Hate speech is any kind of communication that attacks a person or a group based on their characteristics, such as gender, religion and race. Due to the availability of online platforms where people can express their (hateful) opinions, the amount of hate speech is steadily increasing that often leads to offline hate crimes. This paper focuses on understanding and detecting hate speech in underground hacking and extremist forums where cybercriminals and extremists, respectively, communicate with each other, and some of them are associated with criminal activity. Moreover, due to the lengthy posts, it would be beneficial to identify the specific span of text containing hateful content in order to assist site moderators with the removal of hate speech. This paper describes a hate speech dataset composed of posts extracted from HackForums, an online hacking forum, and Stormfront and Incels.co, two extremist forums. We combined our dataset with a Twitter hate speech dataset to train a multi-platform classifier. Our evaluation shows that a classifier trained on multiple sources of data does not always improve the performance compared to a mono-platform classifier. Finally, this is the first work on extracting hate speech spans from longer texts. The paper fine-tunes BERT (Bidirectional Encoder Representations from Transformers) and adopts two approaches – span prediction and sequence labelling. Both approaches successfully extract hateful spans and achieve an F1-score of at least 69%.

In the past few years, high-quality automated text-to-speech synthesis has effectively become a commodity, with easy access to cloud-based APIs provided by a number of major players. At the same time, developments in deep learning have broadened the scope of voice synthesis functionalities that can be delivered, leading to a growth in the range of commercially viable use cases. We take a look at the technology features and use cases that have attracted attention and investment in the past few years, identifying the major players and recent start-ups in the space.

The OffensEval shared tasks organized as part of SemEval-2019–2020 were very popular, attracting over 1300 participating teams. The two editions of the shared task helped advance the state of the art in offensive language identification by providing the community with benchmark datasets in Arabic, Danish, English, Greek, and Turkish. The datasets were annotated using the OLID hierarchical taxonomy, which since then has become the de facto standard in general offensive language identification research and was widely used beyond OffensEval. We present a survey of OffensEval and related competitions, and we discuss the main lessons learned. We further evaluate the performance of Large Language Models (LLMs), which have recently revolutionalized the field of Natural Language Processing. We use zero-shot prompting with six popular LLMs and zero-shot learning with two task-specific fine-tuned BERT models, and we compare the results against those of the top-performing teams at the OffensEval competitions. Our results show that while some LMMs such as Flan-T5 achieve competitive performance, in general LLMs lag behind the best OffensEval systems.

Lots of companies are offering lots of APIs. Reviews are not always as constructive as they could be. Some reviews encourage unproductive work on checkbox features that no one wants. It makes no sense to do the wrong thing badly. Constructive reviews should help focus priorities on what matters. Users care more about a great box opening experience than small improvements in word error rate and BLEU, popular metrics for speech and translation. In 15 minutes or less, can we teach potential users something new (and fun) that does something useful, such as how to translate PowerPoint between English and Chinese, preserving many of the features that are important to PowerPoint such as graphics and animations? See Appendix for the solution.

By increasing the amount of data in computer networks, searching and finding suitable information will be harder for users. One of the most widespread forms of information on such networks are textual documents. So exploring these documents to get information about their content is difficult and sometimes impossible. Multi-document text summarization systems are an aid to producing a summary with a fixed and predefined length, while covering the maximum content of the input documents. This paper presents a novel method for multi-document extractive summarization based on textual entailment relations and sentence compression via formulating the problem as a knapsack problem. In this approach, sentences of documents are ranked according to the extended Tf-Idf method, then entailment scores of selected sentences are computed. Through these scores, the final score of each sentence is calculated. Finally, by decreasing the lengths of sentences via sentence compression, the problem has been solved by greedy and dynamic Programming approaches to the knapsack problem. Experiments on standard summarization datasets and evaluating the results based on the Rouge system show that the suggested method, according to the best of our knowledge, has increased F-measure of query-based summarization systems by two per cent and F-measure of general summarization systems by five per cent.

Verbmobil represents a new generation of speech-to-speech translation systems in which spontaneously spoken language, speaker independence and adaptability as well as the combination of deep and shallow approaches to the analysis and transfer problems are the main features. The project brought together researchers from the fields of signal processing, computational linguistics and artificial intelligence. Verbmobil goes beyond the state-of-the-art in each of these areas, but its main achievement is the seamless integration of them. The first project phase (1993–1996) has been followed up by the second project phase (1997–2000), which aims at applying the results to further languages and at integrating innovative telecooperation techniques. Quite apart from the speech and language processing issues, the size and complexity of the project represent an extreme challenge on the areas of project management and software engineering:

[bull ] 50 researchers from 29 organizations at different sites in different countries are involved in the software development process,

[bull ] to reuse existing software, hardware, knowledge and experience, only a few technical restrictions could be given to the partners.

