Papers by Erik Velldal
NorEval: A Norwegian Language Understanding and Generation Evaluation Benchmark (2025.findings-acl)
Copied to clipboard
Vladislav Mikhailov, Tita Enstad, David Samuel, Hans Christian Farsethås, Andrey Kutuzov, Erik Velldal, Lilja Øvrelid
| Challenge: | NorEval is a new evaluation suite for large-scale standardized benchmarking of Norwegian generative language models (LMs). |
| Approach: | They propose a new evaluation suite for large-scale standardized benchmarking of Norwegian generative language models (LMs) NorEval consists of 24 high-quality human-created datasets, of which five are created from scratch. |
| Outcome: | The evaluation framework and materials are publicly available. |
NoReC: The Norwegian Review Corpus (L18-1)
Copied to clipboard
Erik Velldal, Lilja Øvrelid, Eivind Alexander Bergem, Cathrine Stadsnes, Samia Touileb, Fredrik Jørgensen
| Challenge: | The Norwegian Review Corpus is a dataset of full-text reviews from major news sources. |
| Approach: | This paper presents the Norwegian Review Corpus, created for document-level sentiment analysis. |
| Outcome: | The corpus comprises more than 35,000 full-text reviews from a range of different domains. |
A Fine-grained Sentiment Dataset for Norwegian (2020.lrec-1)
Copied to clipboard
| Challenge: | Using a dataset for fine-grained sentiment analysis in Norwegian, we examine the annotation effort and provide an overview of the developed annotation guidelines. |
| Approach: | They propose a dataset for fine-grained sentiment analysis in Norwegian . they provide an overview of the developed annotation guidelines and analyze inter-annotator agreement . |
| Outcome: | The proposed dataset is the first of its kind for Norwegian and is available online. |
Direct parsing to sentiment graphs (2022.acl-short)
Copied to clipboard
| Challenge: | Existing methods for structured sentiment analysis (SSA) focus on subcomponents of sentiment graphs without explicitly expressing their relations or the polarity. |
| Approach: | They propose a graph-based semantic parser which directly predicts sentiment graphs from text without reliance on lossy conversions to intermediate dependency representations. |
| Outcome: | The proposed model performs on 4 out of 5 standard benchmark sets and compares with dependency-based models on the more structurally complex datasets. |
Measuring Idiomaticity in Text Embedding Models with epsilon-compositionality (2026.eacl-long)
Copied to clipboard
| Challenge: | Existing studies on compositionality of text embedding models have limited understanding of the principle . idioms have traditionally been seen as non-compositional . |
| Approach: | They propose to use formal definitions to define compositionality in text embedding models . they find that most models differentiate between idiomatic and non-idiomatic phrases . |
| Outcome: | The proposed model is able to differentiate between idiomatic and non-idiomatic phrases, the authors show . |
Diachronic word embeddings and semantic shifts: a survey (C18-1)
Copied to clipboard
| Challenge: | Existing methods for tracing time-related semantic shifts with word embedding models lack the cohesion, common terminology and shared practices of more established areas of natural language processing. |
| Approach: | They propose several axes along which these methods can be compared and propose a framework for comparison. |
| Outcome: | The proposed methods are compared with existing methods and outline their main challenges and potential applications. |
Measuring Normative and Descriptive Biases in Language Models Using Census Data (2023.eacl-main)
Copied to clipboard
| Challenge: | a new study examines how gender-based distributions of occupations are reflected in pre-trained language models. |
| Approach: | They propose a method to measure to what degree pre-trained language models are aligned to normative and descriptive occupational distributions. |
| Outcome: | The proposed method is language independent and can be extended to other dimensions of census data and demographic variables. |
Entity-Level Sentiment Analysis (ELSA): An Exploratory Task Survey (2022.coling-1)
Copied to clipboard
| Challenge: | Existing tasks and models for identifying sentiment expressed in text are lacking in identifying overall sentiment . prior work focused on document-level polarity classification, but ELSA is under-explored for longer texts with multiple mentions and opinions towards the same entity. |
| Approach: | They propose to use document-, sentence-, and target-level sentiment analysis to identify overall sentiment expressed towards volitional entities in a document. |
| Outcome: | The proposed task is referred to as "entity-level sentiment analysis" the proposed task performs poorly for longer texts with multiple mentions and opinions . |
If you’ve got it, flaunt it: Making the most of fine-grained sentiment annotations (2021.eacl-main)
Copied to clipboard
| Challenge: | a recent study shows that annotating sentiments is difficult and difficult. |
| Approach: | They propose to integrate holder and expression information into sentiment analysis to improve target extraction . they perform experiments on eight English datasets to determine whether annotating expressions improves target extraction. |
| Outcome: | The proposed approach improves target extraction and classification on English datasets. |
Structured Sentiment Analysis as Dependency Graph Parsing (2021.acl-long)
Copied to clipboard
| Challenge: | Structured sentiment analysis attempts to extract full opinion tuples from a text, but has been subdivided into smaller and smaller sub-tasks, e.g., target extraction or targeted polarity classification. |
| Approach: | They propose a framework which jointly predicts all elements of an opinion tuple and their relations by using dependency graph parsing. |
| Outcome: | The proposed framework improves on five datasets in English, Norwegian, Basque, and Catalan and refining the sentiment graphs with syntactic dependency information further improves results. |
NorNE: Annotating Named Entities for Norwegian (2020.lrec-1)
Copied to clipboard
| Challenge: | Using the annotations of the existing treebank, we have created a dataset for named entity recognition for Norwegian. |
| Approach: | They propose to create a manually annotated corpus of named entities for Norwegian . they propose to add named entity annotations to existing treebank . |
| Outcome: | The proposed dataset extends the annotation of the existing Norwegian Dependency Treebank. |
Compositional Generalization with Grounded Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for combining language models with knowledge graphs struggle with generalization to sequences of unseen lengths and novel combinations of seen base components. |
| Approach: | They propose a procedure for generating natural language questions paired with knowledge graphs that targets different aspects of compositionality and avoids grounding models in information already encoded in their weights. |
| Outcome: | The proposed method fails to generalize to unseen lengths and to novel combinations of seen base components. |
Trained on 100 million words and still in shape: BERT meets British National Corpus (2023.findings-eacl)
Copied to clipboard
| Challenge: | masked language models are trained on ever larger corpora, but pre-training on a modestly-sized but representative, well-balanced, and publicly available corpus can reach better performance than the original BERT model. |
| Approach: | They propose an optimized LM architecture called LTG-BERT that can be used to train a competitive language model on a small and standardizable corpus. |
| Outcome: | The proposed architecture outperforms the original English BERT model on a representative, well-balanced and publicly available corpus. |
Transfer and Multi-Task Learning for Noun–Noun Compound Interpretation (D18-1)
Copied to clipboard
| Challenge: | In computational linguistics, nounnoun compound interpretation is approached as an automatic classification problem. |
| Approach: | They empirically evaluate the utility of transfer and multi-task learning on a challenging semantic classification task. |
| Outcome: | The proposed methods improve the accuracy of a neural classifier and its F1 scores on the less frequent, but more difficult relations. |