Increasing Faithfulness in Knowledge-Grounded Dialogue with Controllable Features

By Hannah Rashkin, David Reitter, Gaurav Singh Tomar, Dipanjan Das. 2021

119 citations recorded by the source corpus.

Knowledge-grounded dialogue systems are intended to convey information that is based on evidence provided in a given source text. We discuss the challenges of training a generative neural dialogue model for such systems that is controlled to stay faithful to the evidence. Existing datasets contain a mix of conversational responses that are faithful to selected evidence as well as more subjective or chit-chat style responses. We propose different evaluation measures to disentangle these different styles of responses by quantifying the informativeness and objectivity. At training time, additional inputs based on these evaluation measures are given to the dialogue model. At generation time, these additional inputs act as stylistic controls that encourage the model to generate responses that are faithful to the provided evidence. We also investigate the usage of additional controls at decoding time using resampling techniques. In addition to automatic metrics, we perform a human evaluation study where raters judge the output of these controlled generation models to be generally more objective and faithful to the evidence compared to baseline dialogue systems.

Subfields: Artificial Intelligence, Natural Language Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Natural Language Processing

Citations in the corpus

Measuring Attribution in Natural Language Generation Models (2021, methodology): “The general trend of results is similar to what was found in the human evaluations of faithfulness and subjectivity in Rashkin et al. (2021).”
Survey of Hallucination in Natural Language Generation (2022, background): “Controllable generation techniques like control code which could be provided manually [50, 144, 194] or predicted automatically [194], and controlled re-sampling [144] are leveraged to improve faithfulness.”
Factuality Enhanced Language Models for Open-Ended Text Generation (2022, background): “[56] Hannah Rashkin, David Reitter, Gaurav Singh Tomar, and Dipanjan Das.”
TRUE: Re-evaluating Factual Consistency Evaluation (2022, methodology): “, 2021) or training data augmentation for controlled generation (Rashkin et al., 2021b).”
A Survey of Controllable Text Generation Using Transformer-based Pre-trained Language Models (2022, unknown): “…Input:dialoguecontent,additionalstructural information(e.g.,persona,emotion,intent,template,etc.) Output:Dialogueresponse Persona[126,127,157,167,168] politeness[36,92],Sentiment[35,113,129,145], Template[55,139], ground-truthreference[97,105,109,147] Storytelling Input:Storyelements”
Rome was built in 1776: A Case Study on Factual Correctness in Knowledge-Grounded Response Generation (2021, background): “While factual correctness has been studied in knowledge-grounded response generation (Li et al., 2019; Shuster et al., 2021; Honovich et al., 2021; Rashkin et al., 2021), there has been very limited work in creating a human annotated dataset and detecting factual consistency for open-domain dialog…”
Evaluating Attribution in Dialogue Systems: The BEGIN Benchmark (2021, unknown): “Fully Attributable These are responses that convey information that can be completely supported by the provided document; this property has been referred in the literature to as faithfulness (Rashkin et al., 2021b; Maynez et al., 2020b; Dziri et al., 2021; Durmus et al., 2020) and attribution (Rashk”
On the Origin of Hallucinations in Conversational Models: Is it the Datasets or the Models? (2022, methodology): “Contrasting this with human gold responses, the models not only hallucinate but also amplify the percentage of hallucinations, except CTRL on W O W.”
DialFact: A Benchmark for Fact-Checking in Dialogue (2021, background): “However, there is no data available for fact-checking in dialogue, and related work mainly focuses on improving factual consistency on knowledge-grounded response generation task (Honovich et al., 2021; Rashkin et al., 2021; Shuster et al., 2021).”