A Survey of Paraphrasing and Textual Entailment Methods

By Ion Androutsopoulos, Prodromos Malakasiotis. 2009

485 citations recorded by the source corpus.

Paraphrasing methods recognize, generate, or extract phrases, sentences, or longer natural language expressions that convey almost the same information. Textual entailment methods, on the other hand, recognize, generate, or extract pairs of natural language expressions, such that a human who reads (and trusts) the first element of a pair would most likely infer that the other element is also true. Paraphrasing can be seen as bidirectional textual entailment and methods from the two areas are often similar. Both kinds of methods are useful, at least in principle, in a wide range of natural language processing applications, including question answering, summarization, text generation, and machine translation. We summarize key ideas from the two areas by considering in turn recognition, generation, and extraction methods, also pointing to prominent articles and resources.

Subfields: Artificial Intelligence, Natural Language Processing, Evaluation & Benchmarks

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Natural Language Processing

In Evaluation & Benchmarks

Citations in the corpus

Hypothesis Only Baselines in Natural Language Inference (2018, background): “, 2008) or manually engineered features (Androutsopoulos and Malakasiotis, 2010). Bowman et al. (2015) sparked a renewed interested in NLI, particularly among deep-learning researchers.”
SciTaiL: A Textual Entailment Dataset from Science Question Answering (2018, methodology): “Due to the small size of these datasets, most earlier approaches relied on hand-designed features and alignment systems (Androutsopoulos and Malakasiotis 2010).”
Adversarial Example Generation with Syntactically Controlled Paraphrase Networks (2018, methodology): “Paraphrase generation (Androutsopoulos and Malakasiotis, 2010; Madnani and Dorr, 2010) has been tackled using many different meth-ods, including those based on hand-crafted rules (McKeown, 1983), synonym substitu-tion (Bolshakov and Gelbukh, 2004), machine translation (Quirk et al., 2004), and,…”
Survey of the State of the Art in Natural Language Generation: Core tasks, applications and evaluation (2017, methodology): “In any case, doing full justice to recent developments in the various text-to-text generation applications is beyond the scope of a single survey, and many of these are covered in other surveys, including those by Mani (2001) and Nenkova and McKeown (2011) for summarisation; Androutsopoulos and Mala”
From Paraphrase Database to Compositional Paraphrase Model and Back (2015, background): “See Androutsopoulos and Malakasiotis (2010) for a survey on approaches for detecting paraphrases.”
ECNU: One Stone Two Birds: Ensemble of Heterogenous Measures for Semantic Relatedness and Textual Entailment (2014, methodology): “Like semantic relatedness, TE task (surveyed in (Androutsopoulos and Malakasiotis, 2009)) is also closely related to STS task since in TE task lots of similarity measures at different levels are exploited to boost classiﬁcation.”
Design and realization of a modular architecture for textual entailment (2013, methodology): “The focus of our work is to provide a general, open architecture for RTE that supports the implementation within a common platform of a wide range of approaches developed (Androutsopoulos and Malakasiotis 2010; Dagan, Roth and Zanzotto 2012).”
Semantic Uncertainty: Linguistic Invariances for Uncertainty Estimation in Natural Language Generation (2023, background): “The idea of formalising semantic equivalence via textual entailment has a long history in linguistics (Culicover, 1968) and NLP (Pad´o et al., 2009; Androutsopoulos & Malakasiotis, 2010).”
Statute Law Information Retrieval and Entailment (2019, methodology): “Due to the small size of these datasets, most earlier approaches relied on hand-designed features and alignment systems [8].”
Deep Learning for Text Style Transfer: A Survey (2020, methodology): “One of the common ways of paraphrasing is syntactic variation, such as “X wrote Y.”, “Y was written by X.”, and “X is the writer of Y.” (Androutsopoulos and Malakasiotis 2010).”
Detecting hallucinations in large language models using semantic entropy (2024, unknown): “We operationalize E ( ⋅ , ⋅ ) using the idea of bidirectional entailment, which has a long history in linguistics 52 and natural language processing 28,53,54 .”
Combining Similarity and Transformer Methods for Case Law Entailment (2019, background): “Early approaches for open-domain textual entailment relied heavily on exploiting surface syntax or lexical relationships [10], and then a wide range of tools, such as word embeddings, logical models, graphical models, rule systems and machine learning were applied [1].”
Stance and Sentiment in Tweets (2016, background): “It has received a lot of attention in the past decade, and we refer the reader to surveys [Androutsopoulos and Malakasiotis 2010; Dagan et al. 2013] and proceedings of recent challenges on recognizing textual entailment [Bentivogli et al. 2011; Marelli et al. 2014; Dzikovska et al. 2016].”