Datasets:
The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.
PierreColombo/miam(複製)
データの内容
英語・フランス語・ドイツ語・イタリア語・スペイン語の話し言葉の対話に、対話行為(dialogue act)のラベルを付けた多言語のベンチマーク(MIAM)のデータ。1 行は一つの発話(Utterance)で、話者(Speaker)、対話行為(Dialogue_Act)、Dialogue_ID などを持つ。Dihana、iLISTEN、LORIA、HCRC MapTask、VERBMOBIL(vm2)の五つのコーパスを集めたもので、自発的な発話、台本のあるシナリオ、共同のタスク遂行を含み、config 名で一つずつ選べる。
データの概要
| 項目 | 値 |
|---|---|
| 行数 | 103,901(5 構成・15 分割の合計) |
| 列数 | 6〜7(構成による) |
| 容量 | 32.66 KB(ファイル 3 本) |
| ファイルの形式 | py 1 |
| 言語 | de, en, es, fr, it |
| 公開 | public |
構成と分割
| 構成(config) | 分割(split)ごとの行数 | 列数 |
|---|---|---|
dihana |
train 19,063、validation 2,123、test 2,361 |
7 |
ilisten |
train 1,986、validation 230、test 971 |
6 |
loria |
train 8,465、validation 942、test 1,047 |
7 |
maptask |
train 25,382、validation 5,221、test 5,335 |
7 |
vm2 |
train 25,060、validation 2,860、test 2,855 |
6 |
列(構成 dihana。構成ごとに列が違う)
| 列 | 型 |
|---|---|
Speaker |
string |
Utterance |
string |
Dialogue_Act |
string |
Dialogue_ID |
string |
File_ID |
string |
Label |
struct{class_label: struct{names: struct{…}}} |
Idx |
int32 |
行数と列は元のカードの YAML(dataset_info)の記載。データセットビューアでは読めなかった。全データセットの一覧は HYUKI5/dataset-catalog にある。
複製について
PierreColombo/miam をリビジョン f138dd789c15a3b25eee181a4dc4d330741c4c39 の時点でそのまま複製したもの。中身は変えていない。判断モデル Laya(convaiinnovations/laya)の微調整に使う候補として、調査メモ(Laya 学習用データセット調査)に載せたデータを集めている。
出典
- 元のデータセット: https://huggingface.co/datasets/PierreColombo/miam
- 複製したリビジョン:
f138dd789c15a3b25eee181a4dc4d330741c4c39(2024-01-18 11:09:00+00:00) - 元のライセンス表記:
['cc-by-sa-4.0']
論文
アノテーションの方法
元のカードにアノテーションの節はあるが、中身は空欄(More Information Needed)だった。
微調整に使うときの注意
- 調査メモの台帳での区分: B(継承条件・gated・モデル由来の条件・上流の規約などが付く)。区分は記載の文言で機械的に付けたもので、法的な判定ではない。学習に入れる前に元のカードとライセンスの原文で確かめる。
- Laya の入力は 2,048 token まで(ModernBERT-large / mmBERT)。長い本文は先頭から黙って切られるので、切れる行は事前に除くか、要約・切り出しをしてから入れる。
- 分割は記事・文書単位で行い、同じ本文から作った別の問いを学習側と較正側に分けない。温度と閾値は学習に使っていない分割で別々に当てる。
調査メモの記述
| メモ | 節 | ライセンス(記載) | 確認 | 注記 |
|---|---|---|---|---|
jev_sources_tasksource_bekko.csv |
CSV(tasksource-jev-typed-decisions) | cc-by-sa-4.0 / license_use=commercial | 確認済 | license_use=commercial / [Jev 元データ source] source_annotation_converted(deterministic recast; no model r |
元のカード(日本語訳)
以下は元のデータセットのカードの本文を日本語に訳したもの(機械翻訳)。訳に疑問があるときは、末尾の原文を確かめる。
データセットカード: MIAM
目次
- データセットの説明
- データセットの概要
- 対応するタスクとリーダーボード
- 言語
- データセットの構造
- データインスタンス
- データフィールド
- データの分割
- データセットの作成
- キュレーションの根拠
- ソースデータ
- アノテーション
- 個人情報とセンシティブな情報
- データの利用上の考慮事項
- データセットの社会的影響
- バイアスについての議論
- その他の既知の制限
- 追加情報
- データセットのキュレーター
- ライセンス情報
- 引用情報
- 貢献
データセットの説明
- ホームページ: [N/A]
- リポジトリ: [N/A]
- 論文: [N/A]
- リーダーボード: [N/A]
- 連絡先: [N/A]
データセットの概要
Multilingual dIalogAct benchMark は、話し言葉を対象に設計された自然言語理解システムの学習、評価、分析のための資源の集合である。データセットは英語、フランス語、ドイツ語、イタリア語、スペイン語である。自発的な発話、台本のあるシナリオ、共同でのタスク遂行を含む様々なドメインを扱う。すべてのデータセットには対話行為ラベルが含まれている。
対応するタスクとリーダーボード
[More Information Needed]
言語
英語、フランス語、ドイツ語、イタリア語、スペイン語。
データセットの構造
データインスタンス
Dihana コーパス
dihana の config でのデータセットの例は次のとおり:
{
'Speaker': 'U',
'Utterance': 'Hola , quería obtener el horario para ir a Valencia',
'Dialogue_Act': 9, # 'Pregunta' ('Request')
'Dialogue_ID': '0',
'File_ID': 'B209_BA5c3',
}
iLISTEN コーパス
