See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/359349231

Natural Language to SQL Queries: A Review

Article in International Journal of Innovations in Science and Technology · February 2022

DOI: 10.33411/IJIST/2022040111

CITATIONS READS

5 2,520

6 authors, including:

Muhammad Shahzaib Baig Azhar Imran

Air University of Islamabad Air University of Islamabad
2 PUBLICATIONS 5 CITATIONS 70 PUBLICATIONS 906 CITATIONS

SEE PROFILE SEE PROFILE

Muhammad Imran Khan

6 PUBLICATIONS 10 CITATIONS

SEE PROFILE

All content following this page was uploaded by Imran Rana Imran on 19 March 2022.

The user has requested enhancement of the downloaded file.

 International Journal of Innovations in Science & Technology

Natural Language to SQL Queries: A Review

Muhammad Shahzaib Baig1, Azhar Imran1, Aman Ullah Yasin1, Abdul Haleem Butt1,
Muhammad Imran Khan1
1
Department of Creative Technologies, Faculty of Computing & AI, Air University,
Islamabad
*Correspondence: Dr. Azhar Imran, azharimran63@gmail.com
Citation | Baig. M. S, Imran. A, Yasin. A. U, Butt. A. H, Khan. M. I, “Natural Language
to SQL Queries: A Review”. International Journal of Innovations in Science and
Technology, Vol 4, Issue 1, pp: 147-162, 2022.
Received | Jan 3, 2022; Revised | Feb 10, 2022 Accepted | Feb 12, 2022; Published |
Feb 22, 2022.
___________________________________________________________________
Abstract.
The relational database is the way of maintaining, storing, and accessing
structured data but in order to access the data in that database the queries need to be
translated in the format of SQL queries. Using natural language rather than SQL has
introduced the advancement of a new kind of handling strategy called Natural Language
Interface to Database frameworks (NLIDB). NLIDB is a stage towards the turn of
events of clever data set frameworks (IDBS) to upgrade the clients in performing
adaptable questioning in data sets. A model that can deduce relational database queries
from natural language. Advanced neural algorithms synthesize the end-to-end SQL to
text relation which results in the accuracy of 80% on the publicly available datasets. In
this paper, we reviewed the existing framework and compared them based on the
aggregation classifier, select column pointer, and the clause pointer. Furthermore, we
discussed the role of semantic parsing and neural algorithm’s contribution in predicting
the aggregation, column pointer, and clause pointer. In particular, people with limited
background knowledge are unable to access databases with ease. Using natural language
interfaces for relational databases is the solution to make natural language to SQL
queries. This paper presents a review of the existing framework to process natural
language to SQL queries and we will also cover some of the speech to SQL model in
discussion section, in order to understand their framework and to highlight the
limitations in the existing models.
Keywords: Natural Language Processing, Structured Query Language (SQL), Text to
Relational database, Natural Language Interface for Databases (NLIDB), Intelligent
Database System (IDBS), Flexible Querying, Database Management System (DBMS).
INTRODUCTION
Natural language interfacing to a relational database is a challenging and
important problem to a coupe. It requires a model that can understand natural language
Feb 2022 | Vol 4|Issue 1 Page | 147
 International Journal of Innovations in Science & Technology
and can translate it into relational database queries structures. Using natural language
rather than SQL has introduced the advancement of a new kind of handling strategy
called Natural Language Interface to Database frameworks (NLIDB). NLIDB is a stage
towards the turn of events of clever data set frameworks (IDBS) to upgrade the clients
in performing adaptable questioning in data sets [9]. A model that can deduce relational
database queries from natural language. Advanced neural algorithms synthesize the end-
to-end SQL to text relation which results in the accuracy of 80% on the publicly available
datasets [1].
In this paper, we will be covering the existing framework and compared based
on the aggregation classifier, select column pointer, and the clause pointers and we will
be looking into the role of semantic parsing and neural algorithm’s contribution in
predicting the aggregation, column pointer, and clause pointer that is the existing
solution to the problem. The LUNAR and LADDER systems were intended for non-
technical users to pose natural language queries regarding moon rock samples and US
Navy ships, respectively, in the 1970s (Woods, 1972) [2]. The fast advancement of
computer hardware and software over the last five decades has had such an impact on
databases that database systems established in the 1970s are no longer even compatible
with the current definition of a database (Bercich 2003; Frank 2018). To meet the
industry's demands, various natural language to database querying frameworks have
been created since then. We discovered intriguing research trends in converting natural
language to database queries domain retrieval time, reduced support for language
portability, and difficult configuration processes by evaluating the development timeline
of such systems [2].
It is not an easy process to translate a natural language question into multiple
database query languages like SQL, Simple Protocol, and RDF Query Language
(SPARQL), because today's databases are diversified, massive in size, and use complex
data storing technologies (Nadkarni, 2011) [3]. Storage engines frequently store data in
several formats, including structured (tabular), No SQL or graph (text), and hybrid. As
a result, distinct query languages are required by underlying storage engines to obtain
the stored data [4]. The diversity of data storage technologies complicates natural
language to database query translation. With the evolution of machine learning
techniques, many frameworks that can efficiently interpret natural language inquiries
have been developed.
Recent work shows that recurrent neural networks stacked with attention and
copying mechanism have greater potential in semantic parsing. The Seq2SQL model
shows that utilizing separate decoders for various pieces of an inquiry (i.e., accumulation
activity, target section, and where predicates) expands forecast precision, and supports
learning further works on the model by permitting it to adapt semantically comparable
inquiries past management.
In current study we have categorized the existing frameworks into subcategories
based on their implementation techniques for comparing their results. Frameworks are
categorized and discussed as SQL-Net, Syntax SQL-Net, Grammar SQL, IR-Net, Edit-
SQL, and RAT-SQL based on their algorithms and performance.

Feb 2022 | Vol 4|Issue 1 Page | 148

 International Journal of Innovations in Science & Technology
After evaluating the existing models for natural language to SQL, we have
discussed existing solutions to perform the speech to SQL queries and overviewed their
model in short and we will also discuss the Debug-It-Yourself (DIY) model in the
analysis and discussion section to evaluate their limitation and advantages along with
hurdles to use them as a practically deploying model for real case scenario, followed up
after all this the we will look into the future extension of the work and improvements
in the existing problem in view of author as closing remark of the discussing section
which includes ontological acyclic directed graphs as one of the solution to improve the
accuracy of such systems.
Existing Frameworks
The structured query language is the most used language for retrieving data from
the structured database, various methods are included in the no-SQL category that are
rule-based syntax analysis, semantic analysis, and pattern matching. Figure 1 shows the
conventional tree for natural language to database approach and Figure 2 discusses the
idea of seq2sql that is generating SQL queries using reinforced learning [4].
Furthermore, we discussed the existing framework for the problem as SQL-Net, Syntax
SQL-Net, Grammar SQL, IR-Net, Edit-SQL, and RAT-SQL.
SQL-Net
The SQL-Net model was introduced to minimize the use of the reinforced
learning as shown in Figure 2, by avoiding seq2seq in cases where an order is of low
significance. SQL-Net is based on a sketch-based approach which is comprised of the
dependency graph based on previous predictions so that previous predictions can be
used in the account of predicting the next queries reducing the use of reinforced learning
[3]. WikiSQL, one of the states of art models uses the SQL-Net approach to solve this
particular problem. Figure 3 shows the framework of WikiSQL. The SQL-Net model
also used a weight-based approach to highlight and give more attention to the significant
words and phrases in sentences. SQL-Net uses the concept of sequence to set prediction
which predicts the name of the columns [5]. It is a part of the generic attention
mechanism for feature attention map on question-based on column based. Predicting
the “Where” clause is the most challenging part of the processing [6].
SQL-Net searches for the “Where” clause and each chosen column and constraint
by predicting the operation and value using a sketch-based approach. Accuracy of SQL-
Net on Wiki-SQL test set is 64.4% and on SPIDER test set is around 12.4%. [6].
Seq2SQL framework inefficiencies (generalized to unseen schema and serializability)
and its Seq-to-Seq model have been improved with the aid of a new technique, namely
Xiaojun's sequence-to-set based model (Xu, 2017). The model was implemented in the
"SQL-NET" program, which uses Seq2SQL as a baseline foundation but does not use
reinforcement learning. A sketch-based technique, similar to SQLizer, has been
designed to parse the NL query, but each sketch contains a dependency neural network
to predict the new sketch based on a prior prediction of a sketch. This new model
increases Seq2SQL outcomes by 9% to 13% across many measures [7].

