Polyjuice: Generating counterfactuals for explaining, evaluating, and improving models
While counterfactual examples are useful for analysis and training of NLP models, current
generation methods either rely on manual labor to create very few counterfactuals, or only …
generation methods either rely on manual labor to create very few counterfactuals, or only …
Agency plus automation: Designing artificial intelligence into interactive systems
J Heer - Proceedings of the National Academy of Sciences, 2019 - National Acad Sciences
Much contemporary rhetoric regards the prospects and pitfalls of using artificial intelligence
techniques to automate an increasing range of tasks, especially those once considered the …
techniques to automate an increasing range of tasks, especially those once considered the …
Sparse sequence-to-sequence models
Sequence-to-sequence models are a powerful workhorse of NLP. Most variants employ a
softmax transformation in both their attention mechanism and output layer, leading to dense …
softmax transformation in both their attention mechanism and output layer, leading to dense …
Assessing the impact of automated suggestions on decision making: Domain experts mediate model errors but take less initiative
A Levy, M Agrawal, A Satyanarayan… - Proceedings of the 2021 …, 2021 - dl.acm.org
Automated decision support can accelerate tedious tasks as users can focus their attention
where it is needed most. However, a key concern is whether users overly trust or cede …
where it is needed most. However, a key concern is whether users overly trust or cede …
Under pressure: translation in times of austerity
J Moorkens - Perspectives, 2017 - Taylor & Francis
The profession of translation is undergoing enormous change, expedited by the global
recession that began in 2007–2008. Government policies and an intensive focus on cost …
recession that began in 2007–2008. Government policies and an intensive focus on cost …
Mapping the design space of human-ai interaction in text summarization
Automatic text summarization systems commonly involve humans for preparing data or
evaluating model performance, yet, there lacks a systematic understanding of humans' roles …
evaluating model performance, yet, there lacks a systematic understanding of humans' roles …
Interactive neural machine translation
Despite the promising results achieved in last years by statistical machine translation, and
more precisely, by the neural machine translation systems, this technology is still not error …
more precisely, by the neural machine translation systems, this technology is still not error …
Neural interactive translation prediction
We present an interactive translation prediction method based on neural machine
translation. Even with the same translation quality of the underlying machine translation …
translation. Even with the same translation quality of the underlying machine translation …
AutoMeTS: the autocomplete for medical text simplification
The goal of text simplification (TS) is to transform difficult text into a version that is easier to
understand and more broadly accessible to a wide variety of readers. In some domains …
understand and more broadly accessible to a wide variety of readers. In some domains …
Reliability and learnability of human bandit feedback for sequence-to-sequence reinforcement learning
We present a study on reinforcement learning (RL) from human bandit feedback for
sequence-to-sequence learning, exemplified by the task of bandit neural machine …
sequence-to-sequence learning, exemplified by the task of bandit neural machine …