In this article we describe the Verbmobil prototype system from a software-engineering perspective. We discuss:

[bull ] the modularized functional architecture,

[bull ] the flexible and extensible software architecture which reflects that functional architecture,

[bull ] the evolutionary process of system integration,

[bull ] the communication-based organizational structure of the project,

[bull ] the evaluation of the system operational by the end of the first project phase.

Humans proficiently interpret the true sense of an ambiguous word by establishing association among words in a sentence. The complete sense of text is also based on implicit information, which is not explicitly mentioned. The absence of this implicit information is a significant problem for a computer program that attempts to determine the correct sense of ambiguous words. In this paper, we propose a novel method to uncover the implicit information that links the words of a sentence. We reveal this implicit information using a graph, which is then used to disambiguate the ambiguous word. The experiments show that the proposed algorithm interprets the correct sense for both homonyms and polysemous words. Our proposed algorithm has performed better than the approaches presented in the SemEval-2013 task for word sense disambiguation and has shown an accuracy of 79.6 percent, which is 2.5 percent better than the best unsupervised approach in SemEval-2007.

This paper proposes a robust text classification and correspondence analysis approach to identification of similar languages. In particular, we propose to use the readily available information of clauses and word length distribution to model similar languages. The modeling and classification are based on the hypothesis that languages are self-adaptive complex systems and hence can be classified by dynamic features describing the system, especially in terms of distributional relations of constituents of a system. For similar languages whose grammatical differences are often subtle, classification based on dynamic system features should be more effective. To test this hypothesis, we considered both regional and genre varieties of Mandarin Chinese for classification. The data are extracted from two comparable balanced corpora to minimize possible confounding factors. The two corpora are the Sinica Corpus from Taiwan and the Lancaster Corpus of Mandarin Chinese from Mainland China, and the two genres are reportage and review. Our text classification and correspondence analysis results show that the linguistically felicitous two-level constituency model combining power functions between word and clauses effectively classifies the two varieties of Chinese for both genres. In addition, we found that genres do have compounding effect on classification of regional varieties. In particular, reportage in two varieties is more likely to be classified than review, corroborating the complex system view of language variations. That is, language variations and changes typically do not take place evenly across the board for the complete language system. This further enhances our hypothesis that dynamic complex system features, such as the power functions captured by the Menzerath–Altmann law, provide effective models in classifications of similar languages.

Deep nets have done well with early adopters, but the future will soon depend on crossing the chasm. The goal of this paper is to make deep nets more accessible to a broader audience including people with little or no programming skills, and people with little interest in training new models. A github is provided with simple implementations of image classification, optical character recognition, sentiment analysis, named entity recognition, question answering (QA/SQuAD), machine translation, speech to text (SST), and speech recognition (STT). The emphasis is on instant gratification. Non-programmers should be able to install these programs and use them in 15 minutes or less (per program). Programs are short (10–100 lines each) and readable by users with modest programming skills. Much of the complexity is hidden behind abstractions such as pipelines and auto classes, and pretrained models and datasets provided by hubs: PaddleHub, PaddleNLP, HuggingFaceHub, and Fairseq. Hubs have different priorities than research. Research is training models from corpora and fine-tuning them for tasks. Users are already overwhelmed with an embarrassment of riches (13k models and 1k datasets). Do they want more? We believe the broader market is more interested in inference (how to run pretrained models on novel inputs) and less interested in training (how to create even more models).

