Papers by Alan Akbik
Fundus: A Simple-to-Use News Scraper Optimized for High Quality Extractions (2024.acl-demos)
Copied to clipboard
| Challenge: | Fundus is a news scraper that extracts news articles from the web with just a few lines of code. |
| Approach: | They introduce Fundus, a news scraper that enables users to obtain news articles with just a few lines of code. |
| Outcome: | The proposed news scraper optimizes for quality and provides a unified interface for newspapers. |
Question Decomposition for Retrieval-Augmented Generation (2025.acl-srw)
Copied to clipboard
| Challenge: | Retrieval-augmented generation (RAG) is effective for question answering tasks . multi-hop questions, such as "Which company among NVIDIA, Apple, and Google made the biggest profit in 2023?" challenge RAG because relevant facts are often distributed across multiple documents . |
| Approach: | They propose a pipeline that incorporates question decomposition to ground large language models in verifiable external sources. |
| Outcome: | The proposed approach improves retrieval and answer accuracy over standard RAG . multi-hop questions often require multiple documents to support the model . |
Improving Online Job Advertisement Analysis via Compositional Entity Extraction (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing work on IE in OJAs has focused on skills extraction, but other information is extracted using job tasks, job titles, and work tools. |
| Approach: | They propose a compositional entity modeling framework for requirement extraction from online job advertisements (OJAs) they annotate a manually annotated dataset of 500 German job ads that captures roles, tools, experience levels, attitudes, and their functional context. |
| Outcome: | The proposed framework can extract requirements from a manually annotated dataset of 500 German job ads. |
Contextual String Embeddings for Sequence Labeling (C18-1)
Copied to clipboard
| Challenge: | Recent advances in language modeling have made it viable to model language as distributions over characters. |
| Approach: | They propose to leverage internal states of a trained character language model to produce a new type of word embeddings. |
| Outcome: | The proposed embeddings outperform the state-of-the-art on four classic sequence labeling tasks. |
NoiseBench: Benchmarking the Impact of Real Label Noise on Named Entity Recognition (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to named entity recognition often contain a significant percentage of incorrect labels for entity types and boundary boundaries. |
| Approach: | They propose a noise-robust learning approach that learns from data with partially incorrect labels. |
| Outcome: | The proposed methods are based on simulated noise and are easier to handle than simulated real noise caused by human error or semi-automatic annotation. |
Evaluating Design Decisions for Dual Encoder-based Entity Disambiguation (2025.acl-long)
Copied to clipboard
| Challenge: | Entity disambiguation (ED) is the task of linking mentions in text to corresponding entries in a knowledge base. |
| Approach: | They propose a document-level Dual Encoder model that embeds mentions and label candidates in a shared embedding space and applies a similarity metric to predict the correct label. |
| Outcome: | The proposed model improves the disambiguation of ambiguous mentions of entities in text to their respective KB entries. |
ZAP: An Open-Source Multilingual Annotation Projection Framework (L18-1)
Copied to clipboard
| Challenge: | Existing frameworks for annotation projection in parallel corpora limit reproducibility and comparison of experiments. |
| Approach: | They propose an open-source framework for annotation projection in parallel corpora . framework is Java-based and includes methods for preprocessing corpors, computations and visualization . |
| Outcome: | The proposed framework is designed for ease-of-use with lightweight APIs. |
Token-Level Metrics for Detecting Incorrect Gold Annotations in Named Entity Recognition (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Annotated datasets for supervised learning often contain incorrect labels, i.e. label noise. |
| Approach: | They compare popular sample metrics for detecting incorrect annotations in named entity recognition (NER) they find that training dynamics metrics work the best overall, and they detect errors that the model has not yet memorized . |
| Outcome: | The proposed measures reduce label noise across noise types by detecting errors in trained models. |
Familarity: Better Evaluation of Zero-Shot Named Entity Recognition by Quantifying Label Shifts in Synthetic Training Data (2025.naacl-long)
Copied to clipboard
| Challenge: | Current research relies on large synthetic datasets to train zero-shot named entity recognition models. |
| Approach: | They propose a metric that captures the semantic similarity between entity types in training and evaluation to estimate label shift. |
| Outcome: | The proposed metric captures semantic similarity between entity types in training and evaluation, and their frequency in training data to provide an estimate of label shift. |
Large-Scale Label Interpretation Learning for Few-Shot Named Entity Recognition (2024.eacl-long)
Copied to clipboard
| Challenge: | Few-shot named entity recognition (NER) uses only a few annotated examples to identify named entities within text. |
| Approach: | They propose to leverage natural language descriptions of each entity type to perform few-shot named entity recognition. |
