Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 1801.01828

Shielding Google's language toxicity model against adversarial attacks

Nestor Rodriguez·Sergio Rojas-Galeano
View original source ↗Read source document ↗Export metadata ↓

Abstract

Lack of moderation in online communities enables participants to incur in personal aggression, harassment or cyberbullying, issues that have been accentuated by extremist radicalisation in the contemporary post-truth politics scenario. This kind of hostility is usually expressed by means of toxic language, profanity or abusive statements. Recently Google has developed a machine-learning-based toxicity model in an attempt to assess the hostility of a comment; unfortunately, it has been suggested that said model can be deceived by adversarial attacks that manipulate the text sequence of the comment. In this paper we firstly characterise such adversarial attacks as using obfuscation and polarity transformations. The former deceives by corrupting toxic trigger content with typographic edits, whereas the latter deceives by grammatical negation of the toxic content. Then, we propose a two--stage approach to counter--attack these anomalies, bulding upon a recently proposed text deobfuscation method and the toxicity scoring model. Lastly, we conducted an experiment with approximately 24000 distorted comments, showing how in this way it is feasible to restore toxicity of the adversarial variants, while incurring roughly on a twofold increase in processing time. Even though novel adversary challenges would keep coming up derived from the versatile nature of written language, we anticipate that techniques combining machine learning and text pattern recognition methods, each one targeting different layers of linguistic features, would be needed to achieve robust detection of toxic language, thus fostering aggression--free digital interaction.

Explore related subjects

cs.CLcs.SI

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Nestor Rodriguez, Sergio Rojas-Galeano. 2018-01-05. Shielding Google's language toxicity model against adversarial attacks. https://arxiv.org/abs/1801.01828

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Hierarchical attention interpretation: an interpretable speech-level transformer for bi-modal depression detection

Depression is a common mental disorder. Automatic depression detection tools using speech, enabled by machine learning, help early screening of depression. This paper addresses two limitations that may hinder the clinical implementations of such tools: noise resulting from segment-level labelling and a lack of model interpretability. We propose a bi-modal speech-level transformer to avoid segment-level labelling and introduce a hierarchical interpretation approach to provide both speech-level and sentence-level interpretations, based on gradient-weighted attention maps derived from all attention layers to track interactions between input features. We show that the proposed model outperforms a model that learns at a segment level ($p$=0.854, $r$=0.947, $F1$=0.897 compared to $p$=0.732, $r$=0.808, $F1$=0.768). For model interpretation, using one true positive sample, we show which sentences within a given speech are most relevant to depression detection; and which text tokens and Mel-spectrogram regions within these sentences are most relevant to depression detection. These interpretations allow clinicians to verify the validity of predictions made by depression detection tools, promoting their clinical implementations.

cs.CL↗
arXiv2026

Cultural Alignment in Large Language Models Using Soft Prompt Tuning

Large Language Model (LLM) alignment is commonly achieved through supervised fine-tuning or reinforcement learning, both of which require labeled or preference data and update model weights. Without targeted cultural adaptation, however, deployed LLMs often exhibit culturally homogeneous behavior that fails to reflect diverse local values. Aligning models to cultural value frameworks such as Hofstede's Value Survey Module (VSM13) presents a distinct challenge: alignment signals are available only as aggregated survey-level scores computed after generating responses to an entire survey, providing no per-token gradient and requiring no preference data by construction. This makes standard gradient-based alignment methods ill-suited to the task. We propose a deployment-friendly approach that encodes cultural behavior in short, tunable soft prompts optimized with Differential Evolution (DE), while keeping model weights frozen and requiring no preference data. At inference, the system inserts the appropriate cultural-specific prompt to adapt model responses for different cultures. Experiments across four countries and four instruction-tuned models show that DE-optimized prompts generally reduce discrepancy with VSM13 reference profiles, improve rank agreement with the World Values Survey (WVS), an independent framework not seen during optimization, and are preferred in blinded pairwise evaluations using majority voting across three LLM judges.

cs.CL↗
arXiv2026

VQ-Logits: Compressing the Output Bottleneck of Large Language Models via Vector Quantized Logits

Large Language Models (LLMs) have achieved remarkable success but face significant computational and memory challenges, particularly due to their extensive output vocabularies. The final linear projection layer, mapping hidden states to vocabulary-sized logits, often constitutes a substantial portion of the model's parameters and computational cost during inference. Existing methods like adaptive softmax or hierarchical softmax introduce structural complexities. In this paper, we propose VQ-Logits, a novel approach that leverages Vector Quantization (VQ) to drastically reduce the parameter count and computational load of the LLM output layer. VQ-Logits replaces the large V * dmodel output embedding matrix with a small, shared codebook of K embedding vectors (K << V ). Each token in the vocabulary is mapped to one of these K codebook vectors. The LLM predicts logits over this compact codebook, which are then efficiently "scattered" to the full vocabulary space using the learned or preassigned mapping. We demonstrate through extensive experiments on standard language modeling benchmarks (e.g., WikiText-103, C4) that VQ-Logits can achieve up to 99% parameter reduction in the output layer and 6x speedup in logit computation, with only a marginal 4% increase in perplexity compared to full softmax baselines. We further provide detailed ablation studies on codebook size, initialization, and learning strategies, showcasing the robustness and effectiveness of our approach.

cs.CL↗