Word embeddings quantify 100 years of gender and ethnic stereotypes

N Garg, L Schiebinger, D Jurafsky… - Proceedings of the …, 2018 - National Acad Sciences
Word embeddings are a powerful machine-learning framework that represents each English
word by a vector. The geometric relationship between these vectors captures meaningful …

Dynamic word embeddings

R Bamler, S Mandt - International conference on Machine …, 2017 - proceedings.mlr.press
We present a probabilistic language model for time-stamped text data which tracks the
semantic evolution of individual words over time. The model represents words and contexts …

On the origins of linear representations in large language models

Y Jiang, G Rajendran, P Ravikumar, B Aragam… - arxiv preprint arxiv …, 2024 - arxiv.org
Recent works have argued that high-level semantic concepts are encoded" linearly" in the
representation space of large language models. In this work, we study the origins of such …

Interpretable word embeddings via informative priors

MH Bodell, M Arvidsson, M Magnusson - arxiv preprint arxiv:1909.01459, 2019 - arxiv.org
Word embeddings have demonstrated strong performance on NLP tasks. However, lack of
interpretability and the unsupervised nature of word embeddings have limited their use …

Case vectors: Spatial representations of the law using document embeddings

E Ash, DL Chen - Law as data, 2019 - hal.science
Recent work in natural language processing represents language objects (words and
documents) as dense vectors that encode the relations between those objects. This paper …

Learning dynamic author representations with temporal language models

E Delasalles, S Lamprier… - 2019 IEEE International …, 2019 - ieeexplore.ieee.org
Language models are at the heart of numerous works, notably in the text mining and
information retrieval communities. These statistical models aim at extracting word …

Dynamic word embeddings

SM Mandt, R Bamler - US Patent 11,068,658, 2021 - Google Patents
Abstract Systems, methods, and articles of manufacture to perform an operation comprising
deriving, based on a corpus of electronic text, a machine learning data model that …

Improving optimization for models with continuous symmetry breaking

R Bamler, S Mandt - International Conference on Machine …, 2018 - proceedings.mlr.press
Many loss functions in representation learning are invariant under a continuous symmetry
transformation. For example, the loss function of word embeddings (Mikolov et al., 2013) …

A dynamic embedding model of the media landscape

J Rappaz, D Bourgeois, K Aberer - The World Wide Web Conference, 2019 - dl.acm.org
Information about world events is disseminated through a wide variety of news channels,
each with specific considerations in the choice of their reporting. Although the multiplicity of …

The Geometry of Categorical and Hierarchical Concepts in Large Language Models

K Park, YJ Choe, Y Jiang, V Veitch - arxiv preprint arxiv:2406.01506, 2024 - arxiv.org
Understanding how semantic meaning is encoded in the representation spaces of large
language models is a fundamental problem in interpretability. In this paper, we study the two …