Feb 2022 | Vol 4|Issue 1 Page | 149

 International Journal of Innovations in Science & Technology

Figure 1. Classification of natural language to SQL query framework [7].

Figure 2. Frame of Natural Language to SQL queries using reinforced learning [5].

Feb 2022 | Vol 4|Issue 1 Page | 150

 International Journal of Innovations in Science & Technology

Figure 3. Frame of WikiSQL using SQL-Net [10].

Syntax SQL-Net
Syntax SQL-Net was designed to generate complex queries with multiple clauses
and cross-domain queries, generalizing previous models to complex database structures
by using the syntax tree networks shown in Figure 4. It uses table-aware encoders and
decoders having SQL generation path history. For handling complex queries at
decoders, the syntax SQL-Net uses SQL grammar for determining modules to be
invoked at each recursive step [9]. Syntax SQL-Net was assessed on the SPIDER dataset
and surpassed the previous model in precision by 7.3% [3].

Feb 2022 | Vol 4|Issue 1 Page | 151

 International Journal of Innovations in Science & Technology

Figure 4. Tree-based query generator using syntax SQL-Net [10]

Grammar SQL
The sequence-to-sequence paradigm used by neural text-to-SQL models usually
decodes at the token level and does not contemplate producing SQL hierarchically from
a grammar. The model provides a schema-dependent language with minimal over-
generation [20]. The shallow parsing expression language is intended to capture the least
amount of SQL required to cover most of the cases in the dataset. Later advancements
introduced non-terminals to context-free grammar for context-sensitive SQL
components to maintain consistency of table, column, and value references [25]. The
model utilizes runtime restrictions during decoding to ensure that only legitimate
programmers are allowed to link separate tables together using a foreign key. As input,
the suggested model accepts a natural language utterance, a database, and grammar
relevant to that speech. The SQL query is then built up repeatedly by the decoder
utilizing the attention technique on the input sequence [26]. Anonymization of
identifiers is also used to replace database IDs that occur in both the natural language
inquiry and the SQL query. Figure 5 shows the implementation of Grammar SQL [27].

Feb 2022 | Vol 4|Issue 1 Page | 152

 International Journal of Innovations in Science & Technology

Figure 5. Grammar SQL proposed model [13]

IR-Net
This model uses a neural approach and called Intermediate Representation (IR)
shown in Figure 6 [16] and it mainly focuses on two main problems in the existing
models i.e. the mismatch between intents in text and the prediction of column names
caused by a tremendous number of out of domain words. Instead of synthesizing SQL
queries end-to-end, IR-Net decomposes natural language into three parts. The schema
linking procedure over a database schema and a question are carried out during the first
phase [31]. IR-Net makes use of Sem-QL, a domain-specific language that acts as a
bridge between SQL and plain language. Each model component performs a unique
role in the Text2SQL job. Natural language input is encoded into an embedding vector
using the NL encoder. These embedding vectors are used to build hidden states with a
bi-directional LSTM. The schema encoder takes a database schema as input and
produces column and table representations. Finally, context-free grammar is employed
by the decoder to construct Sem-QL queries [36].
Zhong et al. introduced "Seq2SQL," a neural network-based framework for
interpreting natural language queries and mapping them to SQL (Structured Query
Language) representations (Zhong, Xiong, and Socher 2017). The suggested system
minimizes query space and enhances system execution accuracy. This framework used
Reinforcement Learning (RL) rewards and cross-entropy loss iteratively on query
execution over the database to create unordered parts of the question that are less
suitable for advancement using cross-entropy misfortune. They released WikiSQL, a
dataset including 87673 hand-explained examples of inquiries and SQL questions spread
across 26521 tables from Wikipedia. Seq2SQL outperforms the best in class semantic
parser by Dong and Lapata by incorporating strategy-based reinforcement learning (RL)
with an inquiry execution condition into WikiSQL (2016) [39].

