Translating Machine Translation
Machine translation (MT) is defined as using software to automatically translate text from one language to another and has been approached in different ways (Qun & Xiaojun, 2015, p. 105). The oldest approaches, rule-based MT (e.g., Babelfish), required language rules (grammatical, syntactic, etc) to be manually programmed into the software (Jiménez-Crespo, 2017; Qun & Xiaojun, 2015). Statistical machine translation, introduced in the late 1980s, relies on probabilistic statistical models that use algorithms to draw out correspondences between parallel texts (Le & Schuster, 2016; Qun & Xiaojun, 2015; Wu et al., 2016).
Deep learning machine translation, the latest in MT approaches, uses advances in machine learning to draw out patterns in raw data sets: rather than relying on pre-coded input or pre-written rules, deep learning MT software constructs (or learns) rules from the linguistic input itself (Lewis-Kraus, 2016; Poibeau, 2017). The updated product is faster, requires less human programming on the front-end, and can better handle longer texts and rare words (Kelleher, 2019; Lewis-Kraus, 2016; Poibeau, 2017; Wu et al., 2016). For example, Google launched a new version of its Google Translate platform that uses a form of deep learning (neural networks) in 2016 (Le & Schuster, 2016). This new approach produces noticeably better translations in languages with sufficient databases, a change that has been documented by the MT industry (Lewis-Kraus, 2016; Wu et al., 2016) as well as language teaching/learning professionals (Briggs, 2018; Ducar & Schocket, 2018; Stapleton & Kin, 2019).