An active learning scenario for interactive machine translation

Abstract

Translation needs have greatly increased during the last years. In many situations, text to be translated constitutes an unbounded stream of data that grows continually with time. An effective approach to translate text documents is to follow an interactive-predictive paradigm in which both the system is guided by the user and the user is assisted by the system to generate error-free translations. Unfortunately, when processing such unbounded data streams even this approach requires an overwhelming amount of manpower. Is in this scenario where the use of active learning techniques is compelling. In this work, we propose different active learning techniques for interactive machine translation. Results show that for a given translation quality the use of active learning allows us to greatly reduce the human effort required to translate the sentences in the stream.

Integrating Online and Active Learning in a Computer-Assisted Translation Workbench

“…This fact constraints the models and techniques that can be used to implement AL. Particularly, we select which sentences should be post-edited by the user according to a sentence-level quality measure based on statistical lexicons (González-Rubio et al, 2012) and, given a new translation example, the parameters of the SMT model are reestimated via the OL techniques described above.…”

“…Specifically, the SMT models are updated in real time from the target translations validated by the user, preventing the system from repeating errors in the translation of similar sentences. Despite the strong potential of these features to improve the user experience (Ortiz-Martínez et al, 2010;González-Rubio et al, 2012;Bertoldi et al, 2013;Denkowski et al, 2014), they are still not widely implemented in CAT systems. To the best of our knowledge, the only exception is (Ortiz-Martínez et al, 2011) where the authors describe the implementation of online learning within an ITP system.…”

Learning Tree Languages

“…A short but nice discussion of using tree versus string language learners for NLP purposes can be found in [155]. Recently, also ideas from active learning found their way into SMT systems; see [92]. Let us finally mention that another topic relevant for NLP, more precisely for the assistance of corpus construction, is that of tree annotation; see [170] for one approach on statistical learning in this context.…”

The New Thot Toolkit for Fully-Automatic and Interactive Statistical Machine Translation

“…Empirical results obtained with THOT and reported in (Ortiz-Martínez et al, 2010;OrtizMartínez, 2011) show that incremental learning allows to significantly reduce the user effort in IMT tasks with respect to that required by a conventional IMT system. Additionally, the incremental learning techniques provided by THOT are currently being used in other sophisticated applications such as active learning for SMT (González-Rubio et al, 2012).…”