Finding Memo: Extractive Memorization in Constrained Sequence Generation Tasks
This addresses a critical issue for users of neural machine translation systems by identifying and mitigating memorization that can lead to unreliable outputs, though it is incremental as it builds on prior work on memorization in constrained NLG tasks.
The paper tackles the problem of extractive memorization in neural machine translation, where models generate exact training data under insufficient context, and demonstrates that this poses a serious threat to reliability, with a simple algorithm developed to elicit non-memorized translations for a large fraction of such samples.
Memorization presents a challenge for several constrained Natural Language Generation (NLG) tasks such as Neural Machine Translation (NMT), wherein the proclivity of neural models to memorize noisy and atypical samples reacts adversely with the noisy (web crawled) datasets. However, previous studies of memorization in constrained NLG tasks have only focused on counterfactual memorization, linking it to the problem of hallucinations. In this work, we propose a new, inexpensive algorithm for extractive memorization (exact training data generation under insufficient context) in constrained sequence generation tasks and use it to study extractive memorization and its effects in NMT. We demonstrate that extractive memorization poses a serious threat to NMT reliability by qualitatively and quantitatively characterizing the memorized samples as well as the model behavior in their vicinity. Based on empirical observations, we develop a simple algorithm which elicits non-memorized translations of memorized samples from the same model, for a large fraction of such samples. Finally, we show that the proposed algorithm could also be leveraged to mitigate memorization in the model through finetuning. We have released the code to reproduce our results at https://github.com/vyraun/Finding-Memo.