Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Machine translation
Poibeau T., The MIT Press, Cambridge, MA, 2017. 296 pp.  Type: Book (978-0-262534-21-5)
Date Reviewed: Aug 20 2018

This small book (approximately size A5), part of the Massachusetts Institute of Technology (MIT) Press “Essential Knowledge” series, captures the essence of the machine translation (MT) field, from its origins to the present. As opposed to review books or textbooks, its focus is a panoramic view of MT, at the cost of in-depth details. The MT field has been active for more than seven decades, and it still faces tough challenges in accuracy and coverage. The book marks important milestones in this journey, including those that occurred before the arrival of computers.

The book spans 15 chapters, starting with an introduction that sets the stage. Chapter 2 examines some foundational issues, for example, characteristics of a good translation and ambiguity. Chapter 3 provides an overview of broad approaches, including rule-based systems, the Vauquois triangle, and statistical MT. The next couple of chapters look at some major pre-computer developments, many of which were a driving force in MT for many years to come. For example, the possibility of a universal language that can serve as translation’s interlingua was explored until recently. The exchanges between Norbert Wiener and Warren Weaver are quite interesting (and rarely covered in MT books).

Chapter 6 is on the biggest negative milestone in the history of MT: the Automatic Language Processing Advisory Committee (ALPAC) report, its impact on MT work around the world, and the consequent MT winter. MT came back with a bang, however, with the arrival of statistical approaches, as they promised at least a partial solution to the expensive knowledge engineering challenge of traditional methods. Chapter 7 discusses the key driver, a parallel corpus, and related issues. Issues of sentence alignment, which led to the notion of example-based MT, are discussed in some detail. Chapter 8 discusses example-based MT, and this discussion continues in chapter 9 with word alignment and statistical MT (SMT). Chapter 10 moves on to the segment-based approach. Since SMT is the most widely accepted approach today, it receives a fairly detailed treatment. Chapter 11 provides a focused review of SMT, outlining its limitations and challenges. SMT has been slowly giving way to deep learning of late, and this is the focus of chapter 12. Given the hype surrounding deep learning, however, this chapter is somewhat lacking. Chapter 13 discusses evaluation models, a topic often neglected in MT. Chapter 14 looks at the MT industry, examining actual use cases in practice. Chapter 15 concludes the book and examines issues like cognitively sound approaches.

The book is a pleasurable read. The language is smooth and coherent. Readers will learn many interesting tidbits, even if they are already familiar with the field of MT. The book will also help readers make connections between the different developments on the MT timeline, including how these developments have influenced the different strands of MT. I would have liked to see a chapter on challenges to MT (for example, the different language phenomena). However, on the whole, I strongly recommend this book for anyone interested in languages, and MT specifically.

More reviews about this item: Amazon, Goodreads

Reviewer:  M Sasikumar Review #: CR146211 (1811-0559)
Bookmark and Share
  Editor Recommended
Machine Translation (I.2.7 ... )
Applications And Expert Systems (I.2.1 )
Would you recommend this review?
Other reviews under "Machine Translation": Date
Making machine learning robust against adversarial inputs
Goodfellow I., McDaniel P., Papernot N.  Communications of the ACM 61(7): 56-66, 2018. Type: Article
Oct 15 2018
Machine translation
Poibeau T.,  The MIT Press, Cambridge, MA, 2017. 296 pp. Type: Book (978-0-262534-21-5), Reviews: (3 of 3)
Sep 28 2018
Learning safe multi-label prediction for weakly labeled data
Wei T., Guo L., Li Y., Gao W.  Machine Learning 107(4): 703-725, 2018. Type: Article
Sep 11 2018

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright © 2000-2019 ThinkLoud, Inc.
Terms of Use
| Privacy Policy