Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper systematically analyzes the performance degradation of offensive language detection models when applied across different datasets and languages, identifying the primary causes of this degradation through a novel framework. The authors introduce a zero-shot transfer loss decomposition to distinguish between dataset and language effects, and propose joint training strategies that balance multilingual capability with source-task performance. Key findings reveal that dataset effects are the dominant factor in performance loss, and that certain joint training approaches can significantly enhance multilingual capabilities while minimizing damage to source-task performance.
Dataset effects are the primary culprit behind the performance drop in offensive language detection, overshadowing language effects and revealing critical trade-offs in model training strategies.
Offensive language detection models generally suffer performance degradation when deployed across datasets and across languages, yet most existing studies stop at reporting this phenomenon and lack a systematic methodology for decomposing the causes of degradation into attributable components and quantifying the cost of remediation. This paper proposes a diagnosis and optimization framework composed of three coordinated technical components. First, a zero-shot transfer loss decomposition that separates the performance degradation from OLID to MLMA into two independently measurable components, namely dataset effect and language effect. Second, a controlled fine-tuning protocol that quantifies both adaptation efficiency and the hidden damage inflicted on the source task by comparing few shot learning curves under continued fine-tuning and cold-start starting points. Third, three joint training strategies incorpo rating temperature sampling and experience replay, which offer a controllable Pareto trade-off between improving multilingual capability and preserving source-task performance. Experiments built on this framework show that the dataset effect dominates the zero-shot transfer loss and substantially outweighs the language effect. Few-shot adaptation without a replay mechanism, though data-efficient, inflicts source task damage 4 to 9 times greater than that of the joint training strategies, and its damage magnitude is highly unstable. The three joint training strategies trade 3.2 to 4.1 percentage points of source-task performance for 8.1 to 42.6 percentage points of multilingual capability gain, forming a clear and controllable Pareto trade-off.