Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,154 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
language:
|
3 |
+
- multilingual
|
4 |
+
license:
|
5 |
+
- cc-by-4.0
|
6 |
+
multilinguality:
|
7 |
+
- multilingual
|
8 |
+
source_datasets:
|
9 |
+
- nluplusplus
|
10 |
+
task_categories:
|
11 |
+
- text-classification
|
12 |
+
pretty_name: multi3-nlu
|
13 |
+
|
14 |
+
---
|
15 |
+
|
16 |
+
# Dataset Card for Multi<sup>3</sup>NLU++
|
17 |
+
|
18 |
+
## Table of Contents
|
19 |
+
- [Dataset Description](#dataset-description)
|
20 |
+
- [Dataset Summary](#dataset-summary)
|
21 |
+
- [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
|
22 |
+
- [Languages](#languages)
|
23 |
+
- [Dataset Structure](#dataset-structure)
|
24 |
+
- [Data Instances](#data-instances)
|
25 |
+
- [Data Fields](#data-fields)
|
26 |
+
- [Data Splits](#data-splits)
|
27 |
+
- [Dataset Creation](#dataset-creation)
|
28 |
+
- [Curation Rationale](#curation-rationale)
|
29 |
+
- [Source Data](#source-data)
|
30 |
+
- [Personal and Sensitive Information](#personal-and-sensitive-information)
|
31 |
+
- [Considerations for Using the Data](#considerations-for-using-the-data)
|
32 |
+
- [Discussion of Biases](#discussion-of-biases)
|
33 |
+
- [Other Known Limitations](#other-known-limitations)
|
34 |
+
- [Additional Information](#additional-information)
|
35 |
+
- [Licensing Information](#licensing-information)
|
36 |
+
- [Citation Information](#citation-information)
|
37 |
+
- [Contact](#contact)
|
38 |
+
|
39 |
+
## Dataset Description
|
40 |
+
|
41 |
+
- **Paper:** [arXiv]()
|
42 |
+
|
43 |
+
|
44 |
+
### Dataset Summary
|
45 |
+
|
46 |
+
Multi<sup>3</sup>NLU++ consists of 3080 utterances per language representing challenges in building multilingual multi-intent multi-domain task-oriented dialogue systems. The domains include banking and hotels. There are 68 unique intents.
|
47 |
+
|
48 |
+
### Supported Tasks and Leaderboards
|
49 |
+
|
50 |
+
- multi-label intent detection
|
51 |
+
- slot filling
|
52 |
+
- cross-lingual language understanding for task-oriented dialogue
|
53 |
+
|
54 |
+
### Languages
|
55 |
+
|
56 |
+
The dataset covers four language pairs in addition to the source dataset in English:
|
57 |
+
Spanish, Turkish, Marathi, Amharic
|
58 |
+
|
59 |
+
## Dataset Structure
|
60 |
+
|
61 |
+
### Data Instances
|
62 |
+
|
63 |
+
Each data instance contains the following features: _text_, _intents_, _uid_, _lang_, and ocassionally _slots_ and _values_
|
64 |
+
|
65 |
+
See the [Multi<sup>3</sup>NLU++ corpus viewer](https://huggingface.co/datasets/uoe-nlp/multi3-nlu/viewer/uoe-nlp--multi3-nlu/train) to explore more examples.
|
66 |
+
|
67 |
+
An example from the Multi<sup>3</sup>NLU++ looks like the following:
|
68 |
+
```
|
69 |
+
{
|
70 |
+
"text": "माझे उद्याचे रिझर्वेशन मला रद्द का करता येणार नाही?",
|
71 |
+
"intents": [
|
72 |
+
"why",
|
73 |
+
"booking",
|
74 |
+
"cancel_close_leave_freeze",
|
75 |
+
"wrong_notworking_notshowing"
|
76 |
+
],
|
77 |
+
"slots": {
|
78 |
+
"date_from": {
|
79 |
+
"text": "उद्याचे",
|
80 |
+
"span": [
|
81 |
+
5,
|
82 |
+
12
|
83 |
+
],
|
84 |
+
"value": {
|
85 |
+
"day": 16,
|
86 |
+
"month": 3,
|
87 |
+
"year": 2022
|
88 |
+
}
|
89 |
+
}
|
90 |
+
},
|
91 |
+
"uid": "hotel_1_1",
|
92 |
+
"lang": "mr"
|
93 |
+
|
94 |
+
}
|
95 |
+
```
|
96 |
+
|
97 |
+
### Data Fields
|
98 |
+
|
99 |
+
- 'text': a string containing the utterance for which the intent needs to be detected
|
100 |
+
- 'intents': the corresponding intent labels
|
101 |
+
- 'uid': unique identifier per language
|
102 |
+
- 'lang': the language of the dataset
|
103 |
+
- 'slots': annotation of the span that needs to be extracted for value extraction with its label and _value_
|
104 |
+
|
105 |
+
|
106 |
+
### Data Splits
|
107 |
+
|
108 |
+
The experiments are done on different k-fold validation setups. The dataset has multiple types of data splits. Please see Section 4 of the paper.
|
109 |
+
|
110 |
+
## Dataset Creation
|
111 |
+
|
112 |
+
### Curation Rationale
|
113 |
+
Existing task-oriented dialogue datasets are 1) predominantly limited to detecting a single intent, 2) focused on a single domain, and 3) include a small set of slot types. Furthermore, the success of task-oriented dialogue is \textbf{4)} often evaluated on a small set of higher-resource languages (i.e., typically English) which does not test how generalisable systems are to the diverse range of the world's languages.
|
114 |
+
Our proposed dataset addresses all these limitations
|
115 |
+
|
116 |
+
|
117 |
+
### Source Data
|
118 |
+
|
119 |
+
|
120 |
+
#### Initial Data Collection and Normalization
|
121 |
+
Please see Section 3 of the paper
|
122 |
+
|
123 |
+
#### Who are the source language producers?
|
124 |
+
The source language producers are authors of [NLU++ dataset](https://arxiv.org/abs/2204.13021). The dataset was professionally translated into our chosen four languages. We used Blend Express and Proz.com to recruit these translators.
|
125 |
+
|
126 |
+
|
127 |
+
### Personal and Sensitive Information
|
128 |
+
|
129 |
+
None. Names are fictional
|
130 |
+
|
131 |
+
|
132 |
+
|
133 |
+
### Discussion of Biases
|
134 |
+
|
135 |
+
We have carefully vetted the examples to exclude the problematic examples.
|
136 |
+
|
137 |
+
### Other Known Limitations
|
138 |
+
The dataset comprises utterances extracted from real dialogues between users and conversational agents as well as synthetic human-authored utterances constructed with the aim of introducing additional combinations of intents and slots. The utterances therefore lack the wider context that would be present in a complete dialogue. As such the dataset cannot be used to evaluate systems with respect to discourse-level phenomena present in dialogue.
|
139 |
+
|
140 |
+
## Additional Information
|
141 |
+
N/A
|
142 |
+
|
143 |
+
### Licensing Information
|
144 |
+
|
145 |
+
The dataset is Creative Commons Attribution 4.0 International (cc-by-4.0)
|
146 |
+
|
147 |
+
### Citation Information
|
148 |
+
|
149 |
+
Coming soon
|
150 |
+
|
151 |
+
### Contact
|
152 |
+
[Nikita Moghe](mailto:nikita.moghe@ed.ac.uk) and E[Liane Guillou](mailto:lguillou@ed.ac.uk)
|
153 |
+
|
154 |
+
Dataset card based on [Allociné](https://huggingface.co/datasets/allocine)
|