Linguistic competence is a specific skill that some exceptionally gifted people possess when learning a foreign language by listening to a native speaker or reading a text written by a native speaker. The comprehension of an utterance comes from a semantics built on the extraction of known words and enriched by the person’s automatically transferred knowledge of the particular language pair phenomena. Instead of learning the foreign language’s vocabulary, grammar, words, connotations, pronunciation, or melody on the whole, the learner picks up an accent and some basic vocabulary necessary for everyday communication and keeps on expanding it gradually, having passive contact (listening or reading only) with the language supported by the knowledge transfer. We usually admire these people who speak so many foreign languages fluently and can switch from one language to another with such astonishing ease.
Machine translation (MT) tries to apply such a schema--that is, transferring knowledge about the source language--to the target language, both for text and speech. Its long research history has focused on making an equivalent of any source text perfect in its target text. Unfortunately, despite the dramatic progress achieved by technologies such as neural network translation and deep learning, automatic translation is still far from the perfection users expect; in particular, it continues to be dependent upon the language pair.
The book’s back cover provides a brief but compete summary of what readers can expect:
A concise, non-technical overview of the development of automatic machine translation, including the different approaches, evaluation issues, and major players in the industry.
My PhD dissertation was in the field of cross-language information retrieval, also called translingual engineering, which relies on MT enormously, so I was wondering if there is anything else unknown to me that I would find in this book. Surely, it addresses the greatest challenges of translation such as diversity of languages, word ambiguity, different understandings of the same sentence by users (or alternatively, more than one target sentence being a perfect naturally sounding translation of the same source sentence), and finally subjectivity of the evaluation measures, all of which I read with interest and trust that nonspecialists will read with an even greater interest. It also sees the connection between cognitive sciences and artificial intelligence that employs this linguistic reasoning in practice.
In 15 chapters, Poibeau provides readers with a history of MT: the first approaches before the 1940s; the 1940s through the 1960s with the advent of computers; and the productive period of the 1980s for Anglo-American countries, initiated by direct and then rule-based translation models to develop language-independent artificial interlingua built on the Latin schema. In parallel corpora we found a new technology, as huge text collections were compared for translation equivalents. The 1990s was a very famous period for statistical IBM bilingual models (French to English, or vice versa). More recently, in 2016, Google introduced neural network MT and initiated the exploration of deep learning based on indexed user utterances that resulted in Google Translate.
All of the approaches are presented in a very simple way and the translation models are supported with representative examples. My impression is that the book covers everything there is to learn about MT. Therefore, I would strongly recommend it to students and MT system users in particular.
More reviews about this item: Amazon, Goodreads