| Outcome: | The proposed model learns to interpret verbalized descriptions of entities using natural language descriptions of their types and their verbalizations. |
OpinionGPT: Modelling Explicit Biases in Instruction-Tuned LLMs (2024.naacl-demo)
Copied to clipboard
| Challenge: | Current research seeks to de-bias such models, or suppress potentially biased answers. |
| Approach: | They present a web demo to test the biases of instruction-tuned Large Language Models . they identify 11 different biase based on a corpus of data . |
| Outcome: | The proposed demo shows that biases in instruction-tuning are explicit and transparent . the demo shows how the model was trained and showcases the web application . |
Parameter-Efficient Fine-Tuning: Is There An Optimal Subset of Parameters to Tune? (2024.findings-eacl)
Copied to clipboard
| Challenge: | Recent research has illuminated the possibility of selective parameter-efficient fine-tuning, which retains the inference speed of the original model and comes at no additional computational cost. |
| Approach: | They propose to selectively update only a small subset of parameters during the fine-tuning process, keeping the remaining parameters frozen during training. |
| Outcome: | The proposed methods retain the inference speed of the original model and come at no additional computational cost. |
Early Detection of Sexual Predators in Chats (2021.acl-long)
Copied to clipboard
| Challenge: | Prior work has attempted to identify grooming chats but only after an incidence has already happened in the context of legal prosecution. |
| Approach: | They propose to analyze a running chat and predict grooming attempts as early as possible . they propose to use a new dataset to evaluate the problem from the point of view of prevention . |
| Outcome: | The proposed model is based on existing datasets and their limitations . it can be used to predict grooming attempts as early as possible . |
Pooled Contextualized Embeddings for Named Entity Recognition (N19-1)
Copied to clipboard
| Challenge: | Contextual string embeddings are a recent type of word embeddable that are useful for sequence labeling tasks. |
| Approach: | They propose a method that dynamically aggregates contextualized embeddings of each unique string . they then use a pooling operation to distill a ”global” word representation from all contextualized instances . |
| Outcome: | The proposed method improves state-of-the-art for named entity recognition tasks. |
Lemma Dilemma: On Lemma Generation Without Domain- or Language-Specific Training Data (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Large language models (LLMs) can generate lemmas in context without prior fine-tuning. |
| Approach: | They compare in-context lemma generation with traditional fully supervised approaches . they use encoder-only supervised methods and cross-lingual methods . |
| Outcome: | The proposed model outperforms the traditional fully supervised approach in the context of lemmatization tasks. |
FLAIR: An Easy-to-Use Framework for State-of-the-Art NLP (N19-4)
Copied to clipboard
| Challenge: | Existing approaches combine word embeddings with character-level features to model additional features such as subword structures and meaning ambiguity. |
| Approach: | They present FLAIR, an NLP framework that enables embeddings of word and document data . they propose a hierarchical learning architecture that concatenates output states of a character-level CNN or RNN with the output states from a task data. |
| Outcome: | The proposed framework hides embedding-specific engineering complexity and allows researchers to "mix and match" various embeddables with little effort. |
FiNERweb: Datasets and Artifacts for Scalable Multilingual Named Entity Recognition (2026.findings-eacl)
Copied to clipboard
| Challenge: | Named entity recognition (NER) is the task of identifying tokens that belong to a predefined set of classes such as "person" or "location" |
| Approach: | They propose a dataset-creation pipeline that scales the teacher-student paradigm to 91 languages and 25 scripts. |
| Outcome: | The proposed model achieves comparable or improved performance in English, Thai, and Swahili despite being trained on 19x less data than strong baselines. |
BEAR: A Unified Framework for Evaluating Relational Knowledge in Causal and Masked Language Models (2024.findings-naacl)
Copied to clipboard
| Challenge: | Existing methods to evaluate LMs rely on objective function and are therefore limited to masked or causal LM types. |
| Approach: | They propose an approach that uses an LM’s inherent ability to estimate the log-likelihood of any given textual statement. |
| Outcome: | The proposed framework can probe for knowledge across different LM types. |
Choose Your Transformer: Improved Transferability Estimation of Transformer Models on Classification Tasks (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing models for NLP tasks require fine-tuning, but it is computationally infeasible. |
| Approach: | They propose an approach that inexpensively estimates a ranking of the expected performance of a given set of transformer language models for a specific task. |
| Outcome: | The proposed model improves the Pearson correlation coefficient between the true model ranks and the estimate. |
Medical Coding with Biomedical Transformer Ensembles and Zero/Few-shot Learning (2022.naacl-industry)
Copied to clipboard
