Machine Translation and Post Editing: Everything You Need to Know

1
Machine Translation and Post Editing

• Everything You Need to Know

2
Machine Translation or MT Definition

• The automatic translation from one natural
  language to another natural language by a
  computer software based on certain rules and
  statistics. There is no human intervention in the
  translation process.

3
Machine Translation Post Editing or MTPE
Definition

• is the process where a human linguists reviews
  and edits the translation done by a machine. The
  human linguists checks and corrects grammar,
  sentence structure, context and tone.

4
A History of Machine Translation

• In 1949 could be deemed as the set-off point for
  MT. In this year Warren Weaver, a researcher at
  the Rockefeller Foundation, presented his
  proposal for machine-based translations, inspired
  by information theory and code-breaking
  techniques during WWII.

5

• Initially, MT served military interests only.
  Later on, however, MT began to be used to cater
  to civilian needs as well.
• Additionally, in the age of globalization, fast
  and efficient translation became a necessity for
  any business.
• As a result, at the turn of the century, many
  companies released online translation services.
  One of them was Babel Fish.
• Their service worked with a limited range of
  languages and its rule-based system lacked
  extra-textual knowledge. This meant it got
  confused by expressions such as see and sea.

6

• This type of MT works with general engines or in
  other words they are used to translate general
  content (hence the name) without specialized
  technology.
• This is also the reason why they fail to
  recognize colloquial expression or artistic
  language. Another influential idea is the
  statistical MT which was initially used for
  Google Translate.
• This type of MT is, so to say, self-learning,
  meaning it continually increases its corpus.
  Whats more, it works on the principle of two
  parallel texts.
• Statistical or vertical MT engines can be
  trained to use specific terminology for a
  specific domain (e.g. travel, manufacturing,
  etc.).

7

• The newest approach is neural MT, which however
  is not a drastic step from SMT. This approach
  utilizes an artificial neural network to predict
  the likelihood of a sequence of words.
• These MT engines will never be as meticulous as
  human translation, although they are constantly
  evolving.
• This resulted in the emergence of MTPE or in
  other words, computers and human translators
  working together to achieve fast quality results.
• In this presentation, we will present you with a
  detailed MT and MTPE guide to making sure you are
  well-acquainted with the latest trends and most
  importantly, know how to incorporate them to fit
  your needs!

8
1. Machine translation vs Machine Translation
Post-editing

• Machine Translation refers to the automated
  process of translation from source material into
  another language. In recent years MT engines have
  improved immensely due to the advent of new
  software and hardware.
• Nowadays these engines can process millions of
  words per hour while still retaining the high
  quality of the translation.
• Moreover, MT engines can be used for nearly every
  language. However, MT has a hard time recognizing
  idioms or more colloquial expressions. Have a
  look at Google Translate

9

• This first example is a wrong translation from
  Bulgarian to English. Google translate fails to
  recognize a rather popular colloquial expression
  in the Bulgarian language. I feel so sleepy that
  I cant see two should be translated simply to
  Im very tired.

10

• This is an excerpt from the book Sie nannten
  sich der Wiener Kreis (They Called Themselves
  the Vienna Circle) by Karl Sigmund. Have a look
  at the phrase den Ungeraden which is translated
  as the odd ones.
• Although that translation is not completely
  wrong, in this context it is not a good choice.
  The German phrase is used to denote the
  politically undesirable people, so the odd
  ones a little too vague.

11
Machine Translation Post Editing

• MTPE is a hybrid between machine translation and
  human translation. This in turn means a twofold
  benefit for the client high speed and high
  quality.
• As could be inferred from its name, the MT is
  followed by a post-editing process which is
  fulfilled by a human translator. Their job is to
  edit the material for better accuracy, flow,
  clarity, and local resonance.
• It might be hard to decide which one would better
  suit your needs MT or MTPE. If this is the
  case, have a look at the pros and cons of each

12

• Mistranslations Note MT is prone to making a
  wide range of mistakes such as
• Wrong word order, gender, or number
• missing/Wrong Articles
• Wrong punctuation
• Words in target text left in the source language
• Once we have cleared out what are the positive
  and negative sides of MT and MTPE, you might have
  already made up your mind about which one you
  want to use. If this is not the case, however, or
  you are still unsure, we have your back!

13

• When you can use MT
• If you need to translate the content as fast as
  possible
• Limited budget
• If a rough translation will be sufficient (e.g.
  low-priority internal messages)
• If you have large volumes of text that need to be
  translated.
• When you can use MTPE
• If you need a culturally sensitive or very
  precise translation
• And if you are translating a text that requires
  technical knowledge (e.g. medicine, mechanics,
  law, etc.)
• If you are working with public-facing materials
  (e.g. ads or product descriptions).