Feb 2022 | Vol 4|Issue 1 Page | 153

 International Journal of Innovations in Science & Technology

Figure 6. Graphical implementation of IR-Net

Edit-SQL
Edit-SQL [16] is primarily concerned with the cross-domain context-dependent
text-to-SQL operation as shown in Figure 7. The authors take advantage of the fact that
adjacent natural language questions are dependent on one another and that
corresponding SQL queries overlap. They make use of this reliance by modifying the
previously anticipated query in order to increase the generating quality. The editing
mechanism accepts SQL sequences as input and reuses generating results at the token
level. To cope with complicated tables in many domains, an utterance-table encoder,
and a table-aware decoder is utilized to include the natural language context and the
schema [34]. User utterances and table structure are encoded using the utterance-table
encoder. To encode speech tokens, a bi-LSTM is employed.

Feb 2022 | Vol 4|Issue 1 Page | 154

 International Journal of Innovations in Science & Technology

Figure 7. Architecture of Edit-SQL form [8]

RAT-SQL
One of the most difficult aspects of converting natural language inquiries into
SQL queries is generalizing to unknown database schemas, flowchart for RAT-SQL is
shown below in figure 8. The generalization is dependent on encoding the database
relations in an accessible manner and modeling alignment between important database
columns in the query. The suggested architecture is built on the relation-aware self-
attention technique, which enables address schema encoding, feature representation,
and schema linking inside a text2SQL encoder [7]. Examine the flow chart below to gain
a fast idea of RAT-encoder-decoder SQL's structure.

Figure 8. Flowchart of RAT-SQL [7]

