Feature Analysis for Paraphrase Recognition and Textual Entailment

In this work, we present a new unsupervised and language-independent methodology to detect the relations of textual generality. For this, we introduce a particular case of Textual Entailment (TE), namely Textual Entailment by Generality (TEG). TE aims to capture primary semantic inference needs across applications in Natural Language Processing (NLP). Since 2005, in the TE Recognition (RTE) task, systems have been asked to automatically judge whether the meaning of a portion of the text, the Text (T), entails the meaning of another text, the Hypothesis (H). Several novel approaches and improvements in TE technologies demonstrated in RTE Challenges are signaling renewed interest towards a more in-depth and better understanding of the core phenomena involved in TE. In line with this direction, in this work, we focus on a particular case of entailment, entailment by generality, to detect the relations of textual generality. In text, there are different kinds of entailments, yielded from different types of implicative reasoning (lexical, syntactical, common sense based), but here, we focus just on TEG, which can be defined as an entailment from a specific statement towards a relatively more general one. Therefore, we have T→GH whenever the premise T entails the hypothesis H, this also being more general than the premise. We propose an unsupervised and language-independent method to recognize TEGs, from a pair ⟨T,H⟩ having an entailment relation. To this end, we introduce an Informative Asymmetric Measure (IAM) called Simplified Asymmetric InfoSimba (AISs), which we combine with different Asymmetric Association Measures (AAM). In this work, we hypothesize about the existence of a particular mode of TE, namely TEG. Thus, the main contribution of our study is highlighting the importance of this inference mechanism. Consequently, the new annotation data seem to be a valuable resource for the community.

In this paper we gauge the utility of general-purpose, open-domain semantic parsing for textual entailment recognition by combining graph-structured meaning representations with semantic technologies and formal reasoning tools. Our approach achieves high precision, and in two case studies we show that when reasoning over n-best analyses from the parser the performance of our system reaches stateof-the-art for rule-based textual entailment systems. 1 Background and Motivation There is a growing interest in recent years in general-purpose semantic parsing into graphbased meaning representations, which provide greater expressive power than tree-based structures. Recent efforts in this spirit include, for example, Abstract Meaning Representation (Banarescu et al., 2013), and Semantic Dependency Parsing (SDP) (Oepen et al., 2014; Oepen et al., 2015). Simultaneously, in the Semantic Web community, a range of generic semantic technologies for storing and processing graph-structured data has been made available, but these have not been much used for natural language processing tasks. We propose a flexible, generic framework for precision-oriented Textual Entailment (TE) recognition that combines semantic parsing, graph-based representations of sentence meaning, and semantic technologies. During the decade since the TE task was defined, (logical) inference-based approaches have made some important contributions to the field. Systems such as Bos and Markert (2006) and Tatu and Moldovan (2006) employ automated proof search over logical representations of the input sentences. Other systems, such as Bar-Haim et al. (2007), apply transformational rules to linguistic representations of the sentence pairs, and determine entailment through graph subsumption. Because inference-based systems are vulnerable to incomplete knowledge in the rule set and errors in the mapping from natural language sentences to logical forms or linguistics representations, and because the definition of the TE task encourages a more relaxed, non-logical notion of entailment, the majority of TE systems have used more robust approaches, however. Our work supports a notion of logical inference for TE by reasoning with formal rules over graph-structured meaning representations, while achieving results that are comparable with robust approaches. We use a freely available, grammar-driven semantic parser and a well-defined reduction of underspecified logical-form meaning representations into variable-free semantic graphs called Elementary Dependency Structures (EDS) (Oepen and Lonning, 2006). We capitalize on a pre-existing storage and search infrastructure for EDSs using generic semantic technologies. For entailment classification, we create inference rules that enrich the EDS graphs, apply the rules with a generic reasoner, and use graph alignment as a decision tool. To test our generic setup, we perform two case studies where we replicate well-performing TE systems, one from the Parser Evaluation using Textual Entailments (PETE) task (Yuret et al., 2010), and one from SemEval 2014 Task 1 (Marelli et al., 2014). The best published results for the PETE task, Lien (2014), were obtained through heuristic rules that align meaning representations based on structural similarity. Lien and Kouylekov (2014) extend the same basic approach for SemEval 2014 by including lexical relations and negation handling. We recast the handwritten heuristic rules from these systems as formal Semantic Web Rule Language (SWRL) rules, and run them with a generic reasoning tool over EDS

