In this paper, we introduce Eval-UA-tion, a set of novel Ukrainian-language datasets aimed at evaluating the performance of language models on the Ukrainian language. The tasks include UA-CBT (inspired by the Children’s Book Test, a fill-in-the-gaps type task aimed at gauging the extent to which a story narrative is understood), UP-Titles (where the online newspaper Ukrainska Pravda‘s articles have to be matched to the correct title among 10 similar ones), and LMentry-static-UA/LMES (inspired by the LMentry benchmark, a set of tasks simple to solve for humans but hard for LMs, such as ‘which of these words is longer’ and ‘what is the fifth word of this sentence’). With the exception of UP-Titles, the tasks are built in a way to minimize contamination and use material unlikely to be present in the training sets of language models, and include a split for few-shot model prompting use that minimizes contamination. For each task human and random baselines are provided.

Recent advancements in self-supervised pre-training of Language Models (LMs) have significantly improved their performance across a wide range of Natural Language Processing (NLP) tasks. Yet, the adaptation of these models to specialized domains remains a critical endeavor, as it enables the models to grasp domain-specific nuances, terminology, and patterns more effectively, thereby enhancing their utility in specialized contexts. This paper presents an in-depth investigation into the training and fine-tuning of German language models specifically for the financial sector. We construct various datasets for training and fine-tuning to examine the impact of different data construction strategies on the models’ performance. Our study provides detailed insights into essential pre-processing steps, including text extraction from PDF documents and language identification, to evaluate their influence on the performance of the language models. Addressing the scarcity of resources in the German financial domain, we also introduce a German Text Classification benchmark dataset, aimed at fostering further research and development in this area. The performance of the trained models is evaluated on two domain-specific tasks, demonstrating that fine-tuning with domain-specific data improves model outcomes, even with limited amounts of domain-specific data.

We introduce a predominantly German corpus comprising 12.5k PDF documents sourced from the financial domain. The corresponding extracted textual data encompasses more than 165 million tokens derived predominantly from German, and to a lesser extent, bilingual documents. We provide detailed information about the document types included in the corpus, such as final terms, base prospectuses, annual reports, information materials, law documents, international financial reporting standards, and monthly reports from the Bundesbank, accompanied by comprehensive statistical analysis. To our knowledge, it is the first non-email German financial corpus available, and we hope it will fill this gap and foster further research in the financial domain both in the German language and in multilingual contexts.

In this paper, we describe a publicly available multilingual evaluation corpus for phrase-level Sentiment Analysis that can be used to evaluate real world applications in an industrial context. This corpus contains data from English and German Internet forums (1000 posts each) focusing on the automotive domain. The major topic of the corpus is connecting and using cellphones to/in cars. The presented corpus contains different types of annotations: objects (e.g. my car, my new cellphone), features (e.g. address book, sound quality) and phrase-level polarities (e.g. the best possible automobile, big problem). Each of the posts has been annotated by at least four different annotators ― these annotations are retained in their original form. The reliability of the annotations is evaluated by inter-annotator agreement scores. Besides the corpus data and format, we provide comprehensive corpus statistics. This corpus is one of the first lexical resources focusing on real world applications that analyze the voice of the customer which is crucial for various industrial use cases.

Based on simple methods such as observing word and part of speech tag co-occurrence and clustering, we generate syntactic parses of sentences in an entirely unsupervised and self-inducing manner. The parser learns the structure of the language in question based on measuring “breaking points” within sentences. The learning process is divided into two phases, learning and application of learned knowledge. The basic learning works in an iterative manner which results in a hierarchical constituent representation of the sentence. Part-of-Speech tags are used to circumvent the data sparseness problem for rare words. The algorithm is applied on untagged data, on manually assigned tags and on tags produced by an unsupervised part of speech tagger. The results are unsurpassed by any self-induced parser and challenge the quality of trained parsers with respect to finding certain structures such as noun phrases.