Research on question answering dates back to the 1960s but has more recently been revisited as part of TREC's evaluation campaigns, where question answering is addressed as a subarea of information retrieval that focuses on specific answers to a user's information need. Whereas document retrieval systems aim to return the documents that are most relevant to a user's query, question answering systems aim to return actual answers to a users question. Despite this difference, question answering systems rely on information retrieval components to identify documents that contain an answer to a user's question. The computationally more expensive answer extraction methods are then applied only to this subset of documents that are likely to contain an answer. As information retrieval methods are used to filter the documents in the collection, the performance of this component is critical as documents that are not retrieved are not analyzed by the answer extraction component. The formulation of queries that are used for retrieving those documents has a strong impact on the effectiveness of the retrieval component. In this paper, we focus on predicting the importance of terms from the original question. We use model tree machine learning techniques in order to assign weights to query terms according to their usefulness for identifying documents that contain an answer. Term weights are learned by inspecting a large number of query formulation variations and their respective accuracy in identifying documents containing an answer. Several linguistic features are used for building the models, including part-of-speech tags, degree of connectivity in the dependency parse tree of the question, and ontological information. All of these features are extracted automatically by using several natural language processing tools. Incorporating the learned weights into a state-of-the-art retrieval system results in statistically significant improvements in identifying answer-bearing documents.

This issue of Natural Language Engineering journal reports on recent achievements in the domain of HPSG-based parsing. Research groups at Saarbrücken, CSLI Stanford and the University of Tokyo have worked on grammar development and processing systems that allow the use of HPSG-based processing in practical application contexts. Much of the research reported here has been collaborative, and all of the work shares a commitment to producing comparable results on wide-coverage grammars with substantial test suites. The focus of this special issue is deliberately narrow, to allow detailed technical reports on the results obtained among the collaborating groups. Thus, the volume cannot aim at providing a complete survey on the current state of the field. This introduction summarizes the research background for the work reported in the issue, and puts the major new approaches and results into perspective. Relationships to similar efforts pursued elsewhere are included, along with a brief summary of the research and development efforts reflected in the volume, the joint reference grammar, and the common sets of reference data.

This paper describes a novel approach to multi-document summarization, which explicitly addresses the problem of detecting, and retaining for the summary, multiple themes in document collections. We place equal emphasis on the processes of theme identification and theme presentation. For the former, we apply Iterative Residual Rescaling (IRR); for the latter, we argue for graphical display elements. IRR is an algorithm designed to account for correlations between words and to construct multi-dimensional topical space indicative of relationships among linguistic objects (documents, phrases, and sentences). Summaries are composed of objects with certain properties, derived by exploiting the many-to-many relationships in such a space. Given their inherent complexity, our multi-faceted summaries benefit from a visualization environment. We discuss some essential features of such an environment.

A part-of-speech tagger is a fundamental and indispensable tool in computational linguistics, typically employed at the critical early stages of processing. Although taggers are widely available that achieve high accuracy in very general domains, these do not perform nearly as well when applied to novel specialized domains, and this is especially true with biological text. We present a stochastic tagger that achieves over 97.44% accuracy on MEDLINE abstracts. A primary component of the tagger is its lexicon which enumerates the permitted parts-of-speech for the 10000 words most frequently occurring in MEDLINE. We present evidence for the conclusion that the lexicon is as vital to tagger accuracy as a training corpus, and more important than previously thought.

We show how the DOP model can be used for fast and robust context-sensitive processing of spoken input in a practical spoken dialogue system called OVIS. OVIS (Openbaar Vervoer Informatie Systeem) – ‘Public Transport Information System’, is a Dutch spoken language information system which operates over ordinary telephone lines. The prototype system is the immediate goal of the NWO Priority Programme ‘Language and Speech Technology’. In this paper, we extend the original Data-Oriented Parsing (DOP) model to context-sensitive interpretation of spoken input. The system we describe uses the OVIS corpus (which consists of 10,000 trees enriched with compositional semantics) to compute from an input word-graph the best utterance together with its meaning. Dialogue context is taken into account by dividing up the OVIS corpus into context-dependent subcorpora. Each system question triggers a subcorpus by which the user answer is analysed and interpreted. Our experiments indicate that the context-sensitive DOP model obtains better accuracy than the original model, allowing for fast and robust processing of spoken input.

Some languages have very few NLP resources, while many of them are closely related to better-resourced languages. This paper explores how the similarity between the languages can be utilised by porting resources from better- to lesser-resourced languages. The paper introduces a way of building a representation shared across related languages by combining cross-lingual embedding methods with a lexical similarity measure which is based on the weighted Levenshtein distance. One of the outcomes of the experiments is a Panslavonic embedding space for nine Balto-Slavonic languages. The paper demonstrates that the resulting embedding space helps in such applications as morphological prediction, named-entity recognition and genre classification.