ilisten の config でのデータセットの例は次のとおり:
{
'Speaker': 'T_11_U11',
'Utterance': 'ok, grazie per le informazioni',
'Dialogue_Act': 6, # 'KIND-ATTITUDE_SMALL-TALK'
'Dialogue_ID': '0',
}
LORIA コーパス
loria の config でのデータセットの例は次のとおり:
{
'Speaker': 'Samir',
'Utterance': 'Merci de votre visite, bonne chance, et à la prochaine !',
'Dialogue_Act': 21, # 'quit'
'Dialogue_ID': '5',
'File_ID': 'Dial_20111128_113927',
}
HCRC MapTask コーパス
maptask の config でのデータセットの例は次のとおり:
{
'Speaker': 'f',
'Utterance': 'is it underneath the rope bridge or to the left',
'Dialogue_Act': 6, # 'query_w'
'Dialogue_ID': '0',
'File_ID': 'q4ec1',
}
VERBMOBIL
vm2 の config でのデータセットの例は次のとおり:
{
'Utterance': 'ja was sind viereinhalb Stunden Bahngerüttel gegen siebzig Minuten Turbulenzen im Flugzeug',
'Utterance': 'Utterance',
'Dialogue_Act': 'Dialogue_Act', # 'INFORM'
'Speaker': 'A',
'Dialogue_ID': '66',
}
データフィールド
dihana の config では、フィールドは次のとおり:
Speaker: 話者の識別子(文字列)。Utterance: 発話(文字列)。Dialogue_Act: 発話の対話行為ラベル。'Afirmacion' (0) [Feedback_positive]、'Apertura' (1) [Opening]、'Cierre' (2) [Closing]、'Confirmacion' (3) [Acknowledge]、'Espera' (4) [Hold]、'Indefinida' (5) [Undefined]、'Negacion' (6) [Feedback_negative]、'No_entendido' (7) [Request_clarify]、'Nueva_consulta' (8) [New_request]、'Pregunta' (9) [Request]、'Respuesta' (10) [Reply] のいずれか。Dialogue_ID: 対話の識別子(文字列)。File_ID: 元ファイルの識別子(文字列)。
ilisten の config では、フィールドは次のとおり:
Speaker: 話者の識別子(文字列)。Utterance: 発話(文字列)。Dialogue_Act: 発話の対話行為ラベル。'AGREE' (0)、'ANSWER' (1)、'CLOSING' (2)、'ENCOURAGE-SORRY' (3)、'GENERIC-ANSWER' (4)、'INFO-REQUEST' (5)、'KIND-ATTITUDE_SMALL-TALK' (6)、'OFFER-GIVE-INFO' (7)、'OPENING' (8)、'PERSUASION-SUGGEST' (9)、'QUESTION' (10)、'REJECT' (11)、'SOLICITATION-REQ_CLARIFICATION' (12)、'STATEMENT' (13)、'TALK-ABOUT-SELF' (14) のいずれか。Dialogue_ID: 対話の識別子(文字列)。
loria の config では、フィールドは次のとおり:
Speaker: 話者の識別子(文字列)。Utterance: 発話(文字列)。Dialogue_Act: 発話の対話行為ラベル。'ack' (0)、'ask' (1)、'find_mold' (2)、'find_plans' (3)、'first_step' (4)、'greet' (5)、'help' (6)、'inform' (7)、'inform_engine' (8)、'inform_job' (9)、'inform_material_space' (10)、'informer_conditioner' (11)、'informer_decoration' (12)、'informer_elcomps' (13)、'informer_end_manufacturing' (14)、'kindAtt' (15)、'manufacturing_reqs' (16)、'next_step' (17)、'no' (18)、'other' (19)、'quality_control' (20)、'quit' (21)、'reqRep' (22)、'security_policies' (23)、'staff_enterprise' (24)、'staff_job' (25)、'studies_enterprise' (26)、'studies_job' (27)、'todo_failure' (28)、'todo_irreparable' (29)、'yes' (30) のいずれか。Dialogue_ID: 対話の識別子(文字列)。File_ID: 元ファイルの識別子(文字列)。
maptask の config では、フィールドは次のとおり:
Speaker: 話者の識別子(文字列)。Utterance: 発話(文字列)。Dialogue_Act: 発話の対話行為ラベル。'acknowledge' (0)、'align' (1)、'check' (2)、'clarify' (3)、'explain' (4)、'instruct' (5)、'query_w' (6)、'query_yn' (7)、'ready' (8)、'reply_n' (9)、'reply_w' (10)、'reply_y' (11) のいずれか。Dialogue_ID: 対話の識別子(文字列)。File_ID: 元ファイルの識別子(文字列)。
vm2 の config では、フィールドは次のとおり:
