aliwister commited on
Commit
d171444
·
verified ·
1 Parent(s): 4537c5f

Add CoSQL-S2T, SParC-S2T and Spider-S2T test sets and dataset card

Browse files
README.md ADDED
@@ -0,0 +1,124 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-sa-4.0
3
+ language:
4
+ - en
5
+ pretty_name: SQL2Text Semantic Captioning Benchmark
6
+ task_categories:
7
+ - text-generation
8
+ tags:
9
+ - sql
10
+ - sql2text
11
+ - semantic-captioning
12
+ - code
13
+ - in-context-learning
14
+ size_categories:
15
+ - n<1K
16
+ configs:
17
+ - config_name: cosql_s2t
18
+ data_files:
19
+ - split: test
20
+ path: cosql_s2t/test-00000-of-00001.parquet
21
+ - config_name: sparc_s2t
22
+ data_files:
23
+ - split: test
24
+ path: sparc_s2t/test-00000-of-00001.parquet
25
+ - config_name: spider_s2t
26
+ data_files:
27
+ - split: test
28
+ path: spider_s2t/test-00000-of-00001.parquet
29
+ ---
30
+
31
+ # SQL2Text: a semantic captioning benchmark
32
+
33
+ Test sets for **SQL2Text**, the task of describing a SQL query in natural language (semantic captioning, the
34
+ reverse of Text2SQL). They were introduced in
35
+
36
+ > Ali Al-Lawati, Jason Lucas, Prasenjit Mitra.
37
+ > [Semantic Captioning: Benchmark Dataset and Graph-Aware Few-Shot In-Context Learning for SQL2Text](https://aclanthology.org/2025.coling-main.536/).
38
+ > COLING 2025.
39
+
40
+ A SQL query can be described correctly in many ways, so a single reference question, as found in Text2SQL
41
+ datasets, is a weak target for the reverse task. Each query here therefore comes with several reference
42
+ captions: three written by GPT-4o through an iterative generate, review and refine prompt, plus the original
43
+ human-written question from the source dataset where that question passed validation.
44
+
45
+ ## Contents
46
+
47
+ Three configurations, each with a single `test` split.
48
+
49
+ | Configuration | Source dataset | Queries | Generated captions | Original questions | Total references |
50
+ | --- | --- | ---: | ---: | ---: | ---: |
51
+ | `cosql_s2t` | CoSQL, dev | 290 | 870 | 280 | 1,150 |
52
+ | `sparc_s2t` | SParC, dev | 289 | 867 | 282 | 1,149 |
53
+ | `spider_s2t` | Spider, dev | 274 | 822 | 273 | 1,095 |
54
+ | Total | | 853 | 2,559 | 835 | 3,394 |
55
+
56
+ ```python
57
+ from datasets import load_dataset
58
+
59
+ ds = load_dataset("aliwister/SQL2Text", "spider_s2t", split="test")
60
+ print(ds[0])
61
+ # {'idx': 'spider-0',
62
+ # 'query': 'SELECT count(*) FROM singer',
63
+ # 'utterances': ['How many singers are there?',
64
+ # 'Please provide the total number of singers.',
65
+ # 'Can you count the total number of singers?',
66
+ # 'How many singers do we have?'],
67
+ # 'query_type': 'aggregate'}
68
+ ```
69
+
70
+ ## Fields
71
+
72
+ | Field | Type | Description |
73
+ | --- | --- | --- |
74
+ | `idx` | string | Identifier of the example, for example `cosql-17`. Identifiers are unique within a configuration and are not contiguous, because some examples were removed during validation. |
75
+ | `query` | string | The SQL query to be captioned. |
76
+ | `utterances` | list of strings | The reference captions. The first three were generated by GPT-4o. When there is a fourth, it is the original question from the source dataset; it is absent where the original question did not pass validation. |
77
+ | `query_type` | string | `simple`, `aggregate` or `nested`, the grouping used for the per-type results in the paper. |
78
+
79
+ ## How the data was built
80
+
81
+ 1. **Source.** Query and question pairs from the development sets of the three source datasets: all 293
82
+ examples of CoSQL and random samples of 293 examples each from SParC and Spider. For the conversational
83
+ datasets CoSQL and SParC, the pair is the final query and utterance of each interaction.
84
+ 2. **Generation.** For every query, GPT-4o followed a three-step prompt: generate three captions, review each
85
+ one for accuracy, clarity, fluency and relevance to the query, then refine them using that feedback. The
86
+ three references kept per query were chosen by AlignScore and BERTScore against the original question.
87
+ 3. **Validation.** Three evaluators with NLP and database experience checked every query together with its
88
+ original question and generated captions for semantic accuracy, preservation of the SQL logic and technical
89
+ precision (Fleiss' kappa 0.802). Inconsistent queries, original questions and captions were removed, which
90
+ left the 853 queries above.
91
+
92
+ The prompt and the full protocol are in the paper.
93
+
94
+ ## Use
95
+
96
+ The paper evaluates captions produced by a model for each `query` against the references in `utterances`, with
97
+ BLEU-4, BERTScore and AlignScore. Few-shot demonstrations in the paper are drawn from the training sets of the
98
+ source datasets, which are not redistributed here. Code for the experiments, including the graph-aware
99
+ demonstration selection (AST-ICL), is at [github.com/aliwister/ast-icl](https://github.com/aliwister/ast-icl).
100
+
101
+ ## Source datasets and licence
102
+
103
+ The queries and original questions come from [Spider](https://yale-lily.github.io/spider),
104
+ [SParC](https://yale-lily.github.io/sparc) and [CoSQL](https://yale-lily.github.io/cosql), all distributed
105
+ under CC BY-SA 4.0. This dataset is released under the same licence. Please also cite the source datasets
106
+ when using it.
107
+
108
+ ## Citation
109
+
110
+ ```bibtex
111
+ @inproceedings{al-lawati-etal-2025-semantic,
112
+ title = "Semantic Captioning: Benchmark Dataset and Graph-Aware Few-Shot In-Context Learning for {SQL}2{T}ext",
113
+ author = "Al Lawati, Ali and
114
+ Lucas, Jason and
115
+ Mitra, Prasenjit",
116
+ booktitle = "Proceedings of the 31st International Conference on Computational Linguistics",
117
+ month = jan,
118
+ year = "2025",
119
+ address = "Abu Dhabi, UAE",
120
+ publisher = "Association for Computational Linguistics",
121
+ url = "https://aclanthology.org/2025.coling-main.536/",
122
+ pages = "8026--8042"
123
+ }
124
+ ```
cosql_s2t/test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f832e30401761ad0fa5cf760c26691b604cec50e2188000d64622928543a1b0f
3
+ size 49561
sparc_s2t/test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0fb6b27e6a26377079295dce6dbdaadd8c718f4f0cd93efcf6d39935d58d58ee
3
+ size 44995
spider_s2t/test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:645d51adddf61a8f62afe24825dae5fd96f55233747c84bcfeb5586ac17ef451
3
+ size 45879