Data filtering strategies are a crucial component to develop safe Large Language Models (LLM), since they support the removal of harmful contents from pretraining datasets. There is a lack of research on the actual impact of these strategies on vulnerable groups to discrimination, though, and their effectiveness has not been yet systematically addressed. In this paper we present a benchmark study of data filtering strategies for harm reduction aimed at providing a systematic evaluation on these approaches. We provide an overview 55 technical reports of English LMs and LLMs to identify the existing filtering strategies in literature and implement an experimental setting to test their impact against vulnerable groups. Our results show that the positive impact that strategies have in reducing harmful contents from documents has the side effect of increasing the underrepresentation of vulnerable groups to discrimination in datasets.

Abstract Human users increasingly communicate with large language models (LLMs), but LLMs suffer from frequent overconfidence in their output, even when its accuracy is questionable, which undermines their trustworthiness and perceived legitimacy. Therefore, there is a need for language models to signal their confidence in order to reap the benefits of human-machine collaboration and mitigate potential harms. Verbalized uncertainty is the expression of confidence with linguistic means, an approach that integrates perfectly into language-based interfaces. Most recent research in natural language processing (NLP) overlooks the nuances surrounding human uncertainty communication and the biases that influence the communication of and with machines. We argue for anthropomimetic uncertainty, the principle that intuitive and trustworthy uncertainty communication requires a degree of imitation of human linguistic behaviors. We present a thorough overview of the research in human uncertainty communication, survey ongoing research in NLP, and perform additional analyses to demonstrate so-far underexplored biases in verbalized uncertainty. We conclude by pointing out unique factors in human-machine uncertainty and outlining future research directions towards implementing anthropomimetic uncertainty. kaleidophon/anthropomimetic-uncertainty

Most research on implicit discourse relation identification has focused on written language, however, it is also crucial to understand these relations in spoken discourse. We introduce a novel method for implicit discourse relation identification across both text and speech, that allows us to extract examples of semantically equivalent pairs of implicit and explicit discourse markers, based on aligning speech+transcripts with subtitles in another language variant. We apply our method to Egyptian Arabic, resulting in a novel high-quality dataset of spoken implicit discourse relations. We present a comprehensive approach to modeling implicit discourse relation classification using audio and text data with a range of different models. We find that text-based models outperform audio-based models, but combining text and audio features can lead to enhanced performance.

Social media are a central part of people’s lives. Unfortunately, many public social media spaces are rife with bullying and offensive language, creating an unsafe environment for their users. In this paper, we present a new dataset for offensive language detection in Albanian. The dataset is composed of user-generated comments on Facebook and YouTube from the channels of selected Kosovo news platforms. It is annotated according to the three levels of the OLID annotation scheme. We also show results of a baseline system for offensive language classification based on a fine-tuned BERT model and compare with the Danish DKhate dataset, which is similar in scope and size. In a transfer learning setting, we find that merging the Albanian and Danish training sets leads to improved performance for prediction on Danish, but not Albanian, on both offensive language recognition and distinguishing targeted and untargeted offence.

In this paper, we describe ParCorFull2.0, a parallel corpus annotated with full coreference chains for multiple languages, which is an extension of the existing corpus ParCorFull (Lapshinova-Koltunski et al., 2018). Similar to the previous version, this corpus has been created to address translation of coreference across languages, a phenomenon still challenging for machine translation (MT) and other multilingual natural language processing (NLP) applications. The current version of the corpus that we present here contains not only parallel texts for the language pair English-German, but also for English-French and English-Portuguese, which are all major European languages. The new language pairs belong to the Romance languages. The addition of a new language group creates a need of extension not only in terms of texts added, but also in terms of the annotation guidelines. Both French and Portuguese contain structures not found in English and German. Moreover, Portuguese is a pro-drop language bringing even more systemic differences in the realisation of coreference into our cross-lingual resources. These differences cause problems for multilingual coreference resolution and machine translation. Our parallel corpus with full annotation of coreference will be a valuable resource with a variety of uses not only for NLP applications, but also for contrastive linguists and researchers in translation studies.

Non-nominal co-reference is much less studied than nominal coreference, partly because of the lack of annotated corpora. We explore the possibility to exploit parallel multilingual corpora as a means of cheap supervision for the classification of three different readings of the English pronoun ‘it’: entity, event or pleonastic, from their translation in several languages. We found that the ‘event’ reading is not very frequent, but can be easily predicted provided that the construction used to translate the ‘it’ example is a pronoun as well. These cases, nevertheless, are not enough to generalize to other types of non-nominal reference.

Research on speaker-adapted neural machine translation (NMT) is scarce. One of the main challenges for more personalized MT systems is finding large enough annotated parallel datasets with speaker information. Rabinovich et al. (2017) published an annotated parallel dataset for EN–FR and EN–DE, however, for many other language pairs no sufficiently large annotated datasets are available.

