David Reitter
David Reitter studies natural language processing in the contexts of dialogue and generative models. His most recent work has solved problems in generating grounded text. Such text appears, for example, in certain chat or summarization systems. Grounded text only makes claims that are based on available evidence. Dr. Reitter has studied ways to detect claims that are not grounded in evidence; these evaluation methods have become the basis for generative models across many Google products.
Reitter's research interests cover diverse areas of computational cognitive science. Reitter, with academic colleagues, started a subfield of psycholinguistics that used large-scale observational datasets to understand how the mind processes language and allows us to engage in conversation.
David Reitter has authored more than 120 papers in both cognitive psychology and computer science, and Aquamacs, a widely used software package. Prof. Reitter joined Google from Penn State, where he directed an NSF-funded research group on computational cognition and language processing. He holds a Ph.D. from the University of Edinburgh, prior degrees in linguistics and computer science, and was a fellow at MIT's Media Lab Europe (working on multimodal user interfaces) and a post-doc at Carnegie Mellon University (working on cognitive modeling).
Research Areas
Authored Publications
Sort By
Preview abstract
Knowledge-grounded dialogue generation is a challenging task because it requires satisfying two fundamental yet often competing constraints: being responsive in a manner that is specific to what the conversation partner has said while also being attributable to an underlying source document. In this work, we bring this trade-off between these two objectives (specificity and attribution) to light and ask the question: Can explicit content planning before the response generation help the model to address this challenge? To answer this question, we design a framework called PLEDGE, which allows us to experiment with various plan variables explored in prior work, supporting both metric-agnostic and metric-aware approaches. While content planning shows promise, our results on whether it can actually help to navigate this trade-off are mixed -- planning mechanisms that are metric-aware (use automatic metrics during training) are better at automatic evaluations but underperform in human judgment compared to metric-agnostic mechanisms. We discuss how this may be caused by over-fitting to automatic metrics and the need for future work to better calibrate these metrics towards human judgment. We hope the observations from our analysis will inform future work that aims to apply content planning in this context.
View details
Measuring Attribution in Natural Language Generation Models
Iulia Turc
Computational Linguistics, 49 (2023), pp. 777-840
Preview abstract
With recent improvements in natural language generation (NLG) models for various applications, it has become imperative to have the means to identify and evaluate whether NLG output is only sharing verifiable information about the external world. In this work, we present a new evaluation framework entitled Attributable to Identified Sources (AIS) for assessing the output of natural language generation models, when such output pertains to the external world. We first define AIS and introduce a two-stage annotation pipeline for allowing annotators to appropriately evaluate model output according to AIS guidelines. We empirically validate this approach on generation datasets spanning three tasks (two conversational QA datasets, a summarization dataset, and a table-to-text dataset) via human evaluation studies that suggest that AIS could serve as a common framework for measuring whether model-generated statements are supported by underlying sources. We release guidelines for the human evaluation studies.
View details
Preview abstract
Despite recent progress, it has been difficult to prevent semantic hallucinations in generative Large Language Models. One common solution to this is augmenting LLMs with a retrieval system and making sure that the generated output is attributable to the retrieved information. Given this new added constraint, it is plausible to expect that the overall quality of the output will be affected, for example, in terms of fluency. Can scaling language models help?
Here we examine the relationship between fluency and attribution in LLMs prompted with retrieved evidence in knowledge-heavy dialog settings. Our experiments were implemented with a set of auto-metrics that are aligned with human preferences. They were used to evaluate a large set of generations, produced under varying parameters of LLMs and supplied context.
We show that larger models tend to do much better in both fluency and attribution, and that (naively) using top-k retrieval versus top-1 retrieval improves attribution but hurts fluency. We next propose a recipe that could allow smaller models to both close the gap with larger models and preserve the benefits of top-k retrieval while avoiding its drawbacks.
View details
Evaluating Attribution in Dialogue Systems: The BEGIN Benchmark
Nouha Dziri
Tal Linzen
Transactions of the Association for Computational Linguistics, 10 (2022), 1066–1083
Preview abstract
Knowledge-grounded dialogue systems powered by large language models often generate responses that, while fluent, are not attributable to a relevant source of information. Progress towards models that do not exhibit this issue requires evaluation metrics that can quantify its prevalence. To this end, we introduce the Benchmark for Evaluation of Grounded INteraction (Begin), comprising 12k dialogue turns generated by neural dialogue systems trained on three knowledge-grounded dialogue corpora. We collect human annotations assessing the extent to which the models’ responses can be attributed to the given background information. We then use Begin to analyze eight evaluation metrics. We find that these metrics rely on spurious correlations, do not reliably distinguish attributable abstractive responses from unattributable ones, and perform substantially worse when the knowledge source is longer. Our findings underscore the need for more sophisticated and robust evaluation metrics for knowledge-grounded dialogue. We make Begin publicly available at https://github.com/google/BEGIN-dataset.
