Large language models, trained on vast datasets, exhibit increased output quality in proportion to the amount of data that is used to train them. This data-driven learning process has brought forth a pressing issue where these models may not only reflect but also amplify gender bias, racism, religious prejudice, and queerphobia present in their training data that may not always be recent. This study explores gender bias in language models trained on Icelandic, focusing on occupation-related terms. Icelandic is a highly grammatically gendered language that favors the masculine when referring to groups of people with indeterminable genders. Our aim is to explore whether language models merely mirror gender distributions within the corresponding professions or if they exhibit biases tied to their grammatical genders. Results indicate a significant overall predisposition towards the masculine but specific occupation terms consistently lean toward a particular gender, indicating complex interplays of societal and linguistic influences.

This study introduces “Ice and Fire,” a Multi-Task Learning (MTL) dataset tailored for sentiment analysis in the Icelandic language, encompassing a wide range of linguistic tasks, including sentiment and emotion detection, as well as identification of toxicity, hate speech, encouragement, sympathy, sarcasm/irony, and trolling. With 261 fully annotated blog comments and 1045 comments annotated in at least one task, this contribution marks a significant step forward in the field of Icelandic natural language processing. It provides a comprehensive dataset for understanding the nuances of online communication in Icelandic and an interface to expand the annotation effort. Despite the challenges inherent in subjective interpretation of text, our findings highlight the positive potential of this dataset to improve text analysis techniques and encourage more inclusive online discourse in Icelandic communities. With promising baseline performances, “Ice and Fire” sets the stage for future research to enhance automated text analysis and develop sophisticated language technologies, contributing to healthier online environments and advancing Icelandic language resources.

This paper presents a crowd-sourcing platform designed to address the need for parallel corpora in the field of Automatic Text Simplification (ATS). ATS aims to automatically reduce the linguistic complexity of text to aid individuals with reading difficulties, such as those with cognitive disorders, dyslexia, children, and non-native speakers. ATS does not only facilitate improved reading comprehension among these groups but can also enhance the preprocessing stage for various NLP tasks through summarization, contextual simplification, and paraphrasing. Our work introduces a language independent, openly accessible platform that crowdsources training data for ATS models, potentially benefiting low-resource languages where parallel data is scarce. The platform can efficiently aid in the collection of parallel corpora by providing a user-friendly data-collection environment. Furthermore, using human crowd-workers for the data collection process offers a potential resource for linguistic research on text simplification practices. The paper discusses the platform’s architecture, built with modern web technologies, and its user-friendly interface designed to encourage widespread participation. Through gamification and a robust admin panel, the platform incentivizes high-quality data collection and engagement from crowdworkers.

In this paper, we present the first Entity Linking corpus for Icelandic. We describe our approach of using a multilingual entity linking model (mGENRE) in combination with Wikipedia API Search (WAPIS) to label our data and compare it to an approach using WAPIS only. We find that our combined method reaches 53.9% coverage on our corpus, compared to 30.9% using only WAPIS. We analyze our results and explain the value of using a multilingual system when working with Icelandic. Additionally, we analyze the data that remain unlabeled, identify patterns and discuss why they may be more difficult to annotate.

Word embedding models have become commonplace in a wide range of NLP applications. In order to train and use the best possible models, accurate evaluation is needed. For extrinsic evaluation of word embedding models, analogy evaluation sets have been shown to be a good quality estimator. We introduce an Icelandic adaptation of a large analogy dataset, BATS, evaluate it on three different word embedding models and show that our evaluation set is apt at measuring the capabilities of such models.

In this paper, we present a novel approach to data collection for natural language processing (NLP), linguistic research and lexicographic work. Using the parlor game Fictionary as a framework, data can be crowd-sourced in a gamified manner, which carries the potential of faster, cheaper and better data when compared to traditional methods due to the engaging and competitive nature of the game. To improve data quality, the game includes a built-in review process where players review each other’s data and evaluate its quality. The paper proposes several games that can be used within this framework, and explains the value of the data generated by their use. These proposals include games that collect named entities along with their corresponding type tags, question-answer pairs, translation pairs and neologism, to name only a few. We are currently working on a digital platform that will host these games in Icelandic but wish to open the discussion around this topic and encourage other researchers to explore their own versions of the proposed games, all of which are language-independent.

Spell checkers and other proofreading software are crucial tools for people with dyslexia and other reading disabilities. Most spell checkers automatically detect spelling mistakes by looking up individual words and seeing if they exist in the vocabulary. However, one of the biggest challenges of automatic spelling correction is how to deal with real-word errors, i.e. spelling mistakes which lead to a real but unintended word, such as when then is written in place of than. These errors account for 20% of all spelling mistakes made by people with dyslexia. As both words exist in the vocabulary, a simple dictionary lookup will not detect the mistake. The only way to disambiguate which word was actually intended is to look at the context in which the word appears. This problem is particularly apparent in languages with rich morphology where there is often minimal orthographic difference between grammatical items. In this paper, we present our novel confusion set corpus for Icelandic and discuss how it could be used for context-sensitive spelling correction. We have collected word pairs from seven different categories, chosen for their homophonous properties, along with sentence examples and frequency information from said pairs. We present a small-scale machine learning experiment using a decision tree binary classification which results range from 73% to 86% average accuracy with 10-fold cross validation. While not intended as a finalized result, the method shows potential and will be improved in future research.