Counterfactual Learning from Bandit Feedback under Deterministic Logging: A Case Study in Statistical Machine Translation

TL;DR

This paper demonstrates that counterfactual learning can be effectively applied to deterministic bandit logs in statistical machine translation by using smoothing techniques, leading to significant translation quality improvements.

Contribution

It introduces methods to enable counterfactual learning from deterministic logs in SMT, overcoming theoretical challenges with smoothing techniques.

Findings

Up to 2 BLEU points improvement in translation quality

Counterfactual learning is feasible with deterministic logs using smoothing

Proposed methods prevent degenerate behavior in risk minimization

Abstract

The goal of counterfactual learning for statistical machine translation (SMT) is to optimize a target SMT system from logged data that consist of user feedback to translations that were predicted by another, historic SMT system. A challenge arises by the fact that risk-averse commercial SMT systems deterministically log the most probable translation. The lack of sufficient exploration of the SMT output space seemingly contradicts the theoretical requirements for counterfactual learning. We show that counterfactual learning from deterministic bandit logs is possible nevertheless by smoothing out deterministic components in learning. This can be achieved by additive and multiplicative control variates that avoid degenerate behavior in empirical risk minimization. Our simulation experiments show improvements of up to 2 BLEU points by counterfactual learning from deterministic bandit feedback.