Segment-based Interactive Machine Translation at a Character Level

Segment-based Interactive Machine Translation at a Character Level Ángel Navarro1 and Miguel Domingo1,2 and Francisco Casacuberta1,2 1PRHLT Research Center Universitat Politècnica de València, Spain {annamar8, midobal, fcn}@prhlt.upv.es 2ValgrAI - Valencian Graduate School and Research Network for Artificial Intelligence, Camí de Vera s/n, 46022 Valencia, Spain Abstract To produce high quality translations, hu- man translators need to review and correct machine translation hypothesis in a process known as post-editing. In order to reduce the human effort of this task, interactive machine translation proposed a collabora- tive framework in which human and ma- chine work together to generate the transla- tions. Among the many protocols proposed throughout the years, the segment-based one established a paradigm in which the post-editor is allowed to validate correct word sequences from a translation hypoth- esis and to introduce a word correction to help the system improve the next hypothe- sis. In this work we propose an extension to this protocol: instead of having to type the complete word correction, the system will complete the user’s correction while they are typing. We evaluated our proposal under a simulated environment, achieving a significant reduction of the human effort. 1 Introduction The machine translation (MT) field has significantly changed over the last few years due to the appear- ance and application of neural models. Thanks to this emergent technology, researchers have been able to accomplish human parity in several MT- related tasks (Toral, 2020). Thus, in the future we might no longer need human translators to review and correct translations hypothesis from an MT system to achieve high-quality translations. Until © 2023 The authors. This article is licensed under a Creative Commons 4.0 licence, no derivative works, attribution, CC- BY-ND. this future arrives, human experts need to be in- volved in the translation process and post-edit the MT system’s output in order to get translations of the required high quality. To alleviate the cost of the post-editing task, in- teractive machine translation (IMT) proposed a col- laborative framework in which human and machine work together to construct the final translation: in- stead of correcting the complete translation hypoth- esis, the expert can provide the system with some feedback which it uses to generate a new hypothesis. This process is repeated until the user is satisfied with the system’s hypothesis. Among the different protocols proposed in the literature, we find segment-based IMT (Domingo et al., 2017; Peris et al., 2017). In this paradigm, the user reviews the system’s translation hypothesis and can validate sequences of words which they consider to be correct. Then, they make a word correction. The system reacts to this feedback by generating a new hypothesis and, thus, starting a new iteration of the process. Figure 1 illustrates an iteration of a segment- based IMT session where the user has to translate a sentence from Spanish to English. Given the hypothesis generated by the MT model, the user starts validating a sequence of correctly translated segments and types the word first to help the system fulfill the …

Content truncated to 500 words