CHiPS: Character Histograms and Positional Signals for Lightweight Authorship Attribution in Romanian Texts

2026-07-24 · Source: Computation and Language · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, quick

Summary

CHiPS is a lightweight character-level authorship attribution method specifically designed for Romanian texts, operating in a closed-set experimental setup where the true author is among candidates. The method integrates two distinct writing style fingerprints: CH-SVM, a character-histogram classifier based on one-character marginal distributions, and FFT12-LR, a positional-signal classifier that uses Fourier/Welch spectral descriptors from impulse trains of selected characters and punctuation. Notably, CHiPS requires no tokenization, syntactic analysis, pretrained language models, or transformer fine-tuning, and avoids character n-grams with n >= 2. On a locked ROST split of 400 files from 392 source-text groups by 10 authors, CHiPS-F achieved 0.9310 accuracy and 0.9341 macro-F1. A variant, CHiPS-R, reached 0.8919 accuracy and 0.8708 macro-F1 on the ROSTories-cleaned corpus (1,248 files, 1,240 groups, 19 authors). The research emphasizes exploring the limits of restricted, transparent character evidence under strict leakage control, rather than claiming peak classification accuracy.

Key takeaway

For NLP Engineers developing authorship attribution systems for low-resource languages like Romanian, CHiPS demonstrates that high accuracy is achievable without heavy computational resources or complex models. You should consider character-level features and positional signals as a lightweight, transparent alternative to transformer-based approaches, especially when strict leakage control and interpretability are priorities. This method avoids tokenization and large pretrained models, simplifying deployment.

Key insights

CHiPS offers a lightweight, tokenization-free authorship attribution method using character histograms and positional signals for Romanian texts.

Principles

• Authorship can be attributed without complex NLP.
• Character-level features offer robust style fingerprints.
• Combining diverse style signals improves accuracy.

Method

CHiPS combines CH-SVM (one-character marginal distributions) and FFT12-LR (Fourier/Welch spectral descriptors from character/punctuation impulse trains). An optional reranker can be used.

In practice

• Apply character-level features for low-resource languages.
• Use CH-SVM for simple, transparent style analysis.
• Explore positional signals for subtle stylistic cues.

Topics

• Authorship Attribution
• Character-level Features
• Romanian Language Processing
• Fourier Transform
• Machine Learning
• Stylometry
• Lightweight Models

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• How Far Can Classical NLP Go? From Bag-of-Words to Stacking on Spooky Author Identification — Classical NLP, with careful feature engineering and stacking, excels at stylistic authorship attribution.
• Letter Lemmatization: One-to-one and Banded RNNs for Reversing Character-Set Simplification and Abbreviation in Medieval Text — Character-level RNNs and similarity heuristics effectively manage character set variations and abbreviations in historical texts, even with limited data.
• Osint at SemEval-2026 Task 13: A Distribution-Aware Framework for Machine-Generated Code Detection and Multi-Source Authorship Attribution — Long-context modeling, distribution-aware sampling, and noise-robust training are crucial for reliable machine-generated code detection.
• Detecting LLM-Generated Tokens in Human--LLM Coauthored Text — A token-level detection method smooths adjacent scores using an adaptive Lepski-type rule to identify LLM-generated content in coauthored text.
• Team Vivek Dhayaal at SemEval-2026 Task 13 Subtask B: Multi-Class Authorship Detection — Competitive conspiracy detection is achievable with lightweight DistilRoBERTa models and minimal resources.
• Lightweight Person-Place Relation Extraction from Historical Newspapers with Dependency Graphs and Proximity Features — Minimum character distance is the dominant signal for person-place relation extraction in historical texts, often outperforming complex features.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computation and Language.