Analysis and Discussion
The models discussed above have their logics and algorithms to be considered.
The main objective of all the models is to improve the accuracy of the natural language
to SQL queries. We have compared all the mentioned models on the SPIDER dataset.
Spider is a large-scale, complicated, cross-domain semantic parsing and text-to-SQL
dataset. It comprises of 10,181 questions and 5,693 distinct sophisticated SQL queries
on 200 databases with numerous tables across 138 domains. In Spider, train and test
Feb 2022 | Vol 4|Issue 1 Page | 155
 International Journal of Innovations in Science & Technology
sets contain a variety of complicated SQL queries and databases. To perform effectively
on it, systems must be able to generalize not only to new SQL queries but also to new
database structures. SQL-Net models are designed using sketch-based algorithms and
reducing the use of reinforced learning to handle simple queries without focusing on
cross-domain SQL queries. Dar et al [2] calculated the accuracy of SQL-Net on different
datasets and the accuracy of SQL-Net on Wiki-SQL test set is 64.4% and on SPIDER
test set is around 12.4%. Syntax SQL-Net surpasses the SQL-Net in the accuracy of
15% absolute increase in the accuracy compared to SQL-Net on SPIDER dataset, the
accuracy of the SQL-Net is 12.4% while the accuracy observed by Tao Yu et al [10] is
27.2% in SPIDER dataset by also handling with complex and cross-domain queries.
Grammar SQL uses long short-term memory to handle natural language to SQL
queries. Grammar SQL can handle complex and cross-domain queries along with good
handling of denotation up to an accuracy of 80% and total accuracy calculated by Kevin
Lin et al is 52% on the SPIDER dataset. According to Jiaqi Guo et al., IRNet surpasses
all baselines by a significant margin. On the test set, it outperforms Syntax SQLNet by
27.0 percent. It also outperforms Syntax SQLNet, which conducts large-scale data
augmentation, by 19.5 percent. When BERT is used, the performance of both Syntax
SQLNet and IRNet improves significantly, and the accuracy gap between them on both
the development and test sets widens and the total accuracy is 44.5% but Jiaqi Guo et
al also used the IRNet with BERT model which gives the accuracy of 52.325%.
According to Zhang et al, the Edit-SQL model in cross-domain text2SQL
generation achieves 32.9% accuracy. Furthermore, using BERT embedding results in a
considerable gain in accuracy, attaining 53.4%. Wang et al mention RAT-SQL
outperforms prior benchmark models by 8.7% and achieves 57.2% on the SPIDER
dataset. Incorporating RAT-SQL with BERT results in a significant boost in
performance and 65.6% accuracy.
The model discussed above have their pros and cons in the context of
application, accuracy and the computational complexity. Models discussed above as
their measure of accuracy increases, they would seem to have more computational
complexity by using ensembling techniques [36]. As discussed earlier, the highest
performing model for under discussion task is the RAT-SQL ensembled with BERT
model achieving the accuracy of 65.6% on cross-domain SQL queries but RAT-SQL
uses the linking schema with relation-based encoder and decoder with BERT increases
its computation complexity [27]. While discussing in context of computational
complexity is less for syntax SQL-Net as it used syntax-based decision trees as compared
to the others but it only achieves the overall accuracy of 12.4% [28]. If we consider the
computational complexity along with the performance the IR-Net with BERT model
seem to perform good with over 50% accuracy but lesser computational complexity
[29].
Table 1. Comparison of accuracies of all frameworks
Dataset Model Overall Accuracy
SPIDER SQL-Net 12.4%
SPIDER Syntax SQL-Net 27.2%
SPIDER Grammar SQL 52%
Feb 2022 | Vol 4|Issue 1 Page | 156
 International Journal of Innovations in Science & Technology
SPIDER IR-Net 44.5%
SPIDER IR-Net with BERT 52.35%
SPIDER Edit-SQL 32.9%
SPIDER Edit-SQL with BERT 53.4%
SPIDER RAT-SQL 57.2%
SPIDER RAT-SQL with BERT 65.6%
While looking at the extensions in the existed model, Arpit Narechania et al [49]
discussed the important of importance of natural language to structured queries
mentioned the Debug-It-Yourself (DIY) approach based on state of art model NL2SQL
in which they provided user with a sandbox where user have the provision to interact
with three elements those are the mapping of the generated query on the basis of given
question, small subset of underlying database and ensembled modal explanation of
generated query, their discussion is the user interaction with the sandbox seems to
efficient for control environment testing but not for actually deploying the models in
professional environment where the feasibility and accessibility come into play. My
analysis for DIY approach using NL2SQL is strongly based on the argument that the
whole point of using systems like NL2SQL is to improve the feasibility element for the
individual who do not have expertise on SQL or any structured query, assumingly user
interaction in DIY is at that state which user is not familiar with either so Debug-It-
Yourself is not a part of benchmark evaluation. Serval authors like Blanning et al [47]
discussed this problem in context of semantic parsing in natural language processing
using LSTM or Bi-LSTM models but the issue that persists is the complex queries or
cross-domain queries, considering it into account it performs sufficiently good but it
cannot handle cross-domain SQL queries.
Another method discussed by Song et al [47] comprised on the speech to SQL
conversion in which they discussed that voice communication is far better approach as
compared to the user giving (typed inputs) to the system and then system converts the
input to SQL query. They mentioned that voice-based assistants for such purposes are
usually 6.7 times faster than the conventional input method. They mentioned voice-
based assistants like Google Home, Siri and Cortana have emerged the market leading
to much more feasible and accessible human computer interactions for general purposes
while considered the under-discussion problem, Song et al., mentioned some existing
models for speech to SQL query conversion, in which they mentioned SpeakQL model
which provides the speech to SQL conversion but this model restricts the voice to be
user to be as query which in my opinion is not an natural language to SQL but instead
SQL query in voice to text deploying it as speech to text.
The other model Song et al., mentioned is EchoQuery which is purposed for
same purpose that is speech to SQL conversion but EchoQuery follows the pattern of
query as “What is the {Aggregation}{Columns(s)} of {Table(s)}?”, and SpeakQL
requires the query to be an exact SQL statement such as “Select Salary From Employees
Where Name Equals John”. So far the models discussed in speech is also not converting
the free speech style to SQL queries so Song el al purposed a solution to the problem
which is based on the Speech Encoder, Schema Encoder, Speech Schema Relation
aware encoder and SQL aware decoder which they tested on benchmarks and gave a
Feb 2022 | Vol 4|Issue 1 Page | 157
 International Journal of Innovations in Science & Technology
cumulative accuracy of 15.29% on speech to text conversion but what is the impact of
accents, slangs and dialects on the results they have not either tested or at least
mentioned it.
These models dealing with speech to SQL are discussed here to globalize the
scope of the study as we are focused on the text to SQL query for this study. As a closing
remarks in discussion section, as the results shows that right at the moment there is no
such Natural Language Model exists which can handle or process the natural language
to structured query efficiently, till now the reason I can see here is the ambiguity in the
language or in other terms you can say it as that single sentence of English can have a
lot of meaning depending on its context and many expressions which cumulatively
defines the meaning and the structure of that particular sentence, so the context of the
sentence may also lead to some false SQL query cases. For future work extension, the
directive graphs or ontologies can be used to test the mentioned problems, it might
reduce the problem as the directive graphs defines the relation, context and structure of
sentence to an extent and I have not witnessed any research paper in this domain. An
example of ontological directive graphs is shown in the Figure 9.

