We introduce UniversalCEFR, a large-scale multilingual multidimensional dataset of texts annotated according to the CEFR (Common European Framework of Reference) scale in 13 languages. To enable open research in both automated readability and language proficiency assessment, UniversalCEFR comprises 505,807 CEFR-labeled texts curated from educational and learner-oriented resources, standardized into a unified data format to support consistent processing, analysis, and modeling across tasks and languages. To demonstrate its utility, we conduct benchmark experiments using three modelling paradigms: a) linguistic feature-based classification, b) fine-tuning pre-trained LLMs, and c) descriptor-based prompting of instruction-tuned LLMs. Our results further support using linguistic features and fine-tuning pretrained models in multilingual CEFR level assessment. Overall, UniversalCEFR aims to establish best practices in data distribution in language proficiency research by standardising dataset formats and promoting their accessibility to the global research community.

Reporting bias is the tendency for speakers to omit unnecessary or obvious information while mentioning things they consider relevant or surprising. In descriptions of people, reporting bias can manifest as a tendency to over report on attributes that deviate from the norm. While social bias in language models has garnered a lot of attention in recent years, a majority of the existing work equates “bias” with “stereotypes”. We suggest reporting bias as an alternative lens through which to study how social attitudes manifest in language models. We present the MARB dataset, a diagnostic dataset for studying the interaction between social bias and reporting bias in language models. We use MARB to evaluate the off-the-shelf behavior of both masked and autoregressive language models and find signs of reporting bias with regards to marginalized identities, mirroring that which can be found in human text. This effect is particularly pronounced when taking gender into account, demonstrating the importance of considering intersectionality when studying social phenomena like biases.

In this paper, we experiment with the effect of different levels of detailedness or granularity—understood as i) the number of classes, and ii) the classes’ semantic depth in the sense of hypernym and hyponym relations — of the annotation of Personally Identifiable Information (PII) on automatic detection and labeling of such information. We fine-tune a Swedish BERT model on a corpus of Swedish learner essays annotated with a total of six PII tagsets at varying levels of granularity. We also investigate whether the presence of grammatical and lexical correction annotation in the tokens and class prevalence have an effect on predictions. We observe that the fewer total categories there are, the better the overall results are, but having a more diverse annotation facilitates fewer misclassifications for tokens containing correction annotation. We also note that the classes’ internal diversity has an effect on labeling. We conclude from the results that while labeling based on the detailed annotation is difficult because of the number of classes, it is likely that models trained on such annotation rely more on the semantic content captured by contextual word embeddings rather than just the form of the tokens, making them more robust against nonstandard language.

The way we relay factual information and the way we present deceptive information as truth differs from the perspective of argumentation. In this paper, we explore whether these differences can be exploited to detect deceptive political news in English. We do this by training a model to detect different kinds of argumentation in online news text. We use sentence embeddings extracted from an argumentation type classification model as features for a deceptive news classifier. This deceptive news classification model leverages the sequence of argumentation types within an article to determine whether it is credible or deceptive. Our approach outperforms other state-of-the-art models while having lower variance. Finally, we use the output of our argumentation model to analyze the differences between credible and deceptive news based on the distribution of argumentation types across the articles. Results of this analysis indicate that credible political news presents statements supported by a variety of argumentation types, while deceptive news relies on anecdotes and testimonial.

Generative language models have been used to study a wide variety of phenomena in NLP. This allows us to better understand the linguistic capabilities of those models and to better analyse the texts that we are working with. However, these studies have mainly focused on text generated by L1 speakers of English. In this paper we study whether linguistic competence of L2 learners of Swedish (through their performance on essay tasks) correlates with the perplexity of a decoder-only model (GPT-SW3). We run two sets of experiments, doing both quantitative and qualitative analyses for each of them. In the first one, we analyse the perplexities of the essays and compare them with the CEFR level of the essays, both from an essay-wide level and from a token level. In our second experiment, we compare the perplexity of an L2 learner essay with a normalised version of it. We find that the perplexity of essays tends to be lower for higher CEFR levels and that normalised essays have a lower perplexity than the original versions. Moreover, we find that different factors can lead to spikes in perplexity, not all of them being related to L2 learner language.

Linguistic data can — and often does — contain PII (Personal Identifiable Information). Both from a legal and ethical standpoint, the sharing of such data is not permissible. According to the GDPR, pseudonymization, i.e. the replacement of sensitive information with surrogates, is an acceptable strategy for privacy preservation. While research has been conducted on the detection and replacement of sensitive data in Swedish medical data using Large Language Models (LLMs), it is unclear whether these models handle PII in less structured and more thematically varied texts equally well. In this paper, we present and discuss the performance of an LLM-based PII-detection system for Swedish learner essays.

