Thibault Sellam

Bio: Thibault Sellam is an academic researcher from Google. The author has contributed to research in topics: Machine translation & SQL. The author has an hindex of 10, co-authored 37 publications receiving 712 citations. Previous affiliations of Thibault Sellam include Columbia University & Centrum Wiskunde & Informatica.

BLEURT: Learning Robust Metrics for Text Generation

Abstract: Text generation has made significant advances in the last few years. Yet, evaluation metrics have lagged behind, as the most popular choices (e.g., BLEU and ROUGE) may correlate poorly with human judgments. We propose BLEURT, a learned evaluation metric based on BERT that can model human judgments with a few thousand possibly biased training examples. A key aspect of our approach is a novel pre-training scheme that uses millions of synthetic examples to help the model generalize. BLEURT provides state-of-the-art results on the last three years of the WMT Metrics shared task and the WebNLG Competition dataset. In contrast to a vanilla BERT-based approach, it yields superior results even when the training data is scarce and out-of-distribution.

BLEURT: Learning Robust Metrics for Text Generation

Abstract: Text generation has made significant advances in the last few years. Yet, evaluation metrics have lagged behind, as the most popular choices (e.g., BLEU and ROUGE) may correlate poorly with human judgment. We propose BLEURT, a learned evaluation metric for English based on BERT. BLEURT can model human judgment with a few thousand possibly biased training examples. A key aspect of our approach is a novel pre-training scheme that uses millions of synthetic examples to help the model generalize. BLEURT provides state-of-the-art results on the last three years of the WMT Metrics shared task and the WebNLG data set. In contrast to a vanilla BERT-based approach, it yields superior results even when the training data is scarce and out-of-distribution.

Sticking to the Facts: Confident Decoding for Faithful Data-to-Text Generation.

Abstract: We address the issue of hallucination in data-to-text generation, i.e., reducing the generation of text that is unsupported by the source. We conjecture that hallucination can be caused by an encoder-decoder model generating content phrases without attending to the source; so we propose a confidence score to ensure that the model attends to the source whenever necessary, as well as a variational Bayes training framework that can learn the score from data. Experiments on the WikiBio (Lebretet al., 2016) dataset show that our approach is more faithful to the source than existing state-of-the-art approaches, according to both PARENT score (Dhingra et al., 2019) and human evaluation. We also report strong results on the WebNLG (Gardent et al., 2017) dataset.

Repairing the Cracked Foundation: A Survey of Obstacles in Evaluation Practices for Generated Text

Abstract: Evaluation practices in natural language generation (NLG) have many known flaws, but improved evaluation approaches are rarely widely adopted. This issue has become more urgent, since neural generation models have improved to the point where their outputs can often no longer be distinguished based on the surface-level features that older metrics rely on. This paper surveys the issues with human and automatic model evaluations and with commonly used datasets in NLG that have been pointed out over the past 20 years. We summarize, categorize, and discuss how researchers have been addressing these issues and what their findings mean for the current state of model evaluations. Building on those insights, we lay out a long-term vision for evaluation research and propose concrete steps for researchers to improve their evaluation processes. Finally, we analyze 66 generation papers from recent NLP conferences in how well they already follow these suggestions and identify which areas require more drastic changes to the status quo.

Meet Charles, big data query advisor

Abstract: In scientific data management and business analytics, the most informative queries are a holy grail. Data collection becomes increasingly simpler, yet data exploration gets significantly harder. Exploratory querying is likely to return an empty or an overwhelming result set. On the other hand, data mining algorithms require extensive preparation, ample time and do not scale well. In this paper, we address this challenge at its core, i.e., how to query the query space associated with a given database. The space considered is formed by conjunctive predicates. To express them, we introduce the Segmentation Description Language (SDL). The user provides a query. Charles, our query advisory system, breaks its extent into meaningful segments and returns the subsequent SDL descriptions. This provides insight into the set described and offers the user directions for further exploration. We introduce a novel algorithm to generate SDL answers. We evaluate them using four orthogonal criteria: homogeneity, simplicity, breadth, and entropy. A prototype implementation has been constructed and the landscape of follow-up research is sketched.

Pattern Recognition and Machine Learning

Abstract: Probability Distributions.- Linear Models for Regression.- Linear Models for Classification.- Neural Networks.- Kernel Methods.- Sparse Kernel Machines.- Graphical Models.- Mixture Models and EM.- Approximate Inference.- Sampling Methods.- Continuous Latent Variables.- Sequential Data.- Combining Models.

When is nearest neighbor meaningful

Abstract: We explore the effect of dimensionality on the nearest neighbor problem. We show that under a broad set of conditions (much broader than independent and identically distributed dimensions), as dimensionality increases, the distance to the nearest data point approaches the distance to the farthest data point. To provide a practical perspective, we present empirical results on both real and synthetic data sets that demonstrate that this effect can occur for as few as 10-15 dimensions. These results should not be interpreted to mean that high-dimensional indexing is never meaningful; we illustrate this point by identifying some high-dimensional workloads for which this effect does not occur. However, our results do emphasize that the methodology used almost universally in the database literature to evaluate high-dimensional indexing techniques is flawed, and should be modified. In particular, most such techniques proposed in the literature are not evaluated versus simple linear scan, and are evaluated over workloads for which nearest neighbor is not meaningful. Often, even the reported experiments, when analyzed carefully, show that linear scan would outperform the techniques being proposed on the workloads studied in high (10-15) dimensionality!.

PaLM: Scaling Language Modeling with Pathways

Abstract: Large language models have been shown to achieve remarkable performance across a variety of natural language tasks using few-shot learning , which drastically reduces the number of task-specific training examples needed to adapt the model to a particular application. To further our understanding of the impact of scale on few-shot learning, we trained a 540-billion parameter, densely activated, Transformer language model, which we call Pathways Language Model (PaLM). We trained PaLM on 6144 TPU v4 chips using Pathways, a new ML system which enables highly efficient training across multiple TPU Pods. We demonstrate continued benefits of scaling by achieving state-of-the-art few-shot learning results on hundreds of language understanding and generation benchmarks. On a number of these tasks, PaLM 540B achieves breakthrough performance, outperforming the finetuned state-of-the-art on a suite of multi-step reasoning tasks, and outperforming average human performance on the recently released BIG-bench benchmark. A significant number of BIG-bench tasks showed discontinuous improvements from model scale, meaning that performance steeply increased as we scaled to our largest model. PaLM also has strong capabilities in multilingual tasks and source code generation, which we demonstrate on a wide array of benchmarks. We additionally provide a comprehensive analysis on bias and toxicity, and study the extent of training data memorization with respect to model scale. Finally, we discuss the ethical considerations related to large language models and discuss potential mitigation strategies.

BLEURT: Learning Robust Metrics for Text Generation

Abstract: Text generation has made significant advances in the last few years. Yet, evaluation metrics have lagged behind, as the most popular choices (e.g., BLEU and ROUGE) may correlate poorly with human judgments. We propose BLEURT, a learned evaluation metric based on BERT that can model human judgments with a few thousand possibly biased training examples. A key aspect of our approach is a novel pre-training scheme that uses millions of synthetic examples to help the model generalize. BLEURT provides state-of-the-art results on the last three years of the WMT Metrics shared task and the WebNLG Competition dataset. In contrast to a vanilla BERT-based approach, it yields superior results even when the training data is scarce and out-of-distribution.