View details
CONQRR: Conversational Query Rewriting for Retrieval with Reinforcement Learning
Ellen Wu
Yi Luan
Hannaneh Hajishirzi
Mari Ostendorf
The 2022 Conference on Empirical Methods in Natural Language Processing (2022)
Preview abstract
Compared to standard retrieval tasks, passage retrieval for conversational question answering (CQA) poses new challenges in understanding the current user question, as each question needs to be interpreted within the dialogue context. Moreover, it can be expensive to re-train well-established retrievers such as search engines that are originally developed for non-conversational queries. To facilitate their use, we develop a query rewriting model CONQRR that rewrites a conversational question in the context into a standalone question. It is trained with a novel reward function to directly optimize towards retrieval using reinforcement learning and can be adapted to any off-the-shelf retriever. CONQRR achieves state-of-the-art results on a recent open-domain CQA dataset containing conversations from three different sources, and is effective for two different off-the-shelf retrievers. Our extensive analysis also shows the robustness of CONQRR to out-of-domain dialogues as well as to zero query rewriting supervision.
View details
Dungeons and Dragons as a Challenge Problem for Artificial Intelligence
Chris Callison-Burch
Lara Martin
NAACL Wordplay Workshop, ACL (2022) (to appear)
Preview abstract
AI researchers have posited Dungeons and Dragons (D&D) as a challenge problem to test systems on various language-related capabilities. In this paper, we frame D&D specifically as a dialogue system challenge, where the tasks are to both generate the next conversational turn in the game and predict the state of the game given the dialogue history. We create a gameplay dataset consisting of nearly 900 games, with a total of 7,000 players, 800,000 dialogue turns, 500,000 dice rolls, and 58 million words. We automatically annotate the data with partial state information about the game play. We train a large language model to generate the next game turn, conditioning it on different information. The LM can respond as a particular character or as the player who runs the game—i.e., the Dungeon Master (DM). It is trained to produce dialogue that is either in-character (roleplaying in the fictional world) or out-of-character (discussing rules or strategy). We perform a human evaluation to determine what factors make the generated output plausible and interesting. We further perform an automatic evaluation to determine how well the model can predict the game state given the history and examine how well tracking the game state improves its ability to produce plausible conversational output.
View details
Increasing Faithfulness in Knowledge-Grounded Dialogue with Controllable Features
Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers) (2021), pp. 704-718
Preview abstract
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.
View details
Preview abstract
Recent models of language have eliminated syntactic-semantic dividing lines. We explore the psycholinguistic implications of this development by comparing different types of sentence embeddings in their ability to encode syntactic constructions. Our study uses contrasting sentence structures known to cause syntactic priming effects, that is, the tendency in humans to repeat sentence structures after recent exposure. We compare how syntactic alternatives are captured by sentence embeddings produced by a neural language model (BERT) or by the composition of word embeddings (BEAGLE, HHM, GloVe). Dative double object vs. prepositional object and active vs. passive sentences are separable in the high-dimensional space of the sentence embeddings and can be classified with a high degree of accuracy. The results lend empirical support to the modern, computational, integrated accounts of semantics and syntax, and they shed light on the information stored at different layers in deep language models such as BERT.
View details
Indirect Associations in Learning Semantic and Syntactic Lexical Relationships
Matthew A. Kelly
Moojan Ghafurian
Robert L. West
Journal of Memory and Language, 115 (2020), pp. 104153
Preview abstract
Computational models of distributional semantics (a.k.a. word embeddings) represent a word’s meaning in terms of its relationships with all other words. We examine what grammatical information is encoded in distributional models and investigate the role of indirect associations. Distributional models are sensitive to associations between words at one degree of separation, such as ‘tiger’ and ‘stripes’, or two degrees of separation, such as ‘soar’ and ‘fly’. By recursively adding higher levels of representations to a computational, holographic model of semantic memory, we construct a distributional model sensitive to associations between words at arbitrary degrees of separation. We find that word associations at four degrees of separation increase the similarity assigned by the model to English words that share part-of-speech or syntactic type. Word associations at four degrees of separation also improve the ability of the model to construct grammatical English sentences. Our model proposes that human memory uses indirect associations to learn part-of-speech and that the basic associative mechanisms of memory and learning support knowledge of both semantics and grammatical structure.
View details
Preview abstract
This paper addresses the role of personality characteristics in decisions on the timing of an action, such as in the context of security and safety choices. Examples of such decisions include when to check log files for intruders and when to monitor financial accounts for fraud or errors. Two behavioral studies (n = 461) are conducted. Individual risk propensity and need for cognition are obtained via scales. The task is a game against an opaque computer opponent in which participants make decisions about the timing of actions in response to an unknown external risk factor. The task is not payoff-neutral w.r.t. risk. Difficulty is varied through the availability of explicitly given or decision-critical information, which is observable visually (Study 1) or in temporal memory (Study 2). Across this problem space, we find that risk propensity is not generally a hindrance in timing tasks. Participants of average risk propensity generally benefit from a high need for cognition, particularly when externalized memory is available, as in Study 1. In the more difficult temporal-estimation task, need for cognition was associated with increased payoffs from task experience. In both tasks, higher risk propensity in participants was associated with increased improvements in payoffs from task experience.
View details
