Kellie Webster

Bernd Bohnet, Vinh Q. Tran, Pat Verga et al. · deepmind, mit

Large language models (LLMs) have shown impressive results while requiring little or no direct supervision. Further, there is mounting evidence that LLMs may have potential in information-seeking scenarios. We believe the ability of an LLM to attribute the text that it generates is likely to be crucial in this setting. We formulate and study Attributed QA as a key first step in the development of attributed LLMs. We propose a reproducible evaluation framework for the task and benchmark a broad set of architectures. We take human annotations as a gold standard and show that a correlated automatic metric is suitable for development. Our experimental work gives concrete answers to two key questions (How to measure attribution?, and How well do current state-of-the-art methods perform on attribution?), and give some hints as to how to address a third (How to build LLMs with attribution?).

Zee Fryer, Vera Axelrod, Ben Packer et al.

A common approach for testing fairness issues in text-based classifiers is through the use of counterfactuals: does the classifier output change if a sensitive attribute in the input is changed? Existing counterfactual generation methods typically rely on wordlists or templates, producing simple counterfactuals that don't take into account grammar, context, or subtle sensitive attribute references, and could miss issues that the wordlist creators had not considered. In this paper, we introduce a task for generating counterfactuals that overcomes these shortcomings, and demonstrate how large language models (LLMs) can be leveraged to make progress on this task. We show that this LLM-based method can produce complex counterfactuals that existing methods cannot, comparing the performance of various counterfactual generation methods on the Civil Comments dataset and showing their value in evaluating a toxicity classifier.

Reinald Kim Amplayo, Kellie Webster, Michael Collins et al.

Large language models (LLMs) have been shown to perform well in answering questions and in producing long-form texts, both in few-shot closed-book settings. While the former can be validated using well-known evaluation metrics, the latter is difficult to evaluate. We resolve the difficulties to evaluate long-form output by doing both tasks at once -- to do question answering that requires long-form answers. Such questions tend to be multifaceted, i.e., they may have ambiguities and/or require information from multiple sources. To this end, we define query refinement prompts that encourage LLMs to explicitly express the multifacetedness in questions and generate long-form answers covering multiple facets of the question. Our experiments on two long-form question answering datasets, ASQA and AQuAMuSe, show that using our prompts allows us to outperform fully finetuned models in the closed book setting, as well as achieve results comparable to retrieve-then-generate open-book models.

Gheorghe Comanici, Eric Bieber, Mike Schaekermann et al. · amazon-science, baidu

In this report, we introduce the Gemini 2.X model family: Gemini 2.5 Pro and Gemini 2.5 Flash, as well as our earlier Gemini 2.0 Flash and Flash-Lite models. Gemini 2.5 Pro is our most capable model yet, achieving SoTA performance on frontier coding and reasoning benchmarks. In addition to its incredible coding and reasoning skills, Gemini 2.5 Pro is a thinking model that excels at multimodal understanding and it is now able to process up to 3 hours of video content. Its unique combination of long context, multimodal and reasoning capabilities can be combined to unlock new agentic workflows. Gemini 2.5 Flash provides excellent reasoning abilities at a fraction of the compute and latency requirements and Gemini 2.0 Flash and Flash-Lite provide high performance at low latency and cost. Taken together, the Gemini 2.X model generation spans the full Pareto frontier of model capability vs cost, allowing users to explore the boundaries of what is possible with complex agentic problem solving.

Nan Du, Yanping Huang, Andrew M. Dai et al.

Scaling language models with more data, compute and parameters has driven significant progress in natural language processing. For example, thanks to scaling, GPT-3 was able to achieve strong results on in-context learning tasks. However, training these large dense models requires significant amounts of computing resources. In this paper, we propose and develop a family of language models named GLaM (Generalist Language Model), which uses a sparsely activated mixture-of-experts architecture to scale the model capacity while also incurring substantially less training cost compared to dense variants. The largest GLaM has 1.2 trillion parameters, which is approximately 7x larger than GPT-3. It consumes only 1/3 of the energy used to train GPT-3 and requires half of the computation flops for inference, while still achieving better overall zero-shot and one-shot performance across 29 NLP tasks.

Maharshi Gor, Kellie Webster, Jordan Boyd-Graber

The goal of question answering (QA) is to answer any question. However, major QA datasets have skewed distributions over gender, profession, and nationality. Despite that skew, model accuracy analysis reveals little evidence that accuracy is lower for people based on gender or nationality; instead, there is more variation on professions (question topic). But QA's lack of representation could itself hide evidence of bias, necessitating QA datasets that better represent global diversity.

Christian Hardmeier, Marta R. Costa-jussà, Kellie Webster et al.

At the Workshop on Gender Bias in NLP (GeBNLP), we'd like to encourage authors to give explicit consideration to the wider aspects of bias and its social implications. For the 2020 edition of the workshop, we therefore requested that all authors include an explicit bias statement in their work to clarify how their work relates to the social context in which NLP systems are used. The programme committee of the workshops included a number of reviewers with a background in the humanities and social sciences, in addition to NLP experts doing the bulk of the reviewing. Each paper was assigned one of those reviewers, and they were asked to pay specific attention to the provided bias statements in their reviews. This initiative was well received by the authors who submitted papers to the workshop, several of whom said they received useful suggestions and literature hints from the bias reviewers. We are therefore planning to keep this feature of the review process in future editions of the workshop.

Tony Sun, Kellie Webster, Apu Shah et al.

