Dialogue Commonsense Graph in Japanese

We proposed a commonsense knowledge graph (CSKG) grounded on dialogues. This repository contains the graph built in Japanese.

Data Format

We defined the eight relations based on categories, time series, and target persons, referring to CICERO.

Relation	Category	Time	Person
Cause	Event	Before	Speaker
Prerequisite	Event	Before	Listener
Subsequent event	Event	After	Speaker
Effect	Event	After	Listener
Motivation	Mental state	Before	Speaker
Guess	Mental state	Before	Listener
Reflection	Mental state	After	Speaker
Reaction	Mental state	After	Listener

Each utterance is annotated with inferences of the eight relations, and each dialogue is stored as a list of the utterances. An example of the annotated utterances is shown below:

{
    "status_id": 1557164592392183808,
    "cause": [
        "暑い暑い"
    ],
    "prerequisite": [
        "確かに全然外に出てない"
    ],
    "subsequent_event": [
        "宿題が終わらない"
    ],
    "effect": [
        "じゃあ家で宿題すればちょうどいいじゃん",
        "外に出ないでエアコンが効いた部屋で過ごす"
    ],
    "motivation": [
        "暑すぎてキツイ",
        "夏が終わらないと困る"
    ],
    "guess": [
        "暑くない",
        "夏がすぐ終わると宿題が終わらなくて困る"
    ],
    "reflection": [
        "暑さには何も勝てないよ"
    ],
    "reaction": [
        "そうかもね",
        "その通りだ"
    ]
}

Note that the utterance texts are replaced by their Twitter IDs.

We annotated 352 dialogues, corresponding to 2,121 utterances, with the inferences. The statistics of the graph is shown below:

Relation	# triples	Avg # infs / utt
Cause	3,060	1.44
Prerequisite	2,728	1.29
Subsequent event	3,001	1.41
Effect	3,276	1.54
Motivation	3,567	1.68
Guess	1,679	0.79
Reflection	1,591	0.75
Reaction	3,564	1.68

Reference

@InProceedings{ide_nlp2023_dcsg,
    author =    "井手竜也 and 榮田亮真 and 河原大輔 and 山崎天 and 李聖哲 and 新里顕大 and 佐藤敏紀",
    title =     "対話に基づく常識知識グラフの構築と対話応答生成に対する適用",
    booktitle = "言語処理学会第29回年次大会",
    year =      "2023",
    url =       "https://www.anlp.jp/proceedings/annual_meeting/2023/pdf_dir/H1-4.pdf"
    note=       "in Japanese"
}

License

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.