This is a python3 project that converts some annotations of the ELITR Minuting Corpus from the original TEXT format into the JSON format, to be consistent with the format of ami-and-icsi-corpora.
The processed corpora can be directly downloaded from here (last update 21/04/2023).
Click ► in below to see example and explication of each annotation.
If you want to run the code yourself, please follow the following instructions.
-
Extract ELITR Minuting Corpus (ELITR-minuting-corpus.zip) under
input/ELITR-minuting-corpus. -
Run the following Python scripts to obtain respective annotations in JSON format under
output/.
- Manual meeting transcriptions (transcript.py). Note: you should run this script first.
example
[
...,
{
"id":"meeting_en_dev_003.u.4",
"speaker":"PERSON12",
"text":"And maybe you, you could write some this, some of the sections, I think.",
"problem":[
"1",
"None"
],
"gender":"male",
"annotator":"annot18"
},
...
]id denotes utterance id, e.g., "meeting_en_dev_003" refers to the meeting id, "u" to "utterance", "4" to the line number of the utterance in orginal transcript file.
speaker and gender denotes the deidentified speaker tags, and their gender, sometimes they both can be UNKNOWN.
The ID numbers are shuffled and unique for each meeting, i.e. PERSON1 denotes the same person across all the files of one meeting but a different person in the files of another meeting.
annotator denotes the annotator of the transcript.
problem denotes problematic or interesting properties of the utterance, e.g., ["1","None"] means there are two annotators, the first one thinks this utterance relates to organizational talk, and the second one find it contains no problematic or interesting properties. problem can be empty [].
1 - Organizational
Organizational talk not directly related to the subject of the meeting
(e.g. discussing technical issues with the video call).
2 - Speech incomprehensible
It is not clear what the speaker is saying.
3 - Other issue
4 - Small talk
Small talk or conversation unrelated to the subject of the meeting
(e.g. discussing the weather).
5 - Censored
This part of the transcript had to be removed for privacy reasons.
The text can contain entities, which are enclosed in square brackets, of types:
PERSON, ORGANIZATION, PROJECT, LOCATION, ANNOTATOR, URL,
NUMBER, PASSWORD, PHONE, PATH, EMAIL, OTHER.
The transcript data also contains the following tags:
<another_language>...</another_language> or <another_language/>
speech in a different language than the rest of the transcript
<typing/>
sounds of typing
<parallel_talk>...</parallel_talk> or <parallel_talk/>
speakers talking over each other
<cough/>
coughing
<other_yawn/>
yawning
<censored/>
section of the transcript has been censored for privacy
or ethical reasons
<laugh/>
laughter
<unintelligible/>
speech is not comprehensible
<other_sigh/>
sighing
<talking_to_self/>
speaker talking to themselves
<other_noise/>
another further unspecified noise
- Extractive summaries (extractive.py)
example
[
{
"id":"meeting_en_dev_003.u.11",
"speaker":"PERSON12",
"text":"So, so let's go on agenda.",
"problem":[
"None",
"None"
],
"gender":"male",
"annotator":"annot18"
},
{
"id":"meeting_en_dev_003.u.12",
"speaker":"PERSON12",
"text":"First we have accepted paper to [OTHER8], to [PROJECT2].",
"problem":[
"None",
"None"
],
"gender":"male",
"annotator":"annot18"
},
...
]One meeting can have zero to multiple extractive summaries, the file(s) are named in two ways:
- [meeting_id]_ORIG.json, summary provided by meeting organizer.
- [meeting_id]_GENER_annot[YY].json, summary provided by the annotator YY.
- Meeting minutes / Abstractive summaries (minute.py)
example
[
...,
{
"id":"meeting_en_dev_004.GENER_annot18.13",
"text":"- [PERSON5] explained [PERSON17] what he wanted to say in his comments"
},
{
"id":"meeting_en_dev_004.GENER_annot18.14",
"text":"- the next meeting is planned on 21th July"
},
...
]One meeting can have zero to multiple minutes, the file(s) are named in two ways:
- [meeting_id]_ORIG.json, the original agenda or minutes, written by meeting organizer.
- [meeting_id]_GENER_annot[YY].json, the minutes files, i.e. summaries written by the annotator YY.