Transformation-Based Learning (TBL) is a relatively new machine learning method that has achieved notable success on language problems. This paper presents a variant of TBL, called Randomized TBL, that overcomes the training time problems of standard TBL without sacrificing accuracy. It includes a set of experiments on part-of-speech tagging in which the size of the corpus and template set are varied. The results show that Randomized TBL can address problems that are intractable in terms of training time for standard TBL. In addition, for language problems such as dialogue act tagging where the most effective features have not been identified through linguistic studies, Randomized TBL allows the researcher to experiment with a large set of templates capturing many potentially useful features and feature interactions.

Electronic written texts used in computer-mediated interactions (emails, blogs, chats, and the like) contain significant deviations from the norm of the language. This paper presents the detail of a system aiming at normalizing the orthography of French SMS messages: after discussing the linguistic peculiarities of these messages and possible approaches to their automatic normalization, we present, compare, and evaluate various instanciations of a normalization device based on weighted finite-state transducers. These experiments show that using an intermediate phonemic representation and training, our system outperforms an alternative normalization system based on phrase-based statistical machine translation techniques.

This paper describes CLIME, a web-based legal advisory system with a multilingual natural language interface. CLIME is a ‘proof-of-concept’ system which answers queries relating to ship-building and ship-operating regulations. Its core knowledge source is a set of such regulations encoded as a conceptual domain model and a set of formalised legal inference rules. The system supports retrieval of regulations via the conceptual model, and assessment of the legality of a situation or activity on a ship according to the legal inference rules. The focus of this paper is on the natural language aspects of the system, which help the user to construct semantically complex queries using WYSIWYM technology, allow the system to produce extended and cohesive responses and explanations, and support the whole interaction through a hybrid synchronous/asynchronous dialogue structure. Multilinguality (English and French) is viewed simply as interface localisation: the core representations are language-neutral, and the system can present extended or local interactions in either language at any time. The development of CLIME featured a high degree of client involvement, and the specification, implementation and evaluation of natural language components in this context are also discussed.

We describe the natural language processing and knowledge representation components of B2, a collaborative system that allows medical students to practice their decision-making skills by considering a number of medical cases that differ from each other in a controlled manner. The underlying decision-support model of B2 uses a Bayesian network that captures the results of prior clinical studies of abdominal pain. B2 generates story-problems based on this model and supports natural language queries about the conclusions of the model and the reasoning behind them. B2 benefits from having a single knowledge representation and reasoning component that acts as a blackboard for intertask communication and cooperation. All knowledge is represented using a propositional semantic network formalism, thereby providing a uniform representation to all components. The natural language component is composed of a generalized augmented transition network parser/grammar and a discourse analyzer for managing the natural language interactions. The knowlege representation component supports the natural language component by providing a uniform representation of the content and structure of the interaction, at the parser, discourse, and domain levels. This uniform representation allows distinct tasks, such as dialog management, domain-specific reasoning, and meta-reasoning about the Bayesian network, to all use the same information source, without requiring mediation. This is important because there are queries, such as Why?, whose interpretation and response requires information from each of these tasks. By contrast, traditional approaches treat each subtask as a “black-box” with respect to other task components, and have a separate knowledge representation language for each. As a result, they have had much more difficulty providing useful responses.

In this paper we show how established object modelling techniques can be used in the creation of spoken dialogue management systems. One of the motivations behind the particular approach adopted here is the observation that, in spoken human-to-human dialogues, certain skillsets and patterns of dialogue evolution are common to many different contexts; other dialogue skills and accompanying real-world knowledge are required only for more specialised transactions within particular business domains. As a starting point for modelling an automated spoken dialogue management system we recommend a use case analysis of the required functionality. The use case analysis encourages the developer to identify generic-specific relationships and interactions between different dialogue management skills. We consider some of the broad philosophies underlying current dialogue management systems and outline practical high-level dialogue behaviour based on mixed-initiative, frame-based processing, combined with a rigorously applied confirmation strategy. On the basis of the use case requirements analysis, we explore a possible design for an object-oriented dialogue management system, indicating the roles and relationships of the various classes that embody the required dialogue functionality, and showing how implemented objects within the system will interact. The manner of this interaction is such as to allow one overall system to process transactions in several business domains. We also indicate some of the advantages of a rule-based implementation: the proposed design is tailored towards such an implementation in Prolog++. An object-oriented development process places high-level, generic dialogue management functionality at the disposal of more specialised ‘expert’ components. Maintainability and extensibility are therefore enhanced: if the developer chooses to refine generic behaviour, it is immediately available to the more specialised components; if new domain-specific expertise is required, it can be added with minimal impact on generic behaviour.