Utterance: 発話(文字列)。Dialogue_Act: 発話の対話行為ラベル。'ACCEPT' (0)、'BACKCHANNEL' (1)、'BYE' (2)、'CLARIFY' (3)、'CLOSE' (4)、'COMMIT' (5)、'CONFIRM' (6)、'DEFER' (7)、'DELIBERATE' (8)、'DEVIATE_SCENARIO' (9)、'EXCLUDE' (10)、'EXPLAINED_REJECT' (11)、'FEEDBACK' (12)、'FEEDBACK_NEGATIVE' (13)、'FEEDBACK_POSITIVE' (14)、'GIVE_REASON' (15)、'GREET' (16)、'INFORM' (17)、'INIT' (18)、'INTRODUCE' (19)、'NOT_CLASSIFIABLE' (20)、'OFFER' (21)、'POLITENESS_FORMULA' (22)、'REJECT' (23)、'REQUEST' (24)、'REQUEST_CLARIFY' (25)、'REQUEST_COMMENT' (26)、'REQUEST_COMMIT' (27)、'REQUEST_SUGGEST' (28)、'SUGGEST' (29)、'THANK' (30) のいずれか。Speaker: 話者(文字列)。Dialogue_ID: 対話の識別子(文字列)。
データの分割
| データセット名 | Train | Valid | Test |
|---|---|---|---|
| dihana | 19063 | 2123 | 2361 |
| ilisten | 1986 | 230 | 971 |
| loria | 8465 | 942 | 1047 |
| maptask | 25382 | 5221 | 5335 |
| vm2 | 25060 | 2860 | 2855 |
データセットの作成
キュレーションの根拠
[More Information Needed]
ソースデータ
初期のデータ収集と正規化
[More Information Needed]
ソース言語の生成者は誰か?
[More Information Needed]
アノテーション
アノテーションの過程
[More Information Needed]
アノテーターは誰か?
[More Information Needed]
個人情報とセンシティブな情報
[More Information Needed]
データの利用上の考慮事項
データセットの社会的影響
[More Information Needed]
バイアスについての議論
[More Information Needed]
その他の既知の制限
[More Information Needed]
追加情報
データセットのキュレーター
匿名。
ライセンス情報
本著作物は Creative Commons Attribution-NonCommercial-ShareAlike 4.0 Unported License のもとで提供されている。
引用情報
@inproceedings{colombo-etal-2021-code,
title = "Code-switched inspired losses for spoken dialog representations",
author = "Colombo, Pierre and
Chapuis, Emile and
Labeau, Matthieu and
Clavel, Chlo{\'e}",
booktitle = "Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing",
month = nov,
year = "2021",
address = "Online and Punta Cana, Dominican Republic",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2021.emnlp-main.656",
doi = "10.18653/v1/2021.emnlp-main.656",
pages = "8320--8337",
abstract = "Spoken dialogue systems need to be able to handle both multiple languages and multilinguality inside a conversation (\textit{e.g} in case of code-switching). In this work, we introduce new pretraining losses tailored to learn generic multilingual spoken dialogue representations. The goal of these losses is to expose the model to code-switched language. In order to scale up training, we automatically build a pretraining corpus composed of multilingual conversations in five different languages (French, Italian, English, German and Spanish) from OpenSubtitles, a huge multilingual corpus composed of 24.3G tokens. We test the generic representations on MIAM, a new benchmark composed of five dialogue act corpora on the same aforementioned languages as well as on two novel multilingual tasks (\textit{i.e} multilingual mask utterance retrieval and multilingual inconsistency identification). Our experiments show that our new losses achieve a better performance in both monolingual and multilingual settings.",
}
貢献
このデータセットを追加してくれた @eusip と @PierreColombo に感謝する。
原文(英語)
Dataset Card for MIAM
Dataset Summary
Multilingual dIalogAct benchMark is a collection of resources for training, evaluating, and analyzing natural language understanding systems specifically designed for spoken language. Datasets are in English, French, German, Italian and Spanish. They cover a variety of domains including spontaneous speech, scripted scenarios, and joint task completion. All datasets contain dialogue act labels.