In this study, we explore the potential of machine learning (ML) models after different text representation methods on the balanced IMDB dataset, which is widely regarded as a gold standard in sentiment analysis, one of the Natural Language processing (NLP) tasks. On the open source IMDB movie reviews dataset, we first undertake data cleaning and text representation with data preprocessing steps. Then, we apply sentiment classification using different ML models. In order to evaluate the models, we used precision (P), recall (R), F1-score (F1), and area under curve (AUC), as well as receiver operating characteristic (ROC). It is worth noting that text feature extraction with Bidirectional Encoder Representations from Transformers (BERT) provided the highest performance in all models, with the SVM model offering particularly promising results. In this model, we observed the following results: ACC 0.9033, F1 0.9308, R 0.9015, R 0.9015, P 0.9072, AUC 0.9638, and ROC 0.96. These findings suggest that NLP techniques and, in particular, machine learning models that employ BERT may offer high levels of accuracy and reliability in text classification problems. It would be beneficial for future studies to validate these findings using BERT on different NLP tasks. This would help to evaluate the effectiveness and applicability of the models in practice.

Textual entailment is a fundamental task in natural language processing. Most approaches for solving this problem use only the textual content present in training data. A few approaches have shown that information from external knowledge sources like knowledge graphs (KGs) can add value, in addition to the textual content, by providing background knowledge that may be critical for a task. However, the proposed models do not fully exploit the information in the usually large and noisy KGs, and it is not clear how it can be effectively encoded to be useful for entailment. We present an approach that complements text-based entailment models with information from KGs by (1) using Personalized PageRank to generate contextual subgraphs with reduced noise and (2) encoding these subgraphs using graph convolutional networks to capture the structural and semantic information in KGs. We evaluate our approach on multiple textual entailment datasets and show that the use of external knowledge helps the model to be robust and improves prediction accuracy. This is particularly evident in the challenging BreakingNLI dataset, where we see an absolute improvement of 5-20% over multiple text-based entailment models.

Paraphrase and textual entailment generation can support natural language processing (NLP) tasks that simulate text understanding, e.g., text summariza- tion, plagiarism detection, or question answering. A paraphrase, i.e., a sentence with the same meaning, conveys a certain piece of information with new words and new syntactic structures. Textual entailment, i.e., an inference that humans will judge most likely true, can employ real-world knowledge in order to make some implicit information explicit. Paraphrases can also be seen as mutual entailments. We present a new system that generates paraphrases and textual entailments from a given text in the Czech language. First, the process is rule-based, i.e., the system analyzes the input text, pro- duces its inner representation, transforms it according to transformation rules, and generates new sentences. Second, the generated sentences are ranked according to a statistical model and only the best ones are output. The decision whether a paraphrase or textual entailment is correct or not is left to humans. For this purpose we designed an annotation game based on a conversation between a detective (the human player) and his assis- tant (the system). The result of such annotation is a collection of annotated pairs text-hypothesis. Currently, the system and the game are intended to collect data in the Czech language. However, the idea can be applied for other languages. So far, we have collected 3,321 H-T pairs. From these pairs, 1,563 were judged cor- rect (47.06 %), 1,238 (37.28 %) were judged incorrect entailments, and 520 (15.66 %) were judged non-sense or unknown.

The advances in deep learning have led to great achievements in many Natural Language Processing (NLP) tasks. With the nature of language, i.e., sequential data, most NLP tasks can be framed into the sequence learning framework, such as text generation. As one of the most important foundations for modern NLP techniques, autoregressive generation models have achieved dominant performance in a great deal of NLP tasks. Therefore, this thesis emphasizes improving the autoregressive generation model for different NLP tasks. While many tasks can naturally fit into the sequence learning framework, some of them, e.g., building discourse parsing tree, require sophisticated designs to fit into neural models. Therefore, this thesis firstly emphasizes a novel unified framework for discourse parsing, which builds a discourse tree in a top-down depth-first manner, and it frames the task as an autoregressive generation task with the goal of each step being the prediction of the node position given a piece of text. The proposed approach is proven effective with extensive empirical experiments. In addition, I extend the above framework by proposing a hierarchical decoder, which leverages the information from parents and siblings of the nodes that are currently processed. The proposed decoder utilizes the nature of the tree structure and further improves the experiment performance on both discourse parsing and dependency parsing tasks. On the other hand, the de facto strategies, i.e., cross entropy loss and teacher forcing, for training the autoregressive generation models have been shown problematic in certain aspects. For example, cross entropy loss, which is one of the widely leveraged training objective functions, often leads to text degeneration in text generation, and teacher forcing suffers from the exposure bias problem, where there exists a mismatch between the training and testing setup. For text degeneration, I introduce a class of diminishing attentions, which enforces the submodularity of the coverage calculated by cross attention in the sequence-to-sequence model. The proposed diminishing attentions achieve notable improvement on several neural text generation tasks, including text summarization, machine translation, and image paragraph generation. Further, I propose a novel training objective, ScaleGrad, to replace cross entropy, which significantly reduces the degeneration problem in different text generation tasks. In fact, ScaleGrad can be extended to problems beyond text degeneration. It provides wide flexibility to inject different inductive biases into the text generation model by directly modifying the gradient information in the output layer. Next, for the exposure bias problem, this thesis introduces a novel type of scheduled sampling based on training accuracy, which requires only minimal hyper-parameter tuning compared to existing scheduled sampling methods. Additionally, a novel imitation loss is proposed to further enforce the model’s generative behavior to match the teacher-forced behavior. Moreover, this thesis demonstrates that reducing exposure bias can improve the robustness of language models against repetition and toxic errors.

