Comet is a dataset which contains 11.5k user-assistant dialogs (totalling 103k utterances), grounded in simulated personal memory graphs.
1 PAPER • NO BENCHMARKS YET
ComFact is a benchmark for commonsense fact linking, where models are given contexts and trained to identify situationally-relevant commonsense knowledge from KGs. The novel benchmark, C-om-Fact, contains ∼293k in-context relevance annotations for common-sense triplets across four stylistically diverse dialogue and storytelling datasets.
5 PAPERS • NO BENCHMARKS YET
Most existing dialogue systems fail to respond properly to potentially unsafe user utterances by either ignoring or passively agreeing with them.
13 PAPERS • 1 BENCHMARK
KETOD (Knowledge-Enriched Task-Oriented Dialogue) is a dataset containing system responses designed for enriching task-oriented dialogues with chit-chat based on relevant entity knowledge. There are a total of 5,324 dialogues with enriched system responses.
4 PAPERS • NO BENCHMARKS YET
ArgSciChat is an argumentative dialogue dataset. It consists of 498 messages collected from 41 dialogues on 20 scientific papers. It can be used to evaluate conversational agents and further encourage research on argumentative scientific agents.
1 PAPER • 2 BENCHMARKS
The main goal of the data collection is to acquire highly natural conversations that cover a wide variety of styles and scenarios. In total, the presented corpus consists of five domains: Food, Hotel, Nightlife, Shopping mall and Sightseeing. Controlled by our various task settings, the collected dialogues cover between one to four domains per dialogue, and are thus of greatly varying length and complexity. There are 808 single-task dialogues that contains a single venue target and 4, 298 multi-task dialogues consisting of at least two to four venue targets. These different venues vary in domains most of the times.
3 PAPERS • 2 BENCHMARKS
Next generation task-oriented dialog systems need to understand conversational contexts with their perceived surroundings, to effectively help users in the real-world multimodal environment. Existing task-oriented dialog datasets aimed towards virtual assistance fall short and do not situate the dialog in the user's multimodal context. To overcome, we present a new dataset for Situated and Interactive Multimodal Conversations, SIMMC 2.0, which includes 11K task-oriented user<->assistant dialogs (117K utterances) in the shopping domain, grounded in immersive and photo-realistic scenes. The dialogs are collected using a two-phase pipeline: (1) A novel multimodal dialog simulator generates simulated dialog flows, with an emphasis on diversity and richness of interactions, (2) Manual paraphrasing of the generated utterances to collect diverse referring expressions. We provide an in-depth analysis of the collected dataset, and describe in detail the four main benchmark tasks we propose. Our
12 PAPERS • 2 BENCHMARKS