Supported Tasks and Leaderboards
[More Information Needed]
Languages
English, French, German, Italian, Spanish.
Dataset Structure
Data Instances
Dihana Corpus
For the dihana configuration one example from the dataset is:
{
'Speaker': 'U',
'Utterance': 'Hola , quería obtener el horario para ir a Valencia',
'Dialogue_Act': 9, # 'Pregunta' ('Request')
'Dialogue_ID': '0',
'File_ID': 'B209_BA5c3',
}
iLISTEN Corpus
For the ilisten configuration one example from the dataset is:
{
'Speaker': 'T_11_U11',
'Utterance': 'ok, grazie per le informazioni',
'Dialogue_Act': 6, # 'KIND-ATTITUDE_SMALL-TALK'
'Dialogue_ID': '0',
}
LORIA Corpus
For the loria configuration one example from the dataset is:
{
'Speaker': 'Samir',
'Utterance': 'Merci de votre visite, bonne chance, et à la prochaine !',
'Dialogue_Act': 21, # 'quit'
'Dialogue_ID': '5',
'File_ID': 'Dial_20111128_113927',
}
HCRC MapTask Corpus
For the maptask configuration one example from the dataset is:
{
'Speaker': 'f',
'Utterance': 'is it underneath the rope bridge or to the left',
'Dialogue_Act': 6, # 'query_w'
'Dialogue_ID': '0',
'File_ID': 'q4ec1',
}
VERBMOBIL
For the vm2 configuration one example from the dataset is:
{
'Utterance': 'ja was sind viereinhalb Stunden Bahngerüttel gegen siebzig Minuten Turbulenzen im Flugzeug',
'Utterance': 'Utterance',
'Dialogue_Act': 'Dialogue_Act', # 'INFORM'
'Speaker': 'A',
'Dialogue_ID': '66',
}
Data Fields
For the dihana configuration, the different fields are:
Speaker: identifier of the speaker as a string.Utterance: Utterance as a string.Dialogue_Act: Dialog act label of the utterance. It can be one of 'Afirmacion' (0) [Feedback_positive], 'Apertura' (1) [Opening], 'Cierre' (2) [Closing], 'Confirmacion' (3) [Acknowledge], 'Espera' (4) [Hold], 'Indefinida' (5) [Undefined], 'Negacion' (6) [Feedback_negative], 'No_entendido' (7) [Request_clarify], 'Nueva_consulta' (8) [New_request], 'Pregunta' (9) [Request] or 'Respuesta' (10) [Reply].Dialogue_ID: identifier of the dialogue as a string.File_ID: identifier of the source file as a string.
For the ilisten configuration, the different fields are:
Speaker: identifier of the speaker as a string.Utterance: Utterance as a string.Dialogue_Act: Dialog act label of the utterance. It can be one of 'AGREE' (0), 'ANSWER' (1), 'CLOSING' (2), 'ENCOURAGE-SORRY' (3), 'GENERIC-ANSWER' (4), 'INFO-REQUEST' (5), 'KIND-ATTITUDE_SMALL-TALK' (6), 'OFFER-GIVE-INFO' (7), 'OPENING' (8), 'PERSUASION-SUGGEST' (9), 'QUESTION' (10), 'REJECT' (11), 'SOLICITATION-REQ_CLARIFICATION' (12), 'STATEMENT' (13) or 'TALK-ABOUT-SELF' (14).Dialogue_ID: identifier of the dialogue as a string.
For the loria configuration, the different fields are:
Speaker: identifier of the speaker as a string.Utterance: Utterance as a string.Dialogue_Act: Dialog act label of the utterance. It can be one of 'ack' (0), 'ask' (1), 'find_mold' (2), 'find_plans' (3), 'first_step' (4), 'greet' (5), 'help' (6), 'inform' (7), 'inform_engine' (8), 'inform_job' (9), 'inform_material_space' (10), 'informer_conditioner' (11), 'informer_decoration' (12), 'informer_elcomps' (13), 'informer_end_manufacturing' (14), 'kindAtt' (15), 'manufacturing_reqs' (16), 'next_step' (17), 'no' (18), 'other' (19), 'quality_control' (20), 'quit' (21), 'reqRep' (22), 'security_policies' (23), 'staff_enterprise' (24), 'staff_job' (25), 'studies_enterprise' (26), 'studies_job' (27), 'todo_failure' (28), 'todo_irreparable' (29), 'yes' (30)Dialogue_ID: identifier of the dialogue as a string.File_ID: identifier of the source file as a string.
