Eduard H. Hovy

ICLR Conference 2021 Conference Paper

Decoupling Global and Local Representations via Invertible Generative Flows

• Xuezhe Ma
• Xiang Kong
• Shanghang Zhang
• Eduard H. Hovy

In this work, we propose a new generative model that is capable of automatically decoupling global and local representations of images in an entirely unsupervised setting, by embedding a generative flow in the VAE framework to model the decoder. Specifically, the proposed model utilizes the variational auto-encoding framework to learn a (low-dimensional) vector of latent variables to capture the global information of an image, which is fed as a conditional input to a flow-based invertible decoder with architecture borrowed from style transfer literature. Experimental results on standard image benchmarks demonstrate the effectiveness of our model in terms of density estimation, image generation and unsupervised representation learning. Importantly, this work demonstrates that with only architectural inductive biases, a generative model with a likelihood-based objective is capable of learning decoupled representations, requiring no explicit supervision. The code for our model is available at \url{https://github.com/XuezheMax/wolf}.

ICLR Conference 2021 Conference Paper

Explaining the Efficacy of Counterfactually Augmented Data

• Divyansh Kaushik
• Amrith Setlur
• Eduard H. Hovy
• Zachary C. Lipton

In attempts to produce machine learning models less reliant on spurious patterns in NLP datasets, researchers have recently proposed curating counterfactually augmented data (CAD) via a human-in-the-loop process in which given some documents and their (initial) labels, humans must revise the text to make a counterfactual label applicable. Importantly, edits that are not necessary to flip the applicable label are prohibited. Models trained on the augmented (original and revised) data appear, empirically, to rely less on semantically irrelevant words and to generalize better out of domain. While this work draws loosely on causal thinking, the underlying causal model (even at an abstract level) and the principles underlying the observed out-of-domain improvements remain unclear. In this paper, we introduce a toy analog based on linear Gaussian models, observing interesting relationships between causal models, measurement noise, out-of-domain generalization, and reliance on spurious signals. Our analysis provides some insights that help to explain the efficacy of CAD. Moreover, we develop the hypothesis that while adding noise to causal features should degrade both in-domain and out-of-domain performance, adding noise to non-causal features should lead to relative improvements in out-of-domain performance. This idea inspires a speculative test for determining whether a feature attribution technique has identified the causal spans. If adding noise (e.g., by random word flips) to the highlighted spans degrades both in-domain and out-of-domain performance on a battery of challenge datasets, but adding noise to the complement gives improvements out-of-domain, this suggests we have identified causal spans. Thus, we present a large scale empirical study comparing spans edited to create CAD to those selected by attention and saliency maps. Across numerous challenge domains and models, we find that the hypothesized phenomenon is pronounced for CAD.

ICLR Conference 2020 Conference Paper

Learning The Difference That Makes A Difference With Counterfactually-Augmented Data

• Divyansh Kaushik
• Eduard H. Hovy
• Zachary C. Lipton

Despite alarm over the reliance of machine learning systems on so-called spurious patterns, the term lacks coherent meaning in standard statistical frameworks. However, the language of causality offers clarity: spurious associations are due to confounding (e.g., a common cause), but not direct or indirect causal effects. In this paper, we focus on natural language processing, introducing methods and resources for training models less sensitive to spurious patterns. Given documents and their initial labels, we task humans with revising each document so that it (i) accords with a counterfactual target label; (ii) retains internal coherence; and (iii) avoids unnecessary changes. Interestingly, on sentiment analysis and natural language inference tasks, classifiers trained on original data fail on their counterfactually-revised counterparts and vice versa. Classifiers trained on combined datasets perform remarkably well, just shy of those specialized to either domain. While classifiers trained on either original or manipulated data alone are sensitive to spurious features (e.g., mentions of genre), models trained on the combined data are less sensitive to this signal. Both datasets are publicly available.

AIJ Journal 1993 Journal Article

Automated discourse generation using discourse structure relations

• Eduard H. Hovy

AAAI Conference 1991 Conference Paper

Automatic Generation of Formatted Text

• Eduard H. Hovy

AIJ Journal 1990 Journal Article

Pragmatics and natural language generation

• Eduard H. Hovy

AAAI Conference 1987 Conference Paper

Interpretation in Generation

• Eduard H. Hovy

The computer maxim garbage in, garbage out is especially true of language generation. When a generator slavishly follows its input topics, it usually produces bad text. In order to find more appropriate forms of expression, generators must be given the ability to interpret their input topics. Often, newly formed interpretations can help generators achieve their pragmatic goals with respect to the hearer. Since interpretation requires inference, generators must exercise some control over the inference process. Some general strategies of control, and some specific techniques geared toward achieving pragmatic goals, are described here.

IJCAI Conference 1985 Conference Paper

Integrating Text Planning and Production in Generation

• Eduard H. Hovy

While the task of language generation seems to separate quite naturally into the two aspects of language generation (text planning and text production), it is necessary to have the planning and the production interact at generator decision points in such a way that the former need not contain explicit syntactic knowledge, and that the latter need not contain explicit goal-related information. This paper describes the decision points, the types of plans that are used in making the decisions, and a process that performs the task. These ideas are embodied in a program.