An approach framework of transfer learning, adversarial training and hierarchical multi-task learning - a case study of disinformation detection with offensive text

With society going online and disinformation getting accepted as a phenomena that we have to live with, there is a growing need to automatically detect offensive text on modern social media platforms. But the lack of enough balanced labeled data, constantly evolving socio-linguistic patterns and ever-changing definition of offensive text make it a challenging task. This is a common pattern witnessed in all disinformation detection tasks such as detection of propaganda, rumour, fake news, hate etc. The work described in this paper improves upon the existing body of techniques by bringing in an approach framework that can surpass the existing benchmarks. Firstly, it addresses the imbalanced and insufficient nature of available labeled dataset. Secondly, learning using relates tasks through multi-task learning has been proved to be an effective approach in this domain but it has the unrealistic requirement of labeled data for all related tasks. The framework presented here suitably uses transfer learning in lieu of multi-task learning to address this issue. Thirdly, it builds a model explicitly addressing the hierarchical nature in the taxonomy of disinformation being detected as that delivers a stronger error feedback to the learning tasks. Finally, the model is made more robust by adversarial training. The work presented in this paper uses offensive text detection as a case study and shows convincing results for the chosen approach. The framework adopted can be easily replicated in other similar learning tasks facing a similar set of challenges.

The South Estonian language islands in the context of the Central Baltic area

This article offers a comparative analysis of several morphosyntactic and phonological features in the South Estonian language islands: Leivu, Lutsi, and Kraasna. The objective is to give an overview of the distribution of selected features, their (in)stability over time, and discuss their form and use in a broader areal context. To achieve this goal, comparative information was also included from the closest cognate varieties (Estonian and the South Estonian varieties, Courland Livonian and Salaca Livonian) and the main contact varieties (Latgalian, Latvian, and Russian). The data analysed in this study originated from various sources: text collections, dictionaries, and language corpora. The results reveal a multitude of linguistic patterns and distribution patterns, which means that the studied varieties are similar to / different from one another in various ways and points to multifaceted contact situations and outcomes in this area. Kokkuvõte. Miina Norvik, Uldis Balodis, Valts Ernštreits, Gunta Kļava, Helle Metslang, Karl Pajusalu, Eva Saar: Lõunaeesti keelesaared Kesk-Balti mõjuväljas. Artikkel esitab lõunaeesti keelesaarte – Leivu, Lutsi ja Kraasna – mitme morfosüntaktilise ja fonoloogilise joone võrdleva analüüsi. Uurimuse eesmärgiks on anda ülevaade valitud joonte levikust ja püsivusest ajas ning arutleda nende vormide ja kasutuse üle laiemas areaalses kontekstis. Selleks võetakse arvesse lähimate sugulaskeelte (eesti ja lõunaeesti, Kuramaa ja Salatsi liivi) ja -murrete ning tähtsamate kontaktkeelte (latgali, läti, vene) esinemusi. Analüüsitakse erinevatest allikatest, mh tekstikogudest, sõna- raamatutest ja keelekorpustest pärit ainest. Uurimistulemused toovad esile mitmesuguseid vormiseoseid ja muutuste levikuviise, osutades uuritud keelte ja murrete omavaheliste kontaktide mitmelaadsusele ning sellest tingitud erinevatele keelesüsteemi arengutele.

From the uncertainty of violence to life after abuse: discursive transitions among female survivors of IPV in online contexts

Intimate Partner Violence (IPV) is undoubtedly one of the most worrying concerns in today’s global societies. Due to the many intertwined factors that explain the persistence of this reality among people from all sorts of backgrounds, finding a uniform strategy to cope with this social issue is far from unproblematic. In this study, I contribute to a growing field of research that examines the discourse of female survivors of IPV in online contexts. The main objective this research pursues is to identify relevant linguistic patterns used by these women to represent themselves and their perpetrators in a publicly-available online forum. More specifically, I seek to ascertain discursive traits that characterise women in an initial stage in contrast to a final stage within an abusive relationship. To this end, I adopt a Corpus-Assisted Discourse Studies approach (CADS) in a digital corpus of circa 136,000 words, which are analysed with the software tool Sketch Engine (SkE). Findings show the most salient discursive traits that characterise IPV online discourse. Additionally, and drawing from verb patterns and their semantic categorisation, I highlight relevant verbal tendencies that connect linguistic textual evidence that contributes to sustaining the power imbalances that also define this social phenomenon.

Language, Social Networks, and the City

Research in Malaysian sociolinguistics has seen much development pertaining to its concerns over language in its multilingual, multiracial, post-colonial community. The majority of existing literature however tends to lean towards traditional ideologies to explicate the language situation and linguistic patterns taking place within society. As influential as they are, there is a growing need for research to extend and move beyond traditional parameters so as to better explicate the roles and values of language in the increasingly mobile, transnational, diverse communities found in the city. This is certainly true in the historical city of George Town, Penang where exists an eclectic mix of heritage and urbanity – a contest for fluid and fixed notions of identity, culture, traditions, and language. One approach which has been used to contribute towards the study of linguistic patterns is Social Network Analysis. A notable application of analysis network structures is attributed to Milroy (1987), where the following has emerged: close-knit and dense networks are resistant to outside influences whereas loose-knit, weaker network links are embracing of change. This paper therefore aims to explain Social Network Analysis as a framework and method, how it has been applied in previous studies, and the potential it holds to analyse language in contemporary, urban communities as is found in cities like George Town, Penang.