For the maptask configuration, the different fields are:
Speaker: identifier of the speaker as a string.Utterance: Utterance as a string.Dialogue_Act: Dialog act label of the utterance. It can be one of 'acknowledge' (0), 'align' (1), 'check' (2), 'clarify' (3), 'explain' (4), 'instruct' (5), 'query_w' (6), 'query_yn' (7), 'ready' (8), 'reply_n' (9), 'reply_w' (10) or 'reply_y' (11).Dialogue_ID: identifier of the dialogue as a string.File_ID: identifier of the source file as a string.
For the vm2 configuration, the different fields are:
Utterance: Utterance as a string.Dialogue_Act: Dialogue act label of the utterance. It can be one of 'ACCEPT' (0), 'BACKCHANNEL' (1), 'BYE' (2), 'CLARIFY' (3), 'CLOSE' (4), 'COMMIT' (5), 'CONFIRM' (6), 'DEFER' (7), 'DELIBERATE' (8), 'DEVIATE_SCENARIO' (9), 'EXCLUDE' (10), 'EXPLAINED_REJECT' (11), 'FEEDBACK' (12), 'FEEDBACK_NEGATIVE' (13), 'FEEDBACK_POSITIVE' (14), 'GIVE_REASON' (15), 'GREET' (16), 'INFORM' (17), 'INIT' (18), 'INTRODUCE' (19), 'NOT_CLASSIFIABLE' (20), 'OFFER' (21), 'POLITENESS_FORMULA' (22), 'REJECT' (23), 'REQUEST' (24), 'REQUEST_CLARIFY' (25), 'REQUEST_COMMENT' (26), 'REQUEST_COMMIT' (27), 'REQUEST_SUGGEST' (28), 'SUGGEST' (29), 'THANK' (30).Speaker: Speaker as a string.Dialogue_ID: identifier of the dialogue as a string.
Data Splits
| Dataset name | Train | Valid | Test |
|---|---|---|---|
| dihana | 19063 | 2123 | 2361 |
| ilisten | 1986 | 230 | 971 |
| loria | 8465 | 942 | 1047 |
| maptask | 25382 | 5221 | 5335 |
| vm2 | 25060 | 2860 | 2855 |
Dataset Creation
Curation Rationale
[More Information Needed]
Source Data
Initial Data Collection and Normalization
[More Information Needed]
Who are the source language producers?
[More Information Needed]
Annotations
Annotation process
[More Information Needed]
Who are the annotators?
[More Information Needed]
Personal and Sensitive Information
[More Information Needed]
Considerations for Using the Data
Social Impact of Dataset
[More Information Needed]
Discussion of Biases
[More Information Needed]
Other Known Limitations
[More Information Needed]
Additional Information
Dataset Curators
Anonymous.
Licensing Information
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 Unported License.
Citation Information
@inproceedings{colombo-etal-2021-code,
title = "Code-switched inspired losses for spoken dialog representations",
author = "Colombo, Pierre and
Chapuis, Emile and
Labeau, Matthieu and
Clavel, Chlo{\'e}",
booktitle = "Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing",
month = nov,
year = "2021",
address = "Online and Punta Cana, Dominican Republic",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2021.emnlp-main.656",
doi = "10.18653/v1/2021.emnlp-main.656",
pages = "8320--8337",
abstract = "Spoken dialogue systems need to be able to handle both multiple languages and multilinguality inside a conversation (\textit{e.g} in case of code-switching). In this work, we introduce new pretraining losses tailored to learn generic multilingual spoken dialogue representations. The goal of these losses is to expose the model to code-switched language. In order to scale up training, we automatically build a pretraining corpus composed of multilingual conversations in five different languages (French, Italian, English, German and Spanish) from OpenSubtitles, a huge multilingual corpus composed of 24.3G tokens. We test the generic representations on MIAM, a new benchmark composed of five dialogue act corpora on the same aforementioned languages as well as on two novel multilingual tasks (\textit{i.e} multilingual mask utterance retrieval and multilingual inconsistency identification). Our experiments show that our new losses achieve a better performance in both monolingual and multilingual settings.",
}
Contributions
Thanks to @eusip and @PierreColombo for adding this dataset.
- Downloads last month
- 46