14
2. Post-editing strategies for MTPE

• The post-editing process depends on your
  expectations which should be specified in
  advance. The three main points to keep in mind
  are time, cost, and quality since they would
  determine your post-editing strategy.
• Light Post-editing (LPE)
• as few edits as possible
• to ensure legible and accurate output
• fast and cost-effective PE approach
• Full Post-editing (FPE)
• a thorough review of the MT
• focuses on consistency of tone, correct grammar,
  and appropriateness of the expressions
• aims to sound natural, native
• slower and more expensive approach than LPE
• high-quality translation

15

• Project-specific Strategy
• requirements are project-dependent
• certain segments can be prioritized if they have
  higher business value
• can be adapted to your expectations for time,
  costs, quality
• Bypassing Post Editing
• could work very well with some materials (e.g.
  internal documentation)
• MT quality is expected to be good
• Low probability for bad translations (e.g. very
  simple sentences or a database of terms)

16
3. Does MT work for all languages and dialects?

• Without a doubt, MT has come a long way since its
  advent in the late 40s. Thanks to the progress in
  the development of processing technology, it is
  constantly getting better. However, MT tends to
  work much better with some language combinations
  than with others.
• Generally, MT performs with widely spoken
  languages very well (e.g. English, German,
  French). Moreover, it is best to use MT between
  English and European languages since they are
  related.
• On the contrary, translations between languages
  like English and Japanese, which are disparate,
  generally do not work as good.

17

• The same goes for dialects. There is no machine
  to translate specifically into Canadian French or
  Colombian Spanish. As a rule of thumb, European
  languages have similar grammatical structures
  since they are more closely related.
• In both cases, however, mistakes still occur
  especially when dealing with languages that have
  more complex grammar and syntax (e.g. Turkish and
  Russian).
• MTPE, on the other hand, can deal with such
  ambiguities, of course, with the help of a human
  translator during the post-editing process.

18
4. Is MTPE suitable for all types of content?

• Generally, MT works best with clear and concise
  content. Ideally, a sentence should have the
  simple grammar and not more than 20 words.
• Headline-style sentence segments and long,
  complex sentences do not work well with MT.
• Also, for MT to provide high-quality
  translations, it is best to work with literal
  texts since it tends to struggle to recognize
  tones such as sarcasm or humor.
• Moreover, double negations (e.g. it wont do you
  any good) or some very colloquial expressions are
  real obstacles for MT.
• If you want to ensure the best results, consider
  revising your text so that is prepared
  specifically for MT.
• When it comes to MTPE, good candidates would be
  instruction manuals, website contents, blog
  posts, or packaging inserts.

19
When to use human translations only

• It is not advisable to use raw MT for materials
  aimed at your audience or such that need to
  reflect your brands message. Creative segments
  such as headlines or slogans should be translated
  by human translators without any machine
  intervention.
• Also, when translating highly technical texts or
  materials containing sensitive issues, jokes, or
  a certain tone, it is preferable to opt for human
  translation.
• In such cases, completely eliminating machines is
  advisable. This is because making sense of and
  correcting machine translation can be more
  time-consuming than doing a human translation
  from scratch.

20
5. How much you can save from the machine
translation?

• To put it shortly, MTPE efficiency is very
  difficult to predict. Without a doubt, MT speeds
  up the translation process.
• Moreover, according to a Stanford University
  study regardless of language pairs or
  translators professional experience, MTPE is
  considerably faster than translating from
  scratch.
• Indeed, most professional translators work more
  quickly as editors than as from-scratch
  translators.
• The initial MT translation can save translators
  up to a few hours of work. This makes MTPE one of
  the quickest ways to get a high-quality
  translation.

21

• As a result, human translation costs a lot more
  than MT since it simply takes longer.
• For common languages like German, French,
  Spanish, and Dutch, MT quality is usually upwards
  of 65.
• When our clients use MTPE, the can save up to 30
  on translation costs.
• By using MTPE for your project, you can take the
  best of both worlds MT and human translation.
• Therefore, MTPE would mean major cost savings
  without having to compromise quality.

22
Conclusion

• Well, MT has come a long way, however, it is
  still not able to do sense to sense
  translation. In fact, its purpose has never been
  to replace human translators, but to make their
  job easier and faster.
• MTPE, on the other hand, is a new approach to
  translation that incorporates computers and human
  translators to grant the client premium quality
  translations at lower rates!
• To get MTPE right, you need the help of
  professionals. From choosing the right MT engine
  for your language pair and subject matter to
  training linguists to post edit, expertise, and
  experience is required.
• Contact us for professional MTPE services.