| Challenge: | Medical coding (MC) is an essential pre-requisite for reliable data retrieval and reporting. |
| Approach: | They propose a method to classify medical terms into standardized alphanumerical terms and codes . they use a combination of traditional BERT-based classification and a zero/few-shot learning approach . |
| Outcome: | The proposed approach outperforms baselines in the few-shot regime. |
Beyond Marginal Distributions: A Framework to Evaluate the Representativeness of Demographic-Aligned LLMs (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing work on marginal distributions and model steering fails to account for deeper latent structures that characterise real populations. |
| Approach: | They propose a framework for evaluating the representativeness of aligned models through multivariate correlation patterns in addition to marginal distributions. |
| Outcome: | The proposed framework compares two model steering techniques against human responses from the World Values Survey. |
PECC: Problem Extraction and Coding Challenges (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing benchmarks evaluate tasks in isolation, yet the extent to which LLMs can understand prose-style tasks, identify the underlying problems, and then generate appropriate code solutions remains mostly unexplored. |
| Approach: | They propose a benchmark derived from Advent Of Code challenges and Project Euler, which requires LLMs to interpret narrative-embedded problems, extract requirements, and generate executable code. |
| Outcome: | The proposed benchmarks show that LLMs can understand prose-style tasks, identify underlying problems, and generate appropriate code solutions in a variety of tasks. |
CleanCoNLL: A Nearly Noise-Free Named Entity Recognition Dataset (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing models achieve F1-scores comparable to or exceed noise level in CoNLL-03 . current models have significant annotation errors, incompleteness, and inconsistencies in the data . |
| Approach: | They propose to add a layer of entity linking annotation to the CoNLL-03 corpus to correct 7.0% of all labels. |
| Outcome: | The proposed approach corrects 7.0% of all labels in the English CoNLL-03 dataset. |
Fabricator: An Open Source Toolkit for Generating Labeled Training Data with Teacher LLMs (2023.emnlp-demo)
Copied to clipboard
| Challenge: | Recent research addresses the bottleneck of producing labeled training data for NLP tasks. |
| Approach: | They propose a method that generates labeled data that can be used to train a downstream NLP model. |
| Outcome: | The proposed model enables an LLM to generate labeled data that can be used to train a downstream NLP model. |
Task-Aware Representation of Sentences for Generic Text Classification (2020.coling-main)
Copied to clipboard
| Challenge: | Existing approaches to text classification use a transformer architecture with a linear layer on top. |
| Approach: | They propose a transformer-based approach that outputs a class distribution for a given prediction problem. |
| Outcome: | The proposed model outperforms existing approaches on small training data and can learn to predict new classes even with no training examples. |
Less is More: Parameter-Efficient Selection of Intermediate Tasks for Transfer Learning (2024.emnlp-main)
Copied to clipboard
| Challenge: | Prior methods producing useful task rankings are infeasible for large source pools . Embedding space maps (ESMs) reduce execution time and disk space usage . |
| Approach: | They introduce Embedded Space Maps (ESMs) that approximate the effect of fine-tuning a language model. |
| Outcome: | The proposed method reduces execution time and disk space usage by 10 and 278, respectively, while retaining high selection performance. |
Pre-Training Curriculum for Multi-Token Prediction in Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Multi-token prediction (MTP) is a pre-training objective for language models . prior work has shown that smaller language models struggle with the MTP objective . |
| Approach: | They propose a curriculum learning strategy that uses multiple prediction heads to predict the next tokens at each prediction step. |
| Outcome: | The proposed curriculum improves performance and output quality while retaining the benefits of self-speculative decoding. |
ZELDA: A Comprehensive Benchmark for Supervised Entity Disambiguation (2023.eacl-main)
Copied to clipboard
| Challenge: | Entity disambiguation (ED) is the task of disambiguating named entity mentions in text to unique entries in a knowledge base. |
| Approach: | They propose a benchmark for entity disambiguation that includes a unified training data set, entity vocabulary, candidate lists and challenging evaluation splits covering 8 different domains. |
| Outcome: | The proposed benchmark is based on a unified training data set, entity vocabulary, candidate lists and evaluation splits covering 8 different domains. |
FEIDEGGER: A Multi-modal Corpus of Fashion Images and Descriptions in German (L18-1)
Copied to clipboard
| Challenge: | Recent years have seen a renewed interest in text-image multi-modality . paired text-picture datasets are often limited to English language text . |
| Approach: | They propose a multi-modal corpus that pairs images and textual descriptions of their content in German to enable study of these challenges. |
| Outcome: | The proposed dataset focuses on the domain of fashion items and their visual descriptions in German. |