Confidence through Attention
Citations Over Time
Abstract
Attention distributions of the generated translations are a useful bi-product of attention-based recurrent neural network translation models and can be treated as soft alignments between the input and output tokens. In this work, we use attention distributions as a confidence metric for output translations. We present two strategies of using the attention distributions: filtering out bad translations from a large back-translated corpus, and selecting the best translation in a hybrid setup of two different translation systems. While manual evaluation indicated only a weak correlation between our confidence score and human judgments, the use-cases showed improvements of up to 2.22 BLEU points for filtering and 0.99 points for hybrid translation, tested on EnglishGerman and EnglishLatvian translation.
Related Papers
- → Personality correlates and utilitarian judgments in the everyday context: Psychopathic traits and differential effects of empathy, social dominance orientation, and dehumanization beliefs(2019)19 cited
- → Amount of altruistic punishment accounts for subsequent emotional gratification in participants with primary psychopathy(2011)14 cited
- Susquehanna Chorale Spring Concert "Roots and Wings"(2017)
- Mediating Role of Unconditional Self-acceptance in Relationship Between College Students’ Perfectionism and Depression(2011)
- → ИСПОЛЬЗОВAНИЕ ПОТЕНЦИAЛA СОЦИAЛЬНЫХ ПAРТНЕРОВ В ПОДГОТОВКЕ БУДУЩИХ ПЕДAГОГОВ(2024)