Automated essay scoring (AES) of second-language learner essays is a high-stakes task as it can affect the job and educational opportunities a student may have access to. Thus, it becomes imperative to make sure that the essays are graded based on the students’ language proficiency as opposed to other reasons, such as personal names used in the text of the essay. Moreover, most of the research data for AES tends to contain personal identifiable information. Because of that, pseudonymization becomes an important tool to make sure that this data can be freely shared. Thus, our systems should not grade students based on which given names were used in the text of the essay, both for fairness and for privacy reasons. In this paper we explore how given names affect the CEFR level classification of essays of second language learners of Swedish. We use essays containing just one personal name and substitute it for names from lists of given names from four different ethnic origins, namely Swedish, Finnish, Anglo-American, and Arabic. We find that changing the names within the essays has no apparent effect on the classification task, regardless of whether a feature-based or a transformer-based model is used.

Linguistic data, a component critical not only for research in a variety of fields but also for the development of various Natural Language Processing (NLP) applications, can contain personal information. As a result, its accessibility is limited, both from a legal and an ethical standpoint. One of the solutions is the pseudonymization of the data. Key stages of this process include the identification of sensitive elements and the generation of suitable surrogates in a way that the data is still useful for the intended task. Within this paper, we conduct an analysis of tagsets that have previously been utilized in anonymization and pseudonymization. We also investigate what kinds of Personally Identifiable Information (PII) appear in various domains. These reveal that none of the analyzed tagsets account for all of the PII types present cross-domain at the level of detailedness seemingly required for pseudonymization. We advocate for a universal system of tags for categorizing PIIs leading up to their replacement. Such categorization could facilitate the generation of grammatically, semantically, and sociolinguistically appropriate surrogates for the kinds of information that are considered sensitive in a given domain, resulting in a system that would enable dynamic pseudonymization while keeping the texts readable and useful for future research in various fields.

Knowing our past can help us better understand our future. The explosive development of NLP in these past few decades has allowed us to study ancient languages and cultures in ways that we couldn’t have done in the past. However, not all languages have received the same level of attention. Despite its popularity in pop culture, the languages spoken in Ancient Egypt have been somewhat overlooked in terms of NLP research. In this paper we give an overview of how NLP has been used to study different variations of the Ancient Egyptian languages. This not only includes Old, Middle, and Late Egyptian but also Demotic and Coptic. We begin our survey paper by giving a short introduction to these languages and their writing systems, before talking about the corpora and lexical resources that are available digitally. We then show the different NLP tasks that have been tackled for different variations of Ancient Egyptian, as well as the approaches that have been used. We hope that our work can stoke interest in the study of these languages within the NLP community.

Multiword expressions (MWEs) are common word combinations which exhibit idiosyncrasies in various linguistic levels. For various downstream natural language processing applications and tasks, the identification and discovery of MWEs has been proven to be potentially practical and useful, but still challenging to codify. In this paper we investigate various, relevant to MWE, resources and tools for Swedish, and, within a specific application scenario, namely ‘vaccine skepticism’, we apply structural topic modelling to investigate whether there are any interpretative advantages of identifying MWEs.

With widespread commercial applications in various domains, sentiment analysis has become a success story for Natural Language Processing (NLP). Still, although sentiment analysis has rapidly progressed during the last years, mainly due to the application of modern AI technologies, many approaches apply knowledge-based strategies, such as lexicon-based, to the task. This is particularly true for analyzing short social media content, e.g., tweets. Moreover, lexicon-based sentiment analysis approaches are usually preferred over learning-based methods when training data is unavailable or insufficient. Therefore, our main goal is to scale-up and apply a lexicon-based approach which can be used as a baseline to Swedish sentiment analysis. All scaled-up resources are made available, while the performance of this enhanced tool is evaluated on two short datasets, achieving adequate results.

Throughout the COVID-19 pandemic, a parallel infodemic has also been going on such that the information has been spreading faster than the virus itself. During this time, every individual needs to access accurate news in order to take corresponding protective measures, regardless of their country of origin or the language they speak, as misinformation can cause significant loss to not only individuals but also society. In this paper we train several machine learning models (ranging from traditional machine learning to deep learning) to try to determine whether news articles come from either a reliable or an unreliable source, using just the body of the article. Moreover, we use a previously introduced corpus of news in Swedish related to the COVID-19 pandemic for the classification task. Given that our dataset is both unbalanced and small, we use subsampling and easy data augmentation (EDA) to try to solve these issues. In the end, we realize that, due to the small size of our dataset, using traditional machine learning along with data augmentation yields results that rival those of transformer models such as BERT.

Natural Language Processing (NLP) systems learn harmful societal biases that cause them to amplify inequality as they are deployed in more and more situations. To guide efforts at debiasing these systems, the NLP community relies on a variety of metrics that quantify bias in models. Some of these metrics are intrinsic, measuring bias in word embedding spaces, and some are extrinsic, measuring bias in downstream tasks that the word embeddings enable. Do these intrinsic and extrinsic metrics correlate with each other? We compare intrinsic and extrinsic metrics across hundreds of trained models covering different tasks and experimental conditions. Our results show no reliable correlation between these metrics that holds in all scenarios across tasks and languages. We urge researchers working on debiasing to focus on extrinsic measures of bias, and to make using these measures more feasible via creation of new challenge sets and annotated test data. To aid this effort, we release code, a new intrinsic metric, and an annotated test set focused on gender bias in hate speech.