly 200 university students participated. In phase 2, roughly 1700 people in general public participated. In both phases, 300 sentences are randomly separated into 10 groups, and each participant rated one of them, consisting of 30 sentences. The results were reported at 25th Annual Meeting of NLP.
|
| Free Research Field |
認知科学，言語学，自然言語処理
|
| Academic Significance and Societal Importance of the Research Achievements |
言語学の課題の一つは，普通に使われうる文と使われない文の境界を確定する事である．前者を容認度可能な文と，容認不可能 (か困難) な文と呼ぶ．困った事に，2種類の表現の区別は理論から中立に行なえない．それにより言語学者が自説に都合の良い結果しか見ないという結果が生じる．これは「確証バイアス」の名で知られる．
これを回避するには，理論的立場に影響されない文を十分な数用意し，非専門家がそれらに与える容認度判断を基本反応として収集しておき，理論的利害が関わる事例の容認度評定の際に参照するしかない．日本語容認度評定データ (ARDJ) を，世界で初めてそのような目的をもつ中立データとして構築した．
|
