Semantic parsing is the task of translating natural language into a formal meaning representation on which a machine can act. Representations may be an executable language such as SQL or more abstract representations such as Abstract Meaning Representation (AMR).
Each AMR is a single rooted, directed graph. AMRs include PropBank semantic roles, within-sentence coreference, named entities and types, modality, negation, questions, quantities, and so on. See.
13,051 sentences
Models are evaluated on the newswire section and the full dataset based on smatch. Systems marked with * are pipeline systems that require other systems (i.e. a dependency parse or a SRL parse) as input.
| Model | F1 Newswire | F1 Full | Paper / Source |
|---|---|---|---|
| Incremental joint model (Zhou et al., 2016)* | 0.71 | 0.66 | AMR Parsing with an Incremental Joint Model |
| Transition-based transducer (Wang et al., 2015)* | 0.70 | 0.66 | Boosting Transition-based AMR Parsing with Refined Actions and Auxiliary Analyzers |
| Imitation learning (Goodman et al., 2016)* | 0.70 | -- | Noise reduction and targeted exploration in imitation learning for Abstract Meaning Representation parsing |
| MT-Based (Pust et al., 2015)* | -- | 0.66 | Parsing English into Abstract Meaning Representation Using Syntax-Based Machine Translation |
| Transition-based parser-Stack-LSTM (Ballesteros and Al-Onaizan, 2017)* | 0.69 | 0.64 | AMR Parsing using Stack-LSTMs |
| Transition-based parser-Stack-LSTM (Ballesteros and Al-Onaizan, 2017) | 0.68 | 0.63 | AMR Parsing using Stack-LSTMs |
19,572 sentences
Models are evaluated based on smatch.
| Model | Smatch | Paper / Source |
|---|---|---|
| Joint model (Lyu and Titov, 2018) | 73.7 | AMR Parsing as Graph Prediction with Latent Alignment |
| Mul-BiLSTM (Foland and Martin, 2017) | 70.7 | Abstract Meaning Representation Parsing using LSTM Recurrent Neural Networks |
| JAMR (Flanigan et al., 2016) | 67.0 | CMU at SemEval-2016 Task 8: Graph-based AMR Parsing with Infinite Ramp Loss |
| CAMR (Wang et al., 2016) | 66.5 | CAMR at SemEval-2016 Task 8: An Extended Transition-based AMR Parser |
| AMREager (Damonte et al., 2017) | 64.0 | An Incremental Parser for Abstract Meaning Representation |
| SEQ2SEQ + 20M (Konstas et al., 2017) | 62.1 | Neural AMR: Sequence-to-Sequence Models for Parsing and Generation |
39,260 sentences
Results are computed over 8 runs. Models are evaluated based on smatch.
| Model | Smatch | Paper / Source |
|---|---|---|
| Joint model (Lyu and Titov, 2018) | 74.4 | AMR Parsing as Graph Prediction with Latent Alignment |
| ChSeq + 100K (van Noord and Bos, 2017) | 71.0 | Neural Semantic Parsing by Character-based Translation: Experiments with Abstract Meaning Representations |
| Neural-Pointer (Buys and Blunsom, 2017) | 61.9 | Oxford at SemEval-2017 Task 9: Neural AMR Parsing with Pointer-Augmented Attention |
5,280 user questions for a flight-booking task:
- Collected and manually annotated with SQL Dahl et al., (1994)
- Modified by Iyer et al., (2017) to reduce nesting
- Bugfixes and changes to a canonical style by Finegan-Dollak et al., (2018)
Example:
| Question | SQL query |
|---|---|
| what flights from any city land at MKE | SELECT DISTINCT FLIGHTalias0.FLIGHT_ID FROM AIRPORT AS AIRPORTalias0 , AIRPORT_SERVICE AS AIRPORT_SERVICEalias0 , CITY AS CITYalias0 , FLIGHT AS FLIGHTalias0 WHERE AIRPORTalias0.AIRPORT_CODE = "MKE" AND CITYalias0.CITY_CODE = AIRPORT_SERVICEalias0.CITY_CODE AND FLIGHTalias0.FROM_AIRPORT = AIRPORT_SERVICEalias0.AIRPORT_CODE AND FLIGHTalias0.TO_AIRPORT = AIRPORTalias0.AIRPORT_CODE ; |
| Model | Question Split | Query Split | Paper / Source | Code |
|---|---|---|---|---|
| Seq2Seq with copying (Finegan-Dollak et al., 2018) | 51 | 32 | Improving Text-to-SQL Evaluation Methodology | Data and System |
| Iyer et al., (2017) | 45 | 17 | Learning a neural semantic parser from user feedback | System |
| Template Baseline (Finegan-Dollak et al., 2018) | 45 | 0 | Improving Text-to-SQL Evaluation Methodology | Data and System |
877 user questions about US geography:
- Collected and manually annotated with Prolog Zelle and Mooney (1996)
- Most questions were converted to SQL by Popescu et al., (2003)
- Remaining question converted to SQL by Giordani and Moschitti (2012), and independently by Iyer et al., (2017)
- Bugfixes and changes to a canonical style by Finegan-Dollak et al., (2018)
Example:
| Question | SQL query |
|---|---|
| what is the biggest city in arizona | SELECT CITYalias0.CITY_NAME FROM CITY AS CITYalias0 WHERE CITYalias0.POPULATION = ( SELECT MAX( CITYalias1.POPULATION ) FROM CITY AS CITYalias1 WHERE CITYalias1.STATE_NAME = "arizona" ) AND CITYalias0.STATE_NAME = "arizona" |
| Model | Question Split | Query Split | Paper / Source | Code |
|---|---|---|---|---|
| Seq2Seq with copying (Finegan-Dollak et al., 2018) | 71 | 20 | Improving Text-to-SQL Evaluation Methodology | Data and System |
| Iyer et al., (2017) | 66 | 40 | Learning a neural semantic parser from user feedback | System |
| Template Baseline (Finegan-Dollak et al., 2018) | 66 | 0 | Improving Text-to-SQL Evaluation Methodology | Data and System |
817 user questions about academic publications, with automatically generated SQL that was checked by asking the user if the output was correct.
- Collected by Iyer et al., (2017)
- Bugfixes and changes to a canonical style by Finegan-Dollak et al., (2018)
Example:
| Question | SQL query |
|---|---|
| What papers has sharon goldwater written ? | SELECT DISTINCT WRITESalias0.PAPERID FROM AUTHOR AS AUTHORalias0 , WRITES AS WRITESalias0 WHERE AUTHORalias0.AUTHORNAME = "sharon goldwater" AND WRITESalias0.AUTHORID = AUTHORalias0.AUTHORID ; |
| Model | Question Split | Query Split | Paper / Source | Code |
|---|---|---|---|---|
| Seq2Seq with copying (Finegan-Dollak et al., 2018) | 59 | 5 | Improving Text-to-SQL Evaluation Methodology | Data and System |
| Template Baseline (Finegan-Dollak et al., 2018) | 52 | 0 | Improving Text-to-SQL Evaluation Methodology | Data and System |
| Iyer et al., (2017) | 44 | 3 | Learning a neural semantic parser from user feedback | System |
4,570 user questions about university course advising, with manually annotated SQL Finegan-Dollak et al., (2018).
Example:
| Question | SQL query |
|---|---|
| Can undergrads take 550 ? | SELECT DISTINCT COURSEalias0.ADVISORY_REQUIREMENT , COURSEalias0.ENFORCED_REQUIREMENT , COURSEalias0.NAME FROM COURSE AS COURSEalias0 WHERE COURSEalias0.DEPARTMENT = \"department0\" AND COURSEalias0.NUMBER = 550 ; |
| Model | Question Split | Query Split | Paper / Source | Code |
|---|---|---|---|---|
| Template Baseline (Finegan-Dollak et al., 2018) | 80 | 0 | Improving Text-to-SQL Evaluation Methodology | Data and System |
| Seq2Seq with copying (Finegan-Dollak et al., 2018) | 70 | 0 | Improving Text-to-SQL Evaluation Methodology | Data and System |
| Iyer et al., (2017) | 41 | 1 | Learning a neural semantic parser from user feedback | System |
The WikiSQL dataset consists of 87,673 examples of questions, SQL queries, and database tables built from 26,521 tables. Train/dev/test splits are provided so that each table is only in one split. Models are evaluated based on accuracy on execute result matches.
Example:
| Question | SQL query |
|---|---|
| How many engine types did Val Musetti use? | SELECT COUNT Engine WHERE Driver = Val Musetti |
| Model | Acc ex | Paper / Source |
|---|---|---|
| TypeSQL+TC (Yu et al., 2018) | 82.6 | TypeSQL: Knowledge-based Type-Aware Neural Text-to-SQL Generation |
| SQLNet (Xu et al., 2017) | 68.0 | Sqlnet: Generating structured queries from natural language without reinforcement learning |
| Seq2SQL (Zhong et al., 2017) | 59.4 | Seq2sql: Generating structured queries from natural language using reinforcement learning |
Spider is a large-scale complex and cross-domain semantic parsing and text-to-SQL dataset. It consists of 10,181 questions and 5,693 unique complex SQL queries on 200 databases with multiple tables covering 138 different domains. In Spider 1.0, different complex SQL queries and databases appear in train and test sets.