Responsible development of technology involves applications being inclusive of the diverse set of users they hope to support. An important part of this is understanding the many ways to refer to a person and being able to fluently change between the different forms as needed. We perform a case study on the singular they, a common way to promote gender inclusion in English. We define a re-writing task, create an evaluation benchmark, and show how a model can be trained to produce gender-neutral English with <1% word error rate with no human-labeled data. We discuss the practical applications and ethical considerations of the task, providing direction for future work into inclusive natural language systems.

Alexander D'Amour, Katherine Heller, Dan Moldovan et al.

ML models often exhibit unexpectedly poor behavior when they are deployed in real-world domains. We identify underspecification as a key reason for these failures. An ML pipeline is underspecified when it can return many predictors with equivalently strong held-out performance in the training domain. Underspecification is common in modern ML pipelines, such as those based on deep learning. Predictors returned by underspecified pipelines are often treated as equivalent based on their training domain performance, but we show here that such predictors can behave very differently in deployment domains. This ambiguity can lead to instability and poor model behavior in practice, and is a distinct failure mode from previously identified issues arising from structural mismatch between training and deployment domains. We show that this problem appears in a wide variety of practical ML pipelines, using examples from computer vision, medical imaging, natural language processing, clinical risk prediction based on electronic health records, and medical genomics. Our results show the need to explicitly account for underspecification in modeling pipelines that are intended for real-world deployment in any domain.

Kellie Webster, Xuezhi Wang, Ian Tenney et al.

Pre-trained models have revolutionized natural language understanding. However, researchers have found they can encode artifacts undesired in many applications, such as professions correlating with one gender more than another. We explore such gendered correlations as a case study for how to address unintended correlations in pre-trained models. We define metrics and reveal that it is possible for models with similar accuracy to encode correlations at very different rates. We show how measured correlations can be reduced with general-purpose techniques, and highlight the trade offs different strategies have. With these results, we make recommendations for training robust models: (1) carefully evaluate unintended correlations, (2) be mindful of seemingly innocuous configuration differences, and (3) focus on general mitigations.

Ana Valeria Gonzalez, Maria Barrett, Rasmus Hvingelby et al.

The one-sided focus on English in previous studies of gender bias in NLP misses out on opportunities in other languages: English challenge datasets such as GAP and WinoGender highlight model preferences that are "hallucinatory", e.g., disambiguating gender-ambiguous occurrences of 'doctor' as male doctors. We show that for languages with type B reflexivization, e.g., Swedish and Russian, we can construct multi-task challenge datasets for detecting gender bias that lead to unambiguously wrong model predictions: In these languages, the direct translation of 'the doctor removed his mask' is not ambiguous between a coreferential reading and a disjoint reading. Instead, the coreferential reading requires a non-gendered pronoun, and the gendered, possessive pronouns are anti-reflexive. We present a multilingual, multi-task challenge dataset, which spans four languages and four NLP tasks and focuses only on this phenomenon. We find evidence for gender bias across all task-language combinations and correlate model bias with national labor market statistics.

Kellie Webster, Emily Pitler

Machine translation systems with inadequate document understanding can make errors when translating dropped or neutral pronouns into languages with gendered pronouns (e.g., English). Predicting the underlying gender of these pronouns is difficult since it is not marked textually and must instead be inferred from coreferent mentions in the context. We propose a novel cross-lingual pivoting technique for automatically producing high-quality gender labels, and show that this data can be used to fine-tune a BERT classifier with 92% F1 for Spanish dropped feminine pronouns, compared with 30-51% for neural machine translation models and 54-71% for a non-fine-tuned BERT model. We augment a neural machine translation model with labels from our classifier to improve pronoun translation, while still having parallelizable translation models that translate a sentence at a time.

Ben Hutchinson, Vinodkumar Prabhakaran, Emily Denton et al.

Building equitable and inclusive NLP technologies demands consideration of whether and how social attitudes are represented in ML models. In particular, representations encoded in models often inadvertently perpetuate undesirable social biases from the data on which they are trained. In this paper, we present evidence of such undesirable biases towards mentions of disability in two different English language models: toxicity prediction and sentiment analysis. Next, we demonstrate that the neural embeddings that are the critical first step in most NLP pipelines similarly contain undesirable biases towards mentions of disability. We end by highlighting topical biases in the discourse about disability which may contribute to the observed model biases; for instance, gun violence, homelessness, and drug addiction are over-represented in texts discussing mental illness.

Hila Gonen, Kellie Webster

The successful application of neural methods to machine translation has realized huge quality advances for the community. With these improvements, many have noted outstanding challenges, including the modeling and treatment of gendered language. While previous studies have identified issues using synthetic examples, we develop a novel technique to mine examples from real world data to explore challenges for deployed systems. We use our method to compile an evaluation benchmark spanning examples for four languages from three language families, which we publicly release to facilitate research. The examples in our benchmark expose where model representations are gendered, and the unintended consequences these gendered representations can have in downstream application.

Kellie Webster, Marta Recasens, Vera Axelrod et al.

Coreference resolution is an important task for natural language understanding, and the resolution of ambiguous pronouns a longstanding challenge. Nonetheless, existing corpora do not capture ambiguous pronouns in sufficient volume or diversity to accurately indicate the practical utility of models. Furthermore, we find gender bias in existing corpora and systems favoring masculine entities. To address this, we present and release GAP, a gender-balanced labeled corpus of 8,908 ambiguous pronoun-name pairs sampled to provide diverse coverage of challenges posed by real-world text. We explore a range of baselines which demonstrate the complexity of the challenge, the best achieving just 66.9% F1. We show that syntactic structure and continuous neural models provide promising, complementary cues for approaching the challenge.