We compare the performance of the APT and AutoPRF metrics for pronoun translation against a manually annotated dataset comprising human judgements as to the correctness of translations of the PROTEST test suite.Although there is some correlation with the human judgements, a range of issues limit the performance of the automated metrics.Instead, we recommend the use of semiautomatic metrics and test suites in place of fully automatic metrics.

Speakers of different languages must attend to and encode strikingly different aspects of the world in order to use their language correctly (Sapir, 1921;Slobin, 1996).One such difference is related to the way gender is expressed in a language.Saying "I am happy" in English, does not encode any additional knowledge of the speaker that uttered the sentence.However, many other languages do have grammatical gender systems and so such knowledge would be encoded.In order to correctly translate such a sentence into, say, French, the inherent gender information needs to be retained/recovered.The same sentence would become either "Je suis heureux", for a male speaker or "Je suis heureuse" for a female one.Apart from morphological agreement, demographic factors (gender, age, etc.) also influence our use of language in terms of word choices or even on the level of syntactic constructions (Tannen, 1991;Pennebaker et al., 2003).We integrate gender information into NMT systems.Our contribution is twofold: (1) the compilation of large datasets with speaker information for 20 language pairs, and (2) a simple set of experiments that incorporate gender information into NMT for multiple language pairs.Our experiments show that adding a gender feature to an NMT system significantly improves the translation quality for some language pairs.

Word segmentation is a low-level NLP task that is non-trivial for a considerable number of languages. In this paper, we present a sequence tagging framework and apply it to word segmentation for a wide range of languages with different writing systems and typological characteristics. Additionally, we investigate the correlations between various typological factors and word segmentation accuracy. The experimental results indicate that segmentation accuracy is positively related to word boundary markers and negatively to the number of unique non-segmental terms. Based on the analysis, we design a small set of language-specific settings and extensively evaluate the segmentation system on the Universal Dependencies datasets. Our model obtains state-of-the-art accuracies on all the UD languages. It performs substantially better on languages that are non-trivial to segment, such as Chinese, Japanese, Arabic and Hebrew, when compared to previous work.

In this paper, we address the problem of predicting one of three functions for the English pronoun 'it': anaphoric, event reference or pleonastic.This disambiguation is valuable in the context of machine translation and coreference resolution.We present experiments using a MAXENT classifier trained on gold-standard data and self-training experiments of an RNN trained on silver-standard data, annotated using the MAXENT classifier.Lastly, we report on an analysis of the strengths of these two models.

We present a character-based model for joint segmentation and POS tagging for Chinese. The bidirectional RNN-CRF architecture for general sequence tagging is adapted and applied with novel vector representations of Chinese characters that capture rich contextual information and lower-than-character level features. The proposed model is extensively evaluated and compared with a state-of-the-art tagger respectively on CTB5, CTB9 and UD Chinese. The experimental results indicate that our model is accurate and robust across datasets in different sizes, genres and annotation schemes. We obtain state-of-the-art performance on CTB5, achieving 94.38 F1-score for joint segmentation and POS tagging.

Historical texts are challenging for natural language processing because they differ linguistically from modern texts and because of their lack of orthographical and grammatical standardisation. We use a character-level neural network to build a part-of-speech (POS) tagger that can process historical data directly without requiring a separate spelling normalisation stage. Its performance in a Swedish verb identification and a German POS tagging task is similar to that of a two-stage model. We analyse the performance of this tagger and a more traditional baseline system, discuss some of the remaining problems for tagging historical data and suggest how the flexibility of our neural tagger could be exploited to address diachronic divergences in morphology and syntax in early modern Swedish with the help of data from closely related languages.

Structural kernels are a flexible learning paradigm that has been widely used in Natural Language Processing. However, the problem of model selection in kernel-based methods is usually overlooked. Previous approaches mostly rely on setting default values for kernel hyperparameters or using grid search, which is slow and coarse-grained. In contrast, Bayesian methods allow efficient model selection by maximizing the evidence on the training data through gradient-based methods. In this paper we show how to perform this in the context of structural kernels by using Gaussian Processes. Experimental results on tree kernels show that this procedure results in better prediction performance compared to hyperparameter optimization via grid search. The framework proposed in this paper can be adapted to other structures besides trees, e.g., strings and graphs, thereby extending the utility of kernel-based methods.

This paper addresses the task of predicting the correct French translations of third-person subject pronouns in English discourse, a problem that is relevant as a prerequisite for machine translation and that requires anaphora resolution.We present an approach based on neural networks that models anaphoric links as latent variables and show that its performance is competitive with that of a system with separate anaphora resolution while not requiring any coreference-annotated training data.This demonstrates that the information contained in parallel bitexts can successfully be used to acquire knowledge about pronominal anaphora in an unsupervised way.