Publications

In this work, we reimagine classical probing to evaluate knowledge transfer from simple source to more complex target tasks. Instead of probing frozen representations from a complex source task on diverse simple target probing tasks (as usually done in probing), we explore the effectiveness of embeddings from multiple simple source tasks on a single target task. We select coreference resolution, a linguistically complex problem requiring contextual understanding, as focus target task, and test the usefulness of embeddings from comparably simpler tasks tasks such as paraphrase detection, named entity recognition, and relation extraction. Through systematic experiments, we evaluate the impact of individual and combined task embeddings. Our findings reveal that task embeddings vary significantly in utility for coreference resolution, with semantic similarity tasks (e.g., paraphrase detection) proving most beneficial. Additionally, representations from intermediate layers of fine-tuned models often outperform those from final layers. Combining embeddings from multiple tasks consistently improves performance, with attention-based aggregation yielding substantial gains. These insights shed light on relationships between task-specific representations and their adaptability to complex downstream tasks, encouraging further exploration of embedding-level task transfer.

Cross-lingual knowledge transfer, especially between high- and low-resource languages, remains challenging in natural language processing (NLP). This study offers insights for improving cross-lingual NLP applications through the combination of parameter-efficient fine-tuning methods. We systematically explore strategies for enhancing cross-lingual transfer through the incorporation of language-specific and task-specific adapters and soft prompts. We present a detailed investigation of various combinations of these methods, exploring their efficiency across 16 languages, focusing on 10 mid- and low-resource languages. We further present to our knowledge the first use of soft prompts for language transfer, a technique we call soft language prompts. Our findings demonstrate that in contrast to claims of previous work, a combination of language and task adapters does not always work best; instead, combining a soft language prompt with a task adapter outperforms most configurations in many cases.

As the Information Retrieval (IR) field increasingly recognizes the importance of inclusivity, addressing the needs of low-resource languages remains a significant challenge. This paper introduces the first large-scale Urdu IR dataset, created by translating the MS MARCO dataset through machine translation. We establish baseline results through zero-shot learning for IR in Urdu and subsequently apply the mMARCO multilingual IR methodology to this newly translated dataset. Our findings demonstrate that the fine-tuned model (Urdu-mT5-mMARCO) achieves a Mean Reciprocal Rank (MRR@10) of 0.247 and a Recall@10 of 0.439, representing significant improvements over zero-shot results and showing the potential for expanding IR access for Urdu speakers. By bridging access gaps for speakers of low-resource languages, this work not only advances multilingual IR research but also emphasizes the ethical and societal importance of inclusive IR technologies. This work provides valuable insights into the challenges and solutions for improving language representation and lays the groundwork for future research, especially in South Asian languages, which can benefit from the adaptable methods used in this study.

Natural language explanations (NLEs) are vital for elucidating the reasoning behind large language model (LLM) decisions. Many techniques have been developed to generate NLEs using LLMs. However, like humans, LLMs might not always produce optimal NLEs on first attempt. Inspired by human learning processes, we introduce CROSS-REFINE, which employs role modeling by deploying two LLMs as generator and critic, respectively. The generator outputs a first NLE and then refines this initial explanation using feedback and suggestions provided by the critic. CROSS-REFINE does not require any supervised training data or additional training. We validate CROSS-REFINE across three NLP tasks using three state-of-the-art open-source LLMs through automatic and human evaluation. We select SELF-REFINE (Madaan et al., 2023) as the baseline, which only utilizes self-feedback to refine the explanations. Our findings from automatic evaluation and a user study indicate that CROSS-REFINE outperforms SELF-REFINE. Meanwhile, CROSS-REFINE can perform effectively with less powerful LLMs, whereas SELF-REFINE only yields strong results with ChatGPT. Additionally, we conduct an ablation study to assess the importance of feedback and suggestions. Both of them play an important role in refining explanations. We further evaluate CROSS-REFINE on a bilingual dataset in English and German.