Figure 9. Ontological Directive Graphs [50]

Conclusion
In this paper, we reviewed the framework to process the natural language to
simple and complex cross-domain queries. Furthermore, we have compared their
algorithms and compared the accuracy of all the models on the same dataset that is
SPIDER in our discussion, the dataset was kept the same throughout the discussion to
evaluate all the frameworks on the same benchmark. The results mentioned in Table 1
show that RAT-SQL model with BERT model is outperforming all the pre-existing
Feb 2022 | Vol 4|Issue 1 Page | 158
 International Journal of Innovations in Science & Technology
frameworks with an accuracy of 65.6% and Edit-SQL with BERT model standing at
second place with an accuracy of 53.4%. In future the RAT-SQL model can be further
improved to achieve higher accuracies in the context of aggregator and in the column
prediction. We also discussed Debug-It-Yourself (DIY) model based on NL2SQL
which improves user experience but have its own limitation relating to the prior
experience of the user so DIY model is not specifically used for individuals with no
prior or limited prior experience. We also discussed the speech to SQL query models in
which the state of art model till now is proposed by Song el al, this speech to SQL query
is not in the scope of the study but we have touched their limitations and advantages in
context of accessibility and feasibility in analysis and discussion section, so far, the
speech to text model seems to perform in limited and restriction environment. The
future work we can extend on the basis of prior studies is to extend the work of natural
language to SQL query using the ontological acyclic directed graphs for processing the
sentences to identify and understand the contextual information.

Acknowledgment. The author expresses gratitude to the Department of Creative

Technologies, Air University, Islamabad, under which the research has been carried out.
Furthermore, all the results mentioned in this paper have been credited to their original
authors and their work is mentioned in the reference.
Conflict of Interest. The author declares that there is no conflict of interests.

REFERENCES
[1] Singh, G., & Solanki, A. (2016). An algorithm to transform natural language into sql
queries for relational databases. Selforganizology, 3(3), 100-116. Sripad, Joshi, and
Laxmaiah E. n.d. 2013. Survey of Natural Language Interface to Databases.
[2] Kim, H., So, B. H., Han, W. S., & Lee, H. (2020). Natural language to SQL: Where
are we today? Proceedings of the VLDB Endowment, 13(10), 1737-1750.
[3] Vig, Jesse, and Kalai Ramea. “Comparison of transfer-learning approaches for
response selection in multi-turn conversations.” Workshop on DSTC7. 2019.
[4] Yu, Tao, et al. “Syntaxsqlnet: Syntax tree networks for the complex and cross-domain
text-to-SQL task.” arXiv preprint arXiv:1810.05237 (2018).
[5] Sun, Zeyu, et al. “A grammar-based structural CNN decoder for code
generation.” Proceedings of the AAAI Conference on Artificial Intelligence. Vol. 33.
2019.
[6] Finegan-Dollak, Catherine, et al. “Improving text-to-SQL evaluation
methodology.” arXiv preprint arXiv:1806.09029 (2018).
[7] Yu, Tao, et al. “Spider: A large-scale human-labeled dataset for complex and cross-
domain semantic parsing and text-to-SQL task.” arXiv preprint
arXiv:1809.08887 (2018).
[8] Hwang, Wonseok, et al. “A comprehensive exploration on wikisql with table-aware
word contextualization.” arXiv preprint arXiv:1902.01069 (2019).
[9] Lin, Kevin, et al. “Grammar-based neural text-to-SQL generation.” arXiv preprint
arXiv:1905.13326 (2019).

Feb 2022 | Vol 4|Issue 1 Page | 159

 International Journal of Innovations in Science & Technology
[10] Maas, Andrew, et al. “Learning word vectors for sentiment analysis.” Proceedings
of the 49th annual meeting of the association for computational linguistics: Human
language technologies. 2011.
[11] Xu, Xiaojun, Chang Liu, and Dawn Song. “Sqlnet: Generating structured queries
from natural language without reinforcement learning.” arXiv preprint
arXiv:1711.04436 (2017).
[12] Gardner, Matt, et al. “Allennlp: A deep semantic natural language processing
platform.” arXiv preprint arXiv:1803.07640 (2018).
[13] Affolter, Katrin, Kurt Stockinger, and Abraham Bernstein. “A Comparative Survey
of Recent Natural Language Interfaces for Databases.” The VLDB Journal 28.5 (2019):
793–819. Crossref. Web.
[14] Sujatha, B., & Raju, S. V. (2016). Natural Language Query Processing for Relational
Database using EFFCN Algorithm. International Journal of Computer Sciences and
Engineering, 4, 49-53.
[15] Sukthankar, N., Maharnawar, S., Deshmukh, P., Haribhakta, Y., & Kamble, V.
(2017). nQuery-A Natural Language Statement to SQL Query Generator. In
Proceedings of ACL 2017, Student Research Workshop (pp. 17-23).
[16] Stefan W., Ellen R., Gabriele S., (1996). Connectionist, Statistical and Symbolic
Approaches to Learning for Natural Language Processing, Springer.
[17] T. Ono, H. Hishigaki, A. Tanigami, T. Takagi, (2001), Automated extraction of
information on proteinprotein interactions from the biological literature,
Bioinformatics. doi:10.1093/bioinformatics/17.2.155.
[18] Warren, D. H., & Pereira, F. C. (1982). An efficient easily adaptable system for
interpreting natural language queries. Computational Linguistics, 8(3-4), 110-122.
[19] Woods, William A, Ronald M Kaplan, and Bonnie Nash-Webber. (1972) The lunar
sciences natural language information system. Bolt, Beranek and Newman,
Incorporated.
[20] Xu, X., Liu, C., & Song, D. (2017). Sqlnet: Generating structured queries from
natural language without reinforcement learning. arXiv preprint arXiv:1711.04436.
[21] Yossi Shani, Tal Cohen, and Yossi Vainshtein. (2016) "Natural Language Interface
for Databases." KUERI.ME. 2016. http://kueri.me/.
[22] Yaghmazadeh, N., Wang, Y., Dillig, I., & Dillig, T. (2017). Sqlizer: Query synthesis
from natural language. Proceedings of the ACM on Programming Languages,
1(OOPSLA), 63.
[23] Zhong, V., Xiong, C., & Socher, R. (2017). Seq2SQL: Generating Structured
Queries from Natural Language using Reinforcement Learning. arXiv preprint
arXiv:1709.00103.
[24] Lin, K., Bogin, B., Neumann, M., Berant, J., & Gardner, M. (2019). Grammar-based
neural text-to-sql generation. arXiv preprint arXiv:1905.13326.
[25] Zhang, Rui, et al. “Editing-Based SQL Query Generation for Cross-Domain
Context-Dependent Questions.” arXiv preprint arXiv:1909.00786 (2019).
[26] Wang, Bailin, et al. “Rat-SQL: Relation-aware schema encoding and linking for text-
to-SQL parsers.” arXiv preprint arXiv:1911.04942 (2019).