Semantic Role Labeling (SRL) is a crucial task in Natural Language Processing (NLP) that plays a vital role in extracting meaningful information from text. In the fields of economics and cybersecurity, accurately identifying and analyzing semantic roles within text is crucial due to the rapid increase in the amount and complexity of textual information. This abstract examines the significant role of SRL and its application in economic and cybersecurity contexts. It discusses the state-of-the-art NLP classifiers used for this purpose. By examining the relationship between language processing and these important areas, we aim to emphasize the importance of SRL in extracting useful information and improving decision-making in a constantly changing digital environment. The aim of the findings is to emphasize the significance of SRL in extracting valuable insights from text, as it serves as a fundamental technique in NLP. It is utilized in the economic context to analyze financial reports, news articles, and economic texts. It assists in decision-making and market analysis. It aids in identifying important participants, actions, and objects in economic discourse, leading to better decision-making and market analysis. In the field of cybersecurity, SRL assists parse and comprehending text data related to security, enabling faster responses to threats. NLP classifiers and machine learning models utilize SRL to automate the analysis of large amounts of text. These techniques are practically significant as they improve the ability to extract actionable insights, assess risks, and make informed decisions by organizing unstructured text data. The process of determining relevant information from a large corpus of data requires an optimal methodological basis. Relevant textual data is collected from sources such as financial reports, news articles, or cybersecurity incident reports. Textual data is cleaned, tokenized, and tagged with part-of-speech labels in preparation for NLP analysis. Human annotators label semantic roles in the text, identifying actors, actions, and objects. This creates a dataset that can be used to train classifiers. NLP classifiers, including machine learning models, are trained using annotated datasets to identify semantic roles. The accuracy and performance of the trained classifiers are evaluated using various metrics. NLP classifiers are used to automatically identify and label semantic roles in new, unseen textual data. The output helps extract insights, such as market trends or security threats, depending on the specific field. Researchers improve classifier models by iteratively training and applying them to increase accuracy.

Recognizing textual entailment comprises the task of determining semantic entailment relations between text fragments. A text fragment entails another text fragment if, from the meaning of the former, one can infer the meaning of the latter. If such relation is bidirectional, then we are in the presence of a paraphrase. Automatically recognizing textual entailment relations captures major semantic inference needs in several natural language processing (NLP) applications. As in many NLP tasks, textual entailment corpora for English abound, while the same is not true for more resource-scarce languages such as Portuguese. Exploiting what seems to be the only Portuguese corpus for textual entailment and paraphrases (the ASSIN corpus), in this paper, we address the task of automatically recognizing textual entailment (RTE) and paraphrases from text written in the Portuguese language, by employing supervised machine learning techniques. We employ lexical, syntactic and semantic features, and analyze the impact of using semantic-based approaches in the performance of the system. We then try to take advantage of the bi-dialect nature of ASSIN to compensate its limited size. With the same aim, we explore modeling the task of recognizing textual entailment and paraphrases as a binary classification problem by considering the bidirectional nature of paraphrases as entailment relationships. Addressing the task as a multi-class classification problem, we achieve results in line with the winner of the ASSIN Challenge. In addition, we conclude that semantic-based approaches are promising in this task, and that combining data from European and Brazilian Portuguese is less straightforward than it may initially seem. The binary classification modeling of the problem does not seem to bring advantages to the original multi-class model, despite the outstanding results obtained by the binary classifier for recognizing textual entailments.