Multilingual anaphora resolution identifies referring expressions and implicit arguments in texts and links to antecedents that cover several languages. In the most challenging setting, cross-lingual anaphora resolution, training data, and test data are in different languages. As knowledge needs to be transferred across languages, this task is challenging, both in the multilingual and cross-lingual setting. We hypothesize that one way to alleviate some of the difficulty of the task is to include multimodal information in the form of images (i.e. frames extracted from instructional videos). Such visual inputs are by nature language agnostic, therefore cross- and multilingual anaphora resolution should benefit from visual information. In this paper, we provide the first multilingual and multimodal dataset annotated with anaphoric relations and present experimental results for end-to-end multimodal and multilingual anaphora resolution. Given gold mentions, multimodal features improve anaphora resolution results by 10 % for unseen languages.

While the impact of tokenization on language modeling is well-researched in richly resourced languages, fewer studies on this topic exist for challenging low-resource languages. In this work, we present the first systematic evaluation of tokenization methods for Georgian, a low-resource language with high morphological complexity. We compare standard subword tokenizers, such as WordPiece, Byte Pair Encoding, SentencePiece with Unigram, and a recently proposed token-free approach. We also investigate the multilingual BERT tokenizer (mBERT), which includes Georgian. In addition to these different classes of tokenization algorithms we also evaluate the impact of different vocabulary sizes, a key parameter for subword tokenizers. We evaluate the performance of all tokenizers on masked language modeling and on four downstream tasks: part-of-speech tagging, named entity recognition, toxicity detection, and sentiment analysis. We observe that larger vocabulary sizes for subword tokenizers generally lead to better performance across most tasks, with a notable exception in the toxicity detection task, where finer subword granularity is more effective. For the remaining tasks, pre-training tokenizers on Georgian text consistently yield better results compared to mBERT. Additionally, the token-free method is consistently outperformed by all other tokenizers. Taken together, our comprehensive evaluation of tokenizers will be highly valuable in making informed tokenization choices in future language model developments for Georgian.

Conversational explainable artificial intelligence (ConvXAI) systems based on large language models (LLMs) have garnered significant interest from the research community in natural language processing (NLP) and human-computer interaction (HCI). Such systems can provide answers to user questions about explanations in dialogues, have the potential to enhance users’ comprehension and offer more information about the decision-making and generation processes of LLMs. Currently available ConvXAI systems are based on intent recognition rather than free chat, as this has been found to be more precise and reliable in identifying users’ intentions. However, the recognition of intents still presents a challenge in the case of ConvXAI, since little training data exist and the domain is highly specific, as there is a broad range of XAI methods to map requests onto. In order to bridge this gap, we present CoXQL, the first dataset for user intent recognition in ConvXAI, covering 31 intents, seven of which require filling multiple slots. Subsequently, we enhance an existing parsing approach by incorporating template validations, and conduct an evaluation of several LLMs on CoXQL using different parsing strategies. We conclude that the improved parsing approach (MP+) surpasses the performance of previous approaches. We also discover that intents with multiple slots remain highly challenging for LLMs.

We investigate the impact of LLMs on political discourse with a particular focus on the influence of generated personas on model responses. We find an echo chamber effect from LLM chatbots when provided with German-language biographical information of politicians and voters in German politics, leading to sycophantic responses and the reinforcement of existing political biases. Findings reveal that personas of certain political party, such as those of the ‘Alternative für Deutschland’ party, exert a stronger influence on LLMs, potentially amplifying extremist views. Unlike prior studies, we cannot corroborate a tendency for larger models to exert stronger sycophantic behaviour. We propose that further development should aim at reducing sycophantic behaviour in LLMs across all sizes and diversifying language capabilities in LLMs to enhance inclusivity.

