Papers by Irina Piontkovskaya

12 papers
GEC-DePenD: Non-Autoregressive Grammatical Error Correction with Decoupled Permutation and Decoding (2023.acl-long)

Copied to clipboard

Challenge: grammatical error correction is an important NLP task that is usually solved with autoregressive sequence-to-sequence models.
Approach: They propose a non-autoregressive approach to grammatical error correction that decouples a permutation network and a decoder network that fills in specific tokens.
Outcome: The proposed approach improves over previously known non-autoregressive methods and reaches the level of autoregressive approaches that do not use language-specific synthetic data generation methods.
SumTitles: a Summarization Dataset with Low Extractiveness (2020.coling-main)

Copied to clipboard

Challenge: Existing methods for extractive summarization of dialogue data are limited by the grammar and structure of the utterances used.
Approach: They propose a low-extractive corpus of movie dialogues for abstractive text summarization . they use an alignment algorithm to construct the corpus and a baseline evaluation .
Outcome: The proposed method is low-extractive and shows high performance in dialogue datasets.
CCT-Code: Cross-Consistency Training for Multilingual Clone Detection and Code Search (2025.naacl-srw)

Copied to clipboard

Challenge: clone detection is crucial in software development for identifying semantically similar code . clones can be found in the same language code snippets, but there is little research on multilingual clonage detection.
Approach: They propose a novel training procedure leveraging cross-lingual similarity to train language models on source code in various programming languages.
Outcome: The proposed method achieves state-of-the-art on C++ and Python clone detection benchmarks with comparable performance on decoder-based models.
Efficient Grammatical Error Correction Via Multi-Task Training and Optimized Training Schedule (2023.emnlp-main)

Copied to clipboard

Challenge: Recent research has focused on using synthetic data for grammatical error correction . lack of annotated training data hinders progress in the field .
Approach: They propose auxiliary tasks that exploit alignment between original and corrected sentences . they propose a sequence-to-sequence problem and perform multi-task training .
Outcome: The proposed auxiliary tasks outperform the best models with a BART-based model on 11B parameters.
Robust AI-Generated Text Detection by Restricted Embeddings (2024.findings-emnlp)

Copied to clipboard

Challenge: Existing approaches for artificial text detection are score-based and classifier-based . however, score-driven methods often rely on a score-derived score.
Approach: They investigate the ability of classifier-based detectors to transfer to unseen generators or semantic domains.
Outcome: The proposed methods improve the out-of-distribution classification score by up to 9% and 14%.
Acceptability Judgements via Examining the Topology of Attention Maps (2022.findings-emnlp)

Copied to clipboard

Challenge: Acceptability judgments are a key component of generative linguistics, but their ability to judge grammatical acceptability has not been explored.
Approach: They propose to exploit the geometric properties of the attention graph to evaluate the grammatical acceptability of sentences using topological data analysis.
Outcome: The proposed approach outperforms nine statistical and Transformer LM baselines on the BLiMP benchmark and the human-level performance on the same benchmark.
Unveiling Intrinsic Dimension of Texts: from Academic Abstract to Creative Story (2026.eacl-long)

Copied to clipboard

Challenge: a new study grounding intrinsic dimension in interpretable text properties is published . entropy-like measures are ubiquitous in training and evaluation, but geometric complexity remains underexplored.
Approach: They propose to ground intrinsic dimension (ID) in interpretable text properties through cross-encoder analysis, linguistic features, and sparse autoencodes.
Outcome: The proposed method shows that scientific prose shows low ID ( 8), encyclopedic content medium ID ( > 9), creative/opinion writing high ID (> 10.5)
Artificial Text Detection via Examining the Topology of Attention Maps (2021.emnlp-main)

Copied to clipboard

Challenge: Existing methods for text detection lack interpretability and robustness towards unseen models.
Approach: They propose three new types of interpretable topological features based on topological data analysis which is currently understudied in the field of NLP.
Outcome: The proposed features outperform count- and neural-based baselines up to 10% on three common datasets and tend to be the most robust towards unseen GPT-style generation models.
Feature-Level Insights into Artificial Text Detection with Sparse Autoencoders (2025.findings-acl)

Copied to clipboard

Challenge: Existing algorithms for AI text detection lack interpretability, limiting their reliability in highstakes applications.
Approach: They extend existing ATD frameworks by using Sparse Autoencoders to extract features from Gemma-2-2b residual stream.
Outcome: The proposed algorithms can extract human-interpretable features from Gemma-2-2b model.
AudioSAE: Towards Understanding of Audio-Processing Models with Sparse AutoEncoders (2026.eacl-long)

Copied to clipboard

Challenge: Feature steering reduces Whisper’s false speech detections by 70% with negligible WER increase, demonstrating real-world applicability.
Approach: They train Sparse Autoencoders across all encoder layers of Whisper and HuBERT and evaluate their stability, interpretability, and practical utility.
Outcome: The proposed models capture general acoustic and semantic information as well as specific events, including environmental noises and paralinguistic sounds, and disentangle them effectively.
Quantifying Logical Consistency in Transformers via Query-Key Alignment (2025.emnlp-main)

Copied to clipboard

Challenge: Existing solutions for multi-step logical reasoning are unreliable . Existing methods generate intermediate steps but provide no internal check of coherence .
Approach: They propose a method that uses internal Query-Key interactions within transformer attention heads as a proxy for logical consistency.
Outcome: The proposed method reveals latent reasoning structure in large language models and provides a mechanistic alternative to ablation-based analysis.
Ask Me Anything in Your Native Language (2022.naacl-main)

Copied to clipboard

Challenge: Cross-lingual question answering systems are becoming more and more important . a new approach can be generalized to more than 20 languages and outperforms previous models by 12% .
Approach: They propose a cross-lingual question answering system that can be generalized to more than 20 languages . their approach can outperform previous models by 12% on multiple languages based on a dataset .
Outcome: The proposed approach outperforms the previous models on multiple languages by 12% . it can be generalized to more than 20 languages and outperformed all previous models by 2% .

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