The format of minute is somewhat free form but is always in the form of bullet points rather than a coherent text summary.
- Alignment / Abstractive summaries (alignment.py)
example
[
{
"abstractive":{
"id":"meeting_en_dev_004.GENER_annot02.9",
"text":"- inform that his part of book is ready to comments, except introduction and morfology."
},
"extractive":[
{
"id":"meeting_en_dev_004.u.94",
"speaker":"PERSON6",
"text":"OK, um, so I'm still missing the introduction so it's stil this, um, this, this bullets which I plan to rewrite into the text like one to put this text away.",
"problem":[
"None",
"None"
],
"gender":"male",
"annotator":"annot18"
},
{
"id":"meeting_en_dev_004.u.95",
"speaker":"PERSON6",
"text":"And I still didn't touch morfology so I still need to rewrite it into sentences.",
"problem":[
"None",
"None"
],
"gender":"male",
"annotator":"annot18"
},
...
]
},
...
]One meeting can have zero to multiple minutes, the file(s) are named in two ways:
- [meeting_id]_ORIG.json, the alignments between
minute/[meeting_id]_ORIG.jsonandtranscript/[meeting_id].json - [meeting_id]_GENER_annot[YY].json, the alignments between
minute/[meeting_id]_GENER_annot[YY].jsonandtranscript/[meeting_id].json
For more details of these annotations, please refer to the README and stats.tsv of the original corpus and the citations in below.
The data is separated into the following directories:
- elitr-minuting-corpus-en: the English data
- elitr-minuting-corpus-cs: the Czech data Meetings are further split into train, dev, test and test2 sets by their ids.
elitr_en_train = ['meeting_en_train_'+f'{idx:03}' for idx in range(1,85)]
elitr_en_dev = ['meeting_en_dev_'+f'{idx:03}' for idx in range(1,11)]
elitr_en_test = ['meeting_en_test_'+f'{idx:03}' for idx in range(1,19)]
elitr_en_test2 = ['meeting_en_test2_'+f'{idx:03}' for idx in range(1,9)]
elitr_cs_train = ['meeting_cs_train_'+f'{idx:03}' for idx in range(1,34)]
elitr_cs_dev = ['meeting_cs_dev_'+f'{idx:03}' for idx in range(1,11)]
elitr_cs_test = ['meeting_cs_test_'+f'{idx:03}' for idx in range(1,11)]
elitr_cs_test2 = ['meeting_cs_test2_'+f'{idx:03}' for idx in range(1,7)]In addition to the original dataset paper, if you find this repository helpful, please consider to cite the publications:
[1] Abstractive Meeting Summarization: A Survey
@article{rennard2022abstractive,
title={Abstractive Meeting Summarization: A Survey},
author={Rennard, Virgile and Shang, Guokan and Hunter, Julie and Vazirgiannis, Michalis},
journal={arXiv preprint arXiv:2208.04163},
year={2022}
}
[2] Energy-based Self-attentive Learning of Abstractive Communities for Spoken Language Understanding
@inproceedings{shang2020energy,
title={Energy-based Self-attentive Learning of Abstractive Communities for Spoken Language Understanding},
author={Shang, Guokan and Tixier, Antoine and Vazirgiannis, Michalis and Lorr{\'e}, Jean-Pierre},
booktitle={Proceedings of the 1st Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 10th International Joint Conference on Natural Language Processing},
pages={313--327},
year={2020}
}
[3] Spoken Language Understanding for Abstractive Meeting Summarization
@phdthesis{shang2021spoken,
title={Spoken Language Understanding for Abstractive Meeting Summarization},
author={Shang, Guokan},
year={2021},
school={Institut polytechnique de Paris}
}
@article{shang2018unsupervised,
title={Unsupervised abstractive meeting summarization with multi-sentence compression and budgeted submodular maximization},
author={Shang, Guokan and Ding, Wensi and Zhang, Zekun and Tixier, Antoine Jean-Pierre and Meladianos, Polykarpos and Vazirgiannis, Michalis and Lorr{\'e}, Jean-Pierre},
journal={arXiv preprint arXiv:1805.05271},
year={2018}
}