In this paper, we investigate the use of large language models (LLMs) to enhance the editorial process of rewriting customer help pages. We introduce a German-language dataset comprising Frequently Asked Question-Answer pairs, presenting both raw drafts and their revisions by professional editors. On this dataset, we evaluate the performance of four large language models (LLM) through diverse prompts tailored for the rewriting task. We conduct automatic evaluations of content and text quality using ROUGE, BERTScore, and ChatGPT. Furthermore, we let professional editors assess the helpfulness of automatically generated FAQ revisions for editorial enhancement. Our findings indicate that LLMs can produce FAQ reformulations beneficial to the editorial process. We observe minimal performance discrepancies among LLMs for this task, and our survey on helpfulness underscores the subjective nature of editors’ perspectives on editorial refinement.

This paper explores the integration of graph knowledge from linguistic ontologies into multilingual Large Language Models (LLMs) using adapters to improve performance for low-resource languages (LRLs) in sentiment analysis (SA) and named entity recognition (NER). Building upon successful parameter-efficient fine-tuning techniques, such as K-ADAPTER and MAD-X, we propose a similar approach for incorporating knowledge from multilingual graphs, connecting concepts in various languages with each other through linguistic relationships, into multilingual LLMs for LRLs. Specifically, we focus on eight LRLs {—} Maltese, Bulgarian, Indonesian, Nepali, Javanese, Uyghur, Tibetan, and Sinhala {—} and employ language-specific adapters fine-tuned on data extracted from the language-specific section of ConceptNet, aiming to enable knowledge transfer across the languages covered by the knowledge graph. We compare various fine-tuning objectives, including standard Masked Language Modeling (MLM), MLM with full-word masking, and MLM with targeted masking, to analyze their effectiveness in learning and integrating the extracted graph data. Through empirical evaluation on language-specific tasks, we assess how structured graph knowledge affects the performance of multilingual LLMs for LRLs in SA and NER, providing insights into the potential benefits of adapting language models for low-resource scenarios.

Initially introduced as a machine translation model, the Transformer architecture has now become the foundation for modern deep learning architecture, with applications in a wide range of fields, from computer vision to natural language processing. Nowadays, to tackle increasingly more complex tasks, Transformer-based models are stretched to enormous sizes, requiring increasingly larger training datasets, and unsustainable amount of compute resources. The ubiquitous nature of the Transformer and its core component, the attention mechanism, are thus prime targets for efficiency research.In this work, we propose an alternative compatibility function for the self-attention mechanism introduced by the Transformer architecture. This compatibility function exploits an overlap in the learned representation of the traditional scaled dot-product attention, leading to a symmetric with pairwise coefficient dot-product attention. When applied to the pre-training of BERT-like models, this new symmetric attention mechanism reaches a score of 79.36 on the GLUE benchmark against 78.74 for the traditional implementation, leads to a reduction of 6% in the number of trainable parameters, and reduces the number of training steps required before convergence by half.

An adverse drug effect (ADE) is any harmful event resulting from medical drug treatment. Despite their importance, ADEs are often under-reported in official channels. Some research has therefore turned to detecting discussions of ADEs in social media. Impressive results have been achieved in various attempts to detect ADEs. In a high-stakes domain such as medicine, however, an in-depth evaluation of a model′s abilities is crucial. We address the issue of thorough performance evaluation in detecting ADEs with hand-crafted templates for four capabilities, temporal order, negation, sentiment and beneficial effect. We find that models with similar performance on held-out test sets have varying results on these capabilities.

This paper presents the dfki-mlst submission for the DialAM shared task (Ruiz-Dolz et al., 2024) on identification of argumentative and illocutionary relations in dialogue. Our model achieves best results in the global setting: 48.25 F1 at the focused level when looking only at the related arguments/locutions and 67.05 F1 at the general level when evaluating the complete argument maps. We describe our implementation of the data pre-processing, relation encoding and classification, evaluating 11 different base models and performing experiments with, e.g., node text combination and data augmentation. Our source code is publicly available.