Narrative Performance and Sociopragmatic Abilities in Preschool Children are Linked to Multimodal Imitation Skills

Over recent decades much research has analyzed the relevance of 9- to 20- month-old infants’ early imitation skills (object- and language-based imitation) for language development. Yet there have been few systematic comparisons of the joint relevance of these imitative behaviors later on in development. This correlational study investigated whether multimodal imitation (gestural, prosodic, and lexical components) and object-based imitation are related to narratives and sociopragmatics in preschoolers. Thirty-one typically developing 3- to 4-year-old children performed four tasks to assess multimodal imitation, object-based imitation, narrative abilities, and sociopragmatic abilities. Results revealed that both narrative and sociopragmatic skills were significantly related to multimodal imitation, but not to object-based imitation, indicating that preschoolers’ ability to imitate socially relevant multimodal cues is strongly related to language and sociocommunicative skills. Therefore, this evidence supports a broader conceptualization of imitation behaviors in the field of language development that systematically integrates prosodic, gestural, and verbal linguistic patterns.

Linguistics and Deception Detection (DD): A Work in Progress

Linguistic Deception Detection DD is a well-established part of forensic linguistics and an area that continues to attract attention on the part of researchers, self-styled experts, and the public at large. In this article, the various approaches to DD within the general field of linguistics are examined. The basic method is to treat language as a form of behaviour and to equate marked linguistic behaviour with other marked forms of behaviour. Such a comparison has been identified in other fields such as psychology and kinesics as being associated with stress linked to the attempt to deceive, typically in such contexts as examined here. Representative authentic examples of some of the most common linguistic indicators of deception that have been identified are discussed, dividing them into two general categories which we here introduce: language as revealer and language as concealer. We will argue that linguistic analysis for DD should be conducted relative to the subject’s individual linguistic patterns of behaviour, not on absolutes related to broad generalisations about what is supposedly normal or unmarked in the population at large. We will also briefly discuss some structured methods for linguistic analysis for DD and the prospect that technology and artificial intelligence will provide the means to automate and digitalise the linguistic DD process. We maintain that caution is advisable when considering these, as DD will, in all probability, always remain a work in progress, with the need for a flexible human evaluator ready to take into account many different aspects of the individual subject and the case in question.

Translation of forms of address from Portuguese to Turkish through English: The case of José Saramago’s A Jangada de Pedra

The study of forms of address in translation is a type of register analysis that provides an interesting insight into the way specific linguistic patterns are transferred from one language to another. This article explores how the forms of address are rendered in the Turkish translation of A Jangada de Pedra (1986) by the Portuguese author José Saramago. Paratextual and textual analyses demon­strate that this work has been translated into Turkish through the English translation of the book, and that the English translation has influenced the choices of the Turkish translator. The findings of the study seem to support the hypothesis that using a mediating language/text that lacks similar forms of address as the ultimate source and the target languages/texts can cause shifts in tenor, which results in a different reading of interpersonal relationships between fictional characters in the target text.

Linguistic Patterns for Code Word Resilient Hate Speech Identification

The permanent transition to online activity has brought with it a surge in hate speech discourse. This has prompted increased calls for automatic detection methods, most of which currently rely on a dictionary of hate speech words, and supervised classification. This approach often falls short when dealing with newer words and phrases produced by online extremist communities. These code words are used with the aim of evading automatic detection by systems. Code words are frequently used and have benign meanings in regular discourse, for instance, “skypes, googles, bing, yahoos” are all examples of words that have a hidden hate speech meaning. Such overlap presents a challenge to the traditional keyword approach of collecting data that is specific to hate speech. In this work, we first introduced a word embedding model that learns the hidden hate speech meaning of words. With this insight on code words, we developed a classifier that leverages linguistic patterns to reduce the impact of individual words. The proposed method was evaluated across three different datasets to test its generalizability. The empirical results show that the linguistic patterns approach outperforms the baselines and enables further analysis on hate speech expressions.

Gratitude communication in academic written acknowledgement

In the existing literature, no attempt has been made to inspect how men and women rhetorically manage their gratitude communications in the academic written discourse. To bridge this knowledge gap, the present article examined how students of different gender construct their thanking acts in the acknowledgements of their M.A. theses. Discrepancies between male and female postgraduates’ employment of linguistic patterns and gratitude themes were compared. The results showed that student writers’ gratitude communications to a certain extent are conditioned by the conventional rhetorical patterns of the academic genre. Remarkable gender variations were evidenced in the students’ selections of lexical items for encoding the thanking expressions, thanking modifiers, and gratitude themes of their acknowledgements. These gender discrepancies in gratitude communications are highly pertinent to the social expectations of masculinity and femininity, the students’ psychological orientations toward the emotion of thanking and their own value priorities.

O milagre da juventude: uma análise linguística de exemplares de discurso publicitário da indústria antienvelhecimento

In this work, we aim to analyze how the cosmetic industry communicates the idea of aging and how it offers solutions for reversing this natural process, in the advertising speeches about face creams, aimed at the female audience. We built a corpus of 112 texts on cosmetic products, collected in beauty catalogs, websites of beauty brands and retail chains in the perfumery and cosmetics market, which have a set of texts in European Portuguese on the theme of “aging”. From a perspective of Lexical-Semantic analysis with contributions of the Discourse Analysis Pragmatics we tried to identify linguistic patterns to understand which discursive strategies were manipulated when advertising products promise to reverse the natural aging process. We observed that brands have products where the name is not always self-explanatory of the function it proposes and without great commitment to the possible solution, as aging and the respective physical marks are understood as the result of a process that the skin suffered and not the person.