The Spider dataset can be accessed and leaderboard can be accessed here.
Restaurants - 378 questions about restaurants, their cuisine and locations, collected by Tang and Mooney (2000), converted to SQL by [Popescu et al., (2003)]((http://doi.acm.org/10.1145/604045.604070) and Giordani and Moschitti (2012), improved and converted to canonical style by Finegan-Dollak et al., (2018)
Example:
| Question | SQL query |
|---|---|
| where is a restaurant in alameda ? | SELECT LOCATIONalias0.HOUSE_NUMBER , RESTAURANTalias0.NAME FROM LOCATION AS LOCATIONalias0 , RESTAURANT AS RESTAURANTalias0 WHERE LOCATIONalias0.CITY_NAME = "alameda" AND RESTAURANTalias0.ID = LOCATIONalias0.RESTAURANT_ID ; |
| Model | Question Split | Query Split | Paper / Source | Code |
|---|---|---|---|---|
| Iyer et al., (2017) | 100 | 8 | Learning a neural semantic parser from user feedback | System |
| Seq2Seq with copying (Finegan-Dollak et al., 2018) | 100 | 4 | Improving Text-to-SQL Evaluation Methodology | Data and System |
| Template Baseline (Finegan-Dollak et al., 2018) | 95 | 0 | Improving Text-to-SQL Evaluation Methodology | Data and System |
Academic - 196 questions about publications generated by enumerating all of the different queries possible with the Microsoft Academic Search interface, then writing questions for each query Li and Jagadish (2014). Improved and converted to a cononical style by Finegan-Dollak et al., (2018).
Example:
| Question | SQL query |
|---|---|
| return me the homepage of PVLDB | SELECT JOURNALalias0.HOMEPAGE FROM JOURNAL AS JOURNALalias0 WHERE JOURNALalias0.NAME = "PVLDB" ; |
| Model | Question Split | Query Split | Paper / Source | Code |
|---|---|---|---|---|
| Seq2Seq with copying (Finegan-Dollak et al., 2018) | 81 | 74 | Improving Text-to-SQL Evaluation Methodology | Data and System |
| Iyer et al., (2017) | 76 | 70 | Learning a neural semantic parser from user feedback | System |
| Template Baseline (Finegan-Dollak et al., 2018) | 0 | 0 | Improving Text-to-SQL Evaluation Methodology | Data and System |
Yelp - 128 user questions about the Yelp website Yaghmazadeh et al., 2017. Improved and converted to a cononical style by Finegan-Dollak et al., (2018).
Example:
| Question | SQL query |
|---|---|
| List all businesses with rating 3.5 | SELECT BUSINESSalias0.NAME FROM BUSINESS AS BUSINESSalias0 WHERE BUSINESSalias0.RATING = 3.5 ; |
| Model | Question Split | Query Split | Paper / Source | Code |
|---|---|---|---|---|
| Seq2Seq with copying (Finegan-Dollak et al., 2018) | 12 | 4 | Improving Text-to-SQL Evaluation Methodology | Data and System |
| Iyer et al., (2017) | 6 | 6 | Learning a neural semantic parser from user feedback | System |
| Template Baseline (Finegan-Dollak et al., 2018) | 1 | 0 | Improving Text-to-SQL Evaluation Methodology | Data and System |
IMDB - 131 user questions about the Internet Movie Database Yaghmazadeh et al., 2017. Improved and converted to a cononical style by Finegan-Dollak et al., (2018).
Example:
| Question | SQL query |
|---|---|
| What year was the movie " The Imitation Game " produced | SELECT MOVIEalias0.RELEASE_YEAR FROM MOVIE AS MOVIEalias0 WHERE MOVIEalias0.TITLE = "The Imitation Game" ; |
| Model | Question Split | Query Split | Paper / Source | Code |
|---|---|---|---|---|
| Seq2Seq with copying (Finegan-Dollak et al., 2018) | 26 | 9 | Improving Text-to-SQL Evaluation Methodology | Data and System |
| Iyer et al., (2017) | 10 | 4 | Learning a neural semantic parser from user feedback | System |
| Template Baseline (Finegan-Dollak et al., 2018) | 0 | 0 | Improving Text-to-SQL Evaluation Methodology | Data and System |