TRILOGUE: A Trilingual Spoken Dialogue Fact-Checking Benchmark with Evidence and Paired Audio

SummaryAI generated

On September 7, 2026, researchers released TRILOGUE, a three-language spoken dialogue fact-checking benchmark dataset covering English, Russian, and Kazakh. This dataset includes 12,000 dialogue pairs, 187,000 rounds of interaction, and 390 hours of paired audio recordings, with ASR transcripts aligned with word-level timestamps. It also contains nearly 5,000 human-recorded Russian and Kazakh dialogue files. TRILOGUE supports verifiability checks, source article retrieval, and verification tasks based on statements, gold evidence, and retrieved evidence. Baseline experiments show that ASR degradation and cross-language transfer are particularly difficult in Kazakh, but using the retrieved source evidence can significantly narrow the gap with gold-evidence verification.

Related eventsRELATED EVENTS

Key entitiesKEY ENTITIES

TRILOGUE

SignalsSIGNALS

Keyword heat

TRILOGUE1

All reports (1)SOURCES

↗

A arXiv cs.CL en 2026-09-07 12:00

The researchers released TRILOGUE, a benchmark dataset for verification of three-language spoken dialogue facts, containing 12,000 sets of dialogues, 187,000 rounds of interaction, and 390 hours of paired audio. This dataset covers English, Russian, and Kazakh languages, provides ASR transcripts and word-level timestamp alignments, and includes nearly 5,000 human-recorded Russian and Kazakh dialogue files. TRILOGUE supports verification detection, retrieval of source articles, and validation tasks based on statements, gold evidence, and retrieved evidence. Baseline experiments show that ASR degradation and cross-language transfer are particularly difficult in Kazakh, and the use of retrieved source evidence can significantly reduce the gap with gold evidence validation.