Adapting Text Embeddings for Causal Inference

By Victor Veitch, Dhanya Sridhar, David M. Blei. 2019

129 citations recorded by the source corpus.

Subfields: Natural Language Processing, Causal Inference & Discovery

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Natural Language Processing

1. Brendan T. O'Connor
2. Yulia Tsvetkov
3. Katherine A. Keith
4. David D. Jensen
5. Jacob Eisenstein
6. Anjalie Field
7. Diyi Yang
8. Nadav Oved

In Causal Inference & Discovery

1. Katherine A. Keith
2. David D. Jensen
3. Brendan T. O'Connor
4. Victor Veitch
5. Amir Feder
6. Roi Reichart
7. Nadav Oved
8. Zach Wood-Doughty

Citations in the corpus

• CausaLM: Causal Model Explanation Through Counterfactual Language Models (2020, background): “Recently, this seems to be changing, with substantial work being done on the intersection of causal inference and NLP (Tan, Lee, and Pang 2014; Fong and Grimmer 2016; Egami et al. 2018; Wood-Doughty, Shpitser, and Dredze 2018; Veitch, Sridhar, and Blei 2019).”
• Deconfounded Image Captioning: A Causal Retrospect (2020, background): “For example, disen-tangled causal mechanisms are exploited to design more robust disentanglement measurements [58], generate coun-terfactual samples [59], [60], and alleviate negative effects brought by different biases [61], [62], [63], [64], [65].”
• Text and Causal Inference: A Review of Using Text to Remove Confounding from Causal Estimates (2020, methodology): “To construct Table 1, we started with three seed papers: Roberts et al. (2020), Veitch et al. (2019), and Wood-Doughty et al. (2018).”
• Causal Inference in Natural Language Processing: Estimation, Prediction, Interpretation and Beyond (2021, methodology): “Recently, Veitch et al. (2020) adapted pre-trained language models and supervised topic models with multiple classification heads for binary treatment and coun-terfactual outcomes.”
• Unsupervised Discovery of Implicit Gender Bias (2020, background): “In the first scenario, text maybe be a confounding variable that needs to be controlled in order to measure the effect of a non-text variable (Roberts et al., Forthcoming; Veitch et al., 2019).”
• Open-Source Embedding Models: A Comprehensive Survey of Techniques, Benchmarks, and Applications (2026, unknown): “Even minor alterations in the input text, such as typos, grammatical errors, or changes in phrasing, can sometimes result in significantly different embeddings [88].”