Text extraction is an important part of natural language processing (NLP) tasks. Most NLP tasks like text classification, machine translation, text-to-speech, text-based language identification, text summarization, and named-entity recognition involve the use of textual data. Such data is limited for low-resourced languages making it difficult to experiment advanced NLP techniques on these languages. This paper presents a Python-based toolkit for text analysis and text extraction from different types of images, documents, and audio files. The toolkit is built as a library that has functions that can be imported and utilized for text extraction.

Trie-based rule processing for clinical NLP: A use-case study of n-trie, making the ConText algorithm more efficient and scalable.

Researches in Computational Linguistics (CL) and Natural Language Processing (NLP) have been increasingly dissociated from each other. Empirical techniques in NLP show good performances in some tasks when large amount of data (with annotation) are available. However, in order for these techniques to be adapted easily to new text types or domains, or for similar techniques to be applied to more complex tasks such as text entailment than POS taggers, parsers, etc., rational understanding of language is required. Engineering techniques have to be underpinned by scientific understanding. In this paper, taking grammar in CL and parsing in NLP as an example, we will discuss how to re-integrate these two research disciplines. Research results of our group on parsing are presented to show how grammar in CL is used as the backbone of a parser.KeywordsNatural Language ProcessingLexical EntryDerivation TreeElementary TreeComputational LinguisticThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Text similarity measures are used in multiple tasks such as plagiarism detection, information ranking and recognition of paraphrases and textual entailment. While recent advances in deep learning highlighted the relevance of sequential models in natural language generation, existing similarity measures do not fully exploit the sequential nature of language. Examples of such similarity measures include n-grams and skip-grams overlap which rely on distinct slices of the input texts. In this paper we present a novel text similarity measure inspired from a common representation in DNA sequence alignment algorithms. The new measure, called TextFlow, represents input text pairs as continuous curves and uses both the actual position of the words and sequence matching to compute the similarity value. Our experiments on 8 different datasets show very encouraging results in paraphrase detection, textual entailment recognition and ranking relevance.

Long Short-Term Memory (LSTM) networks are unique to exercise data in its memory cell with long-term memory as Natural Language Processing (NLP) tasks have inklings of intensive time and computational power due to their complex structures like magnitude language model Transformer required to pre-train and learn billions of data performing different NLP tasks. In this paper, a dynamic chaotic model is proposed for the objective of transforming neurons states in network with neural dynamic characteristics by restructuring LSTM as Chaotic Neural Oscillatory-Long-Short Term Memory (CNO-LSTM), where neurons in LSTM memory cells are weighed in substitutes by oscillatory neurons to speed up computational training of language model and improve text classification accuracy for real-world applications. From the implementation perspective, five popular datasets of general text classification including binary, multi classification and multi-label classification are used to compare with mainstream baseline models on NLP tasks. Results showed that the performance of CNO-LSTM, a simplified model structure and oscillatory neurons state in exercising different types of text classification tasks are above baseline models in terms of evaluation index such as Accuracy, Precision, Recall and F1. The main contributions are time reduction and improved accuracy. It achieved approximately 46.76% of the highest reduction training time and 2.55% accuracy compared with vanilla LSTM model. Further, it achieved approximately 35.86% in time reduction compared with attention model without oscillatory indicating that the model restructure has reduced GPU dependency to improve training accuracy.

Towards a Novel Weakly Supervised Joint Approach of Named Entity Recognition and Normalization for Noisy Text

<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Context:</i> User stories have been widely accepted as artifacts to capture the user requirements in agile software development. They are short pieces of texts in a semi-structured format that express requirements. Natural language processing (NLP) techniques offer a potential advantage in user story applications. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Objective:</i> Conduct a systematic literature review to capture the current state-of-the-art of NLP research on user stories. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Method:</i> The search strategy is used to obtain relevant papers from SCOPUS, ScienceDirect, IEEE Xplore, ACM Digital Library, SpringerLink, and Google Scholar. Inclusion and exclusion criteria are applied to filter the search results. We also use the forward and backward snowballing techniques to obtain more comprehensive results. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Results:</i> The search results identified 718 papers published between January 2009 to December 2020. After applying the inclusion/exclusion criteria and the snowballing technique, we identified 38 primary studies that discuss NLP techniques in user stories. Most studies used NLP techniques to extract aspects of who, what, and why from user stories. The purpose of NLP studies in user stories is broad, ranging from discovering defects, generating software artifacts, identifying the key abstraction of user stories, and tracing links between model and user stories. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Conclusion:</i> NLP can help system analysts manage user stories. Implementing NLP in user stories has many opportunities and challenges. Considering the exploration of NLP techniques and rigorous evaluation methods is required to obtain quality research. As with NLP research in general, the ability to understand a sentence’s context continues to be a challenge.