Papers by Vlad Niculae

12 papers
Representation Collapse in Machine Translation Through the Lens of Angular Dispersion (2026.findings-eacl)

Copied to clipboard

Challenge: Effective training of Transformer models for sequential language tasks is difficult due to various forms of collapse of the internal representations learned.
Approach: They propose to use angular dispersion to analyze representation collapse at different levels of discrete and continuous transformers throughout training.
Outcome: The proposed method mitigates collapse and improves translation quality.
Sparse Sequence-to-Sequence Models (P19-1)

Copied to clipboard

Challenge: Sequence-to-sequence models are dense and assigning nonzero probability to implausible outputs.
Approach: They propose a new family of -entmax transformations that includes softmax and sparsemax as particular cases and is sparser for any > 1 . they provide fast algorithms to evaluate these transformations and their gradients, which scale well for large vocabulary sizes.
Outcome: The proposed models are able to produce sparse alignments and assign nonzero probability to short list of plausible outputs, sometimes rendering beam search exact.
Understanding the Mechanics of SPIGOT: Surrogate Gradients for Latent Structure Learning (2020.emnlp-main)

Copied to clipboard

Challenge: Latent structure models can mitigate the error propagation and annotation bottleneck in pipeline systems, while uncovering linguistic insights about the data.
Approach: They propose a latent structure model with a pullback of the downstream learning objective.
Outcome: The proposed model outperforms the known and proposed model in the same family and yields new insights for practitioners and revealing intriguing failure cases.
Unlocking Latent Discourse Translation in LLMs Through Quality-Aware Decoding (2026.eacl-long)

Copied to clipboard

Challenge: Large language models (LLMs) struggle to adequately handle discourse phenomena, such as pronoun resolution and lexical cohesion at the document level.
Approach: They propose to use minimum Bayes risk decoding to extract discourse knowledge from LLMs and propose to apply QAD to enhance the semantic richness of translations.
Outcome: The proposed method outperforms other methods and enhances the semantic richness of translations and aligns them more closely with human preferences.
Latent Structure Models for Natural Language Processing (P19-4)

Copied to clipboard

Challenge: Latent structure models are a powerful tool for compositional data modeling and pipelines.
Approach: This tutorial will cover recent advances in discrete latent structure models . it will discuss their motivation, potential, and limitations .
Outcome: This tutorial will cover recent advances in discrete latent structure models . it will discuss their motivation, potential, and limitations .
Towards Dynamic Computation Graphs via Sparse Latent Structure (D18-1)

Copied to clipboard

Challenge: Existing approaches to learn latent structure are limited by factorization assumptions or end-to-end differentiability.
Approach: They propose a method that allows for end-to-end learning of latent structure predictors jointly with a downstream predictor.
Outcome: The proposed method allows for unrestricted dynamic graph construction from the global latent structure while maintaining differentiability.
On Measuring Context Utilization in Document-Level MT Systems (2024.findings-eacl)

Copied to clipboard

Challenge: Current studies on document-level translation evaluation focus on sentence-level models which are inadequate for capturing improvements in discourse phenomena.
Approach: They propose to complement accuracy-based evaluation with measures of context utilization.
Outcome: The proposed model can be used to handle context-dependent discourse phenomena using an automatic annotation tool.
Viewing Knowledge Transfer in Multilingual Machine Translation Through a Representational Lens (2023.findings-emnlp)

Copied to clipboard

Challenge: Existing studies show that translation quality alone is not sufficient for measuring knowledge transfer in multilingual neural machine translation.
Approach: They propose a method that measures representational similarities between languages to measure knowledge transfer.
Outcome: The proposed method improves translation quality for low- and mid-resource languages across multiple datasets and models.
Angular Dispersion Accelerates k-Nearest Neighbors Machine Translation (2025.findings-emnlp)

Copied to clipboard

Challenge: k-nearest neighbors machine translation is a well-established strategy for increasing translation performance . high computational cost and memory requirements make it difficult to perform exhaustive searches in large data stores .
Approach: They propose to augment neural machine translation with external memory at decoding time . they propose to encourage angular dispersion of the neural hidden representations of contexts .
Outcome: The proposed method is expensive and high memory-intensive . it is based on the performance properties of k-NN lookup data structures.
The Unreasonable Effectiveness of Random Target Embeddings for Continuous-Output Neural Machine Translation (2024.naacl-short)

Copied to clipboard

Challenge: Continuous-output neural machine translation models are trained to predict the continuous representation based on distances between vectors.
Approach: They propose a continuous-output neural machine translation (CoNMT) approach that uses random output embeddings to outperform laboriously pre-trained models.
Outcome: The proposed strategy outperforms pre-trained embeddings on large datasets and is strongest for rare words due to the geometry of their embedders.
kNN For Whisper And Its Effect On Bias And Speaker Adaptation (2025.findings-naacl)

Copied to clipboard

Challenge: Speech recognition performance varies by language, domain, speaker characteristics such as accent . fine-tuning models on any of these categories may lead to catastrophic forgetting .
Approach: They propose a nonparametric method that adapts to any category without training the model.
Outcome: The proposed method improves the performance of a transformer end-to-end speech model on transcribed audio in over 680,000 hours of training data on a wide variety of languages.
Adaptively Sparse Transformers (D19-1)

Copied to clipboard

Challenge: Recent architectures, notably the Transformer, learn powerful context-aware word representations through layered, multi-headed attention.
Approach: They propose an adaptively sparse Transformer where attention heads have flexible, context-dependent sparsity patterns.
Outcome: The proposed model improves interpretability and head diversity when compared to softmax-based models on machine translation datasets.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations