Utility-Preserving De-Identification for Math Tutoring: Investigating Numeric Ambiguity in the MathEd-PII Benchmark Dataset

Educational dialogue is a valuable but sensitive resource for research: the same transcripts that capture authentic learning often capture personally identifiable information (PII) entangled with curricular content, where "Riemann" may refer to a real student or to a mathematical concept. Existing approaches force a tradeoff between governance and accuracy. Commercial Large Language Models (LLMs) can handle this ambiguity but require sending student data to third parties, while local named entity recognition (NER) systems preserve governance but over-redact curricular terms. We propose a fully local cascade framework that reframes de-identification from open-ended entity recognition to constrained privacy triage. A recall-first union proposer combines two lightweight encoders with deterministic rules to over-generate candidate spans; a context-aware reviewer then makes a binary Redact/Keep decision for each candidate using surrounding dialogue and speaker role. We evaluate three reviewer configurations against same-family LLM-only baselines and a commercial API on math tutoring transcripts from two large platforms. The strongest local configuration reaches 0.958 macro F1, compared with 0.767 for a same-family LLM-only baseline and 0.706 for the commercial API, while running entirely on a single laptop. On a targeted challenge set of curricular-personal name ambiguity, the same configuration degrades by only 0.03 F1 versus 0.19 to 0.25 for smaller reviewers. These results suggest that for educational de-identification, problem formulation matters more than model scale.

Conversational data collected in domains such as healthcare or social sciences is a valuable resource for research and automated analysis. However, responsible data sharing requires the detection and removal of personally identifiable and sensitive information to protect individual privacy. To support the development and evaluation of automatic de-identification systems, we present DialogPII, a multilingual dataset of synthetic dialogs and speech-derived transcripts for personal information detection. DialogPII covers eight interaction scenarios (emergency calls, medical anamnesis interviews, therapy sessions, insurance communication, customer support, clinical interviews regarding an AI-supported dashboard, police reports, and group therapy discussions), 19 entity types, and 11 languages (English, Arabic, Finnish, French, German, Hindi, Italian, Polish, Portuguese, Spanish, and Turkish). Dialogs were generated semi-automatically using large language models, manually curated for plausibility and diversity, and localized to country- and city-specific contexts. All dialogs were additionally converted to speech via text-to-speech synthesis, transcribed with Whisper, and annotated through automatic projection and manual correction, yielding aligned written and speech-derived resources across all languages. We further release baseline multilingual named entity recognition models and provide technical validation through inter-annotator agreement analysis, translation quality evaluation, annotation projection assessment, and benchmark experiments with transformer-based sequence labeling models.

Benchmark infrastructure for personally identifiable information (PII) detection remains limited: existing corpora cover few entity types, use ad hoc generation conditions, and do not show which surface conditions cause detector failures. We present REDACT, a systematically controlled multilingual PII benchmark with 13,427 records, 324,078 entity annotations, 51 entity types, 4,127 surface-form patterns, and 25 languages across 9 scripts. A strength-2 covering-array sampler controls nine generation axes: domain, format, difficulty, length, density, code-switching, language, adjacency, and co-occurrence. Three entity-level metadata fields (disclosure status, disclosure form, and a GDPR-aligned sensitivity tier) enable stratified evaluation beyond aggregate or per-type F1. From the full benchmark, we evaluate five detectors (Presidio, GLiNER, the OpenAI Privacy Filter, GPT-4.1, and Claude Sonnet 4.6) on a locked, language-stratified sample of 1,000 records. Aggregate F1 masks an architecture-dependent failure structure: the rule-based detector performs poorly on the highest-stakes data, including HIGH-sensitivity categories (recall 0.07) and non-verbatim disclosure forms, while the LLM detectors remain more robust, with the HIGH tier as their strongest sensitivity slice. A three-model reference-free LLM-as-judge assessment corroborates that sensitivity-tier assignment is the task's hardest axis. We release the benchmark, schema, prompts, and stratified evaluation harness.