The pretraining of state-of-the-art large language models now requires trillions of words of text, which is orders of magnitude more than available for the vast majority of languages. While including text in more than one language is an obvious way to acquire more pretraining data, multilinguality is often seen as a curse, and most model training efforts continue to focus near-exclusively on individual large languages. We believe that multilinguality can be a blessing: when the lack of training data is a constraint for effectively training larger models for a target language, augmenting the dataset with other languages can offer a way to improve over the capabilities of monolingual models for that language. In this study, we introduce Poro 34B, a 34 billion parameter model trained for 1 trillion tokens of Finnish, English, and programming languages, and demonstrate that a multilingual training approach can produce a model that substantially advances over the capabilities of existing models for Finnish and excels in translation, while also achieving competitive performance in its class for English and programming languages. We release the model parameters, scripts, and data under open licenses at https://huggingface.co/LumiOpen/Poro-34B.

As LLMs gain more popularity as chatbots and general assistants, methods have been developed to enable LLMs to follow instructions and align with human preferences. These methods have found success in the field, but their effectiveness has not been demonstrated outside of high-resource languages. In this work, we discuss our experiences in post-training an LLM for instruction-following for English and Finnish. We use a multilingual LLM to translate instruction and preference datasets from English to Finnish. We perform instruction tuning and preference optimization in English and Finnish and evaluate the instruction-following capabilities of the model in both languages. Our results show that with a few hundred Finnish instruction samples we can obtain competitive performance in Finnish instruction-following. We also found that although preference optimization in English offers some cross-lingual benefits, we obtain our best results by using preference data from both languages. We release our model, datasets, and recipes under open licenses at https://huggingface.co/LumiOpen/Poro-34B-chat-OpenAssistant.

This paper presents the results of the SHROOM, a shared task focused on detecting hallucinations: outputs from natural language generation (NLG) systems that are fluent, yet inaccurate. Such cases of overgeneration put in jeopardy many NLG applications, where correctness is often mission-critical. The shared task was conducted with a newly constructed dataset of 4000 model outputs labeled by 5 annotators each, spanning 3 NLP tasks: machine translation, paraphrase generation and definition modeling.The shared task was tackled by a total of 58 different users grouped in 42 teams, out of which 26 elected to write a system description paper; collectively, they submitted over 300 prediction sets on both tracks of the shared task. We observe a number of key trends in how this approach was tackled—many participants rely on a handful of model, and often rely either on synthetic data for fine-tuning or zero-shot prompting strategies. While a majority of the teams did outperform our proposed baseline system, the performances of top-scoring systems are still consistent with a random handling of the more challenging items.

Grounding has been argued to be a crucial component towards the development of more complete and truly semantically competent artificial intelligence systems. Literature has divided into two camps: While some argue that grounding allows for qualitatively different generalizations, others believe it can be compensated by mono-modal data quantity. Limited empirical evidence has emerged for or against either position, which we argue is due to the methodological challenges that come with studying grounding and its effects on NLP systems. In this paper, we establish a methodological framework for studying what the effects are—if any—of providing models with richer input sources than text-only. The crux of it lies in the construction of comparable samples of populations of models trained on different input modalities, so that we can tease apart the qualitative effects of different input sources from quantifiable model performances. Experiments using this framework reveal qualitative differences in model behavior between cross-modally grounded, cross-lingually grounded, and ungrounded models, which we measure both at a global dataset level as well as for specific word representations, depending on how concrete their semantics is.

This paper presents M3L-Contrast—a novel multimodal multilingual (M3L) neural topic model for comparable data that maps texts from multiple languages and images into a shared topic space. Our model is trained jointly on texts and images and takes advantage of pretrained document and image embeddings to abstract the complexities between different languages and modalities. As a multilingual topic model, it produces aligned language-specific topics and as multimodal model, it infers textual representations of semantic concepts in images. We demonstrate that our model is competitive with a zero-shot topic model in predicting topic distributions for comparable multilingual data and significantly outperforms a zero-shot model in predicting topic distributions for comparable texts and images. We also show that our model performs almost as well on unaligned embeddings as it does on aligned embeddings.

We tackle the problem of neural headline generation in a low-resource setting, where only limited amount of data is available to train a model. We compare the ideal high-resource scenario on English with results obtained on a smaller subset of the same data and also run experiments on two small news corpora covering low-resource languages, Croatian and Estonian. Two options for headline generation in a multilingual low-resource scenario are investigated: a pretrained multilingual encoder-decoder model and a combination of two pretrained language models, one used as an encoder and the other as a decoder, connected with a cross-attention layer that needs to be trained from scratch. The results show that the first approach outperforms the second one by a large margin. We explore several data augmentation and pretraining strategies in order to improve the performance of both models and show that while we can drastically improve the second approach using these strategies, they have little to no effect on the performance of the pretrained encoder-decoder model. Finally, we propose two new measures for evaluating the performance of the models besides the classic ROUGE scores.