Feb 2022 | Vol 4|Issue 1 Page | 160

 International Journal of Innovations in Science & Technology
[27] Dar, Hafsa Shareef, et al. “Frameworks for Querying Databases Using Natural
Language: A Literature Review.” arXiv preprint arXiv:1909.01822 (2019).
[28] Popescu, A. M., Etzioni, O., & Kautz, H. (2003, January). Towards a theory of
natural language interfaces to databases. In Proceedings of the 8th international conference on
Intelligent user interfaces (pp. 149-157)
[29] Uma, M., Sneha, V., Sneha, G., Bhuvana, J., & Bharathi, B. (2019, February).
Formation of SQL from natural language query using NLP. In 2019 International
Conference on Computational Intelligence in Data Science (ICCIDS) (pp. 1-5). IEEE
[30] Sukthankar, N., Maharnawar, S., Deshmukh, P., Haribhakta, Y., & Kamble, V.
(2017, July). nQuery-A natural language statement to SQL query generator. In Proceedings
of ACL 2017, Student Research Workshop (pp. 17-23)
[31] Montgomery, C. A. (1972, August). Is natural language an unnatural query language?
In Proceedings of the ACM annual conference-Volume 2 (pp. 1075-1078)
[32] Iqbal, R., Murad, M. A. A., Selamat, M. H., & Azman, A. (2012, March). Negation
query handling engine for natural language interfaces to ontologies. In 2012 International
Conference on Information Retrieval & Knowledge Management (pp. 249-253). IEEE.
[33] Mukherjee, P., Chattopadhyay, A., Chakraborty, B., & Nandi, D. (2021). Natural
language query handling using extended knowledge provider system. International Journal
of Knowledge-based and Intelligent Engineering Systems, 25(1), 1-19
[34] Huang, P. S., Wang, C., Singh, R., Yih, W. T., & He, X. (2018). Natural language to
structured query generation via meta-learning. arXiv preprint arXiv:1803.02400
[35] Small, D. W., & Weldon, L. J. (1983). An experimental comparison of natural and
structured query languages. Human Factors, 25(3), 253-263
[36] Koutrika, G., Simitsis, A., & Ioannidis, Y. E. (2010, March). Explaining structured
queries in natural language. In 2010 IEEE 26th International Conference on Data Engineering
(ICDE 2010) (pp. 333-344). IEEE
[37] Gur, I., Yavuz, S., Su, Y., & Yan, X. (2018, July). Dialsql: Dialogue based structured
query generation. In Proceedings of the 56th Annual Meeting of the Association for Computational
Linguistics (Volume 1: Long Papers) (pp. 1339-1349)
[38] Kaplan, S. J. (1984). Designing a portable natural language database query
system. ACM Transactions on Database Systems (TODS), 9(1), 1-19
[39] Yaghmazadeh, N., Wang, Y., Dillig, I., & Dillig, T. (2017). SQLizer: query synthesis
from natural language. Proceedings of the ACM on Programming Languages, 1(OOPSLA), 1-
26
[40] Yaghmazadeh, N., Wang, Y., Dillig, I., & Dillig, T. (2017). SQLizer: query synthesis
from natural language. Proceedings of the ACM on Programming Languages, 1(OOPSLA), 1-
26
[41] Androutsopoulos, I., Ritchie, G. D., & Thanisch, P. (1995). Natural language
interfaces to databases–an introduction. Natural language engineering, 1(1), 29-81
[42] Kate, A., Kamble, S., Bodkhe, A., & Joshi, M. (2018, March). Conversion of natural
language query to SQL query. In 2018 Second International Conference on Electronics,
Communication and Aerospace Technology (ICECA) (pp. 488-491). IEEE.

Feb 2022 | Vol 4|Issue 1 Page | 161

 International Journal of Innovations in Science & Technology
[43] Song, Y., Wong, R. C. W., Zhao, X., & Jiang, D. (2022). Speech-to-SQL: Towards
Speech-driven SQL Query Generation from Natural Language Question. arXiv preprint
arXiv:2201.01209.
[44] Sujatha, B., & Raju, S. V. (2014). A Flexible and Efficient Natural Language Query
interface to databases. International Journal of Computer Science and Information
Technologies, 5(5), 6464-6467.
[45] Dekleva, S. M. (1994). Is natural language querying practical? ACM SIGMIS
Database: the DATABASE for Advances in Information Systems, 25(2), 24-36.
[46] Narechania, A., Fourney, A., Lee, B., & Ramos, G. (2021, April). DIY: Assessing
the correctness of natural language to sql systems. In 26th International Conference on
Intelligent User Interfaces (pp. 597-607).
[47] Blanning, R. W. (1986). A System for natural language communication between a
decision model and its users. IFAC Proceedings Volumes, 19(17), 77-85.
[48] Amble, T. (2000, April). BusTUC-a natural language bus route oracle. In Sixth
Applied Natural Language Processing Conference (pp. 1-6).
[49] Narechania, A., Fourney, A., Lee, B., & Ramos, G. (2021, April). DIY: Assessing
the correctness of natural language to sql systems. In 26th International Conference on
Intelligent User Interfaces (pp. 597-607).
[50] Zhang, X., Cheng, G., & Qu, Y. (2007). Ontology summarization based on rdf
sentence graph. WWW '07.

Copyright © by authors and 50Sea. This work is licensed under

Creative Commons Attribution 4.0 International License.

Feb 2022 | Vol 4|Issue 1 Page | 162

View publication stats