Papers by Artur Kulmizev
Label and Explanation Variation in LLM-Based Annotation: a Case Study in Natural Language Inference (2026.acl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) have shown considerable promise for annotation purposes, but questions remain about their ability to capture human label variation (HLV) label variation is genuine disagreement between annotators observed across NLP tasks. |
| Approach: | They investigate how label and explanation variation manifests within and across LLMs with respect to the Natural Language Inference task. |
| Outcome: | The proposed models generate label distributions similar to humans but exhibit distinct, idiosyncratic judgments and disagreement patterns. |
Attention Can Reflect Syntactic Structure (If You Let It) (2021.eacl-main)
Copied to clipboard
| Challenge: | a recent study has attempted to decode linguistic structure from the Transformer . but, much of the work focused on English, a language with rigid word order and a lack of inflectional morphology. |
| Approach: | They propose to fine-tune a feature encoder for BERT to learn linguistic structure from its multi-head attention mechanism. |
| Outcome: | The proposed model can decode full trees above baseline accuracy from single attention heads across languages. |
What can we learn from Semantic Tagging? (D18-1)
Copied to clipboard
| Challenge: | a recent study shows that multi-task learning improves performance of NLP tasks by exploiting similarities between tasks. |
| Approach: | They employ semantic tagging as an auxiliary task for three NLP tasks . they compare full neural network sharing, partial neural network shared and learning what to share . |
| Outcome: | The proposed model improves for part-of-speech tagging, universal dependency parsing and natural language inference. |
MGAD: Multilingual Generation of Analogy Datasets (L18-1)
Copied to clipboard
| Challenge: | Existing methods for word embedding evaluation are computationally expensive and task-specific. |
| Approach: | They propose a minimally supervised method for generating word embedding evaluation datasets for a large number of languages using existing dependency treebanks and parsers. |
| Outcome: | The proposed method evaluates three popular word embedding algorithms against these datasets and shows that their performance varies between syntactic categories. |
Higher-order Comparisons of Sentence Encoder Representations (D19-1)
Copied to clipboard
| Challenge: | a technique developed by neuroscientists compares activity patterns of different measurement modalities . a recent study examined the correspondence between popular pretrained language encoders and human processing difficulty . |
| Approach: | They employ a technique to compare activity patterns of different measurement modalities . they establish a correspondence between widely-employed pretrained language encoders and human processing difficulty . |
| Outcome: | The proposed technique can be used to compare representational geometries of neural models . it does not require large training samples and is not prone to overfitting, authors say . |
Word Order Does Matter and Shuffled Language Models Know It (2022.acl-long)
Copied to clipboard
| Challenge: | Recent studies have shown that language models pretrained and/or fine-tuned on randomly permuted sentences exhibit competitive performance on GLUE, putting into question the importance of word order information. |
| Approach: | They propose a transformer-based BERT architecture that uses a fixed, sinusoidal position embedding added to each token embeddable to compensate for this absence of linear order. |
| Outcome: | The proposed model retains word order information because of the dependencies between sentence length and unigram probabilities. |
Deep Contextualized Word Embeddings in Transition-Based and Graph-Based Dependency Parsing - A Tale of Two Parsers Revisited (D19-1)
Copied to clipboard
| Challenge: | In recent years, dependency parsing has shifted from discrete features to neural networks and continuous representations. |
| Approach: | They propose to use deep contextualized word embeddings to pack information about global sentence structure into local feature representations to make the two approaches virtually equivalent in terms of accuracy and error profile. |
| Outcome: | The proposed model improves the accuracy and error profile of transition-based and graph-based dependency parsers on 13 languages. |
Investigating UD Treebanks via Dataset Difficulty Measures (2023.eacl-main)
Copied to clipboard
| Challenge: | Treebanks annotated with Universal Dependencies (UD) are currently available for over 100 languages and are only partially reflected in parser evaluations via accuracy metrics like LAS. |
| Approach: | They propose to use dataset cartography, V-information, and minimum description length to analyze UD treebanks using three accuracy-free methods to provide insights about them. |
| Outcome: | The proposed methods provide insights about UD treebanks that would remain undetected if only LAS was considered. |
How Good is Your Wikipedia? Auditing Data Quality for Low-resource and Multilingual NLP (2026.acl-long)
Copied to clipboard
Kushal Tatariya, Artur Kulmizev, Wessel Poelman, Esther Ploeger, Marcel Bollmann, Johannes Bjerva, Jiaming Luo, Heather Lent, Miryam de Lhoneux
| Challenge: | Wikipedia’s perceived high quality and broad language coverage have established it as a fundamental resource in NLP. |
| Approach: | They propose a data filtering procedure which removes a large percentage of Wikipedia's data and a 4-level quality ranking of the site. |
| Outcome: | The results show that the proposed filtering procedure outperforms the raw Wikipedia models in three language modelling scenarios. |
Do Neural Language Models Show Preferences for Syntactic Formalisms? (2020.acl-main)
Copied to clipboard
| Challenge: | Recent work on interpretability of deep neural language models concludes that many properties of natural language syntax are encoded in their representational spaces. |
| Approach: | They propose to examine whether syntactic structure adheres to a surface-syntactical or deep syntaktic style of analysis. |
| Outcome: | The proposed model prefers Universal Dependencies (UD) over Surface-Syntactic Universal Dependency (SUD) with interesting variations across languages and layers. |
Positional Artefacts Propagate Through Masked Language Model Embeddings (2021.acl-long)
Copied to clipboard
| Challenge: | Existing word embedding models have a uniform pitfall in assigning a static vector to a word type. |
| Approach: | They propose a neuron-level analysis method to investigate the source of this information by comparing outlier neurons within BERT and RoBERTa’s hidden state vectors. |
| Outcome: | The proposed method pre-trains the RoBERTa-based models and shows that the outliers disappear without positional embeddings. |