In this paper, we present the participation of the EMBEDDIA team in the SemEval-2022 Task 8 (Multilingual News Article Similarity). We cover several techniques and propose different methods for finding the multilingual news article similarity by exploring the dataset in its entirety. We take advantage of the textual content of the articles, the provided metadata (e.g., titles, keywords, topics), the translated articles, the images (those that were available), and knowledge graph-based representations for entities and relations present in the articles. We, then, compute the semantic similarity between the different features and predict through regression the similarity scores. Our findings show that, while our proposed methods obtained promising results, exploiting the semantic textual similarity with sentence representations is unbeatable. Finally, in the official SemEval-2022 Task 8, we ranked fifth in the overall team ranking cross-lingual results, and second in the English-only results.

This paper presents tools and data sources collected and released by the EMBEDDIA project, supported by the European Union’s Horizon 2020 research and innovation program. The collected resources were offered to participants of a hackathon organized as part of the EACL Hackashop on News Media Content Analysis and Automated Report Generation in February 2021. The hackathon had six participating teams who addressed different challenges, either from the list of proposed challenges or their own news-industry-related tasks. This paper goes beyond the scope of the hackathon, as it brings together in a coherent and compact form most of the resources developed, collected and released by the EMBEDDIA project. Moreover, it constitutes a handy source for news media industry and researchers in the fields of Natural Language Processing and Social Science.

Team Name: team-8 Embeddia Tool: Cross-Lingual Document Retrieval Zosa et al. Dataset: Estonian and Latvian news datasets abstract: Contemporary news media face increasing amounts of available data that can be of use when prioritizing, selecting and discovering new news. In this work we propose a methodology for retrieving interesting articles in a cross-border news discovery setting. More specifically, we explore how a set of seed documents in Estonian can be projected in Latvian document space and serve as a basis for discovery of novel interesting pieces of Latvian news that would interest Estonian readers. The proposed methodology was evaluated by Estonian journalist who confirmed that in the best setting, from top 10 retrieved Latvian documents, half of them represent news that are potentially interesting to be taken by the Estonian media house and presented to Estonian readers.

Moderation of reader comments is a significant problem for online news platforms. Here, we experiment with models for automatic moderation, using a dataset of comments from a popular Croatian newspaper. Our analysis shows that while comments that violate the moderation rules mostly share common linguistic and thematic features, their content varies across the different sections of the newspaper. We therefore make our models topic-aware, incorporating semantic features from a topic model into the classification decision. Our results show that topic information improves the performance of the model, increases its confidence in correct outputs, and helps us understand the model’s outputs.

We address the problem of linking related documents across languages in a multilingual collection. We evaluate three diverse unsupervised methods to represent and compare documents: (1) multilingual topic model; (2) cross-lingual document embeddings; and (3) Wasserstein distance. We test the performance of these methods in retrieving news articles in Swedish that are known to be related to a given Finnish article. The results show that ensembles of the methods outperform the stand-alone methods, suggesting that they capture complementary characteristics of the documents

This paper describes the approaches used by the Discovery Team to solve SemEval-2020 Task 1 - Unsupervised Lexical Semantic Change Detection. The proposed method is based on clustering of BERT contextual embeddings, followed by a comparison of cluster distributions across time. The best results were obtained by an ensemble of this method and static Word2Vec embeddings. According to the official results, our approach proved the best for Latin in Subtask 2.

Dynamic topic models (DTMs) capture the evolution of topics and trends in time series data. Current DTMs are applicable only to monolingual datasets. In this paper we present the multilingual dynamic topic model (ML-DTM), a novel topic model that combines DTM with an existing multilingual topic modeling method to capture cross-lingual topics that evolve across time. We present results of this model on a parallel German-English corpus of news articles and a comparable corpus of Finnish and Swedish news articles. We demonstrate the capability of ML-DTM to track significant events related to a topic and show that it finds distinct topics and performs as well as existing multilingual topic models in aligning cross-lingual topics.

This paper is a part of a collaboration between computer scientists and historians aimed at development of novel tools and methods to improve analysis of historical newspapers. We present a case study of ideological terms ending with -ism suffix in nineteenth century Finnish newspapers. We propose a two-step procedure to trace differences in word usages over time: training of diachronic embeddings on several time slices and when clustering embeddings of selected words together with their neighbours to obtain historical context. The obtained clusters turn out to be useful for historical studies. The paper also discuss specific difficulties related to development historian-oriented tools.