Machine translation, in recent times, has
become a buzzword in the translation industry for a lot of good reasons. Its
technology has enabled translation companies to offer faster turnaround times
for projects while at the same time ensuring substantial cost savings.

But in-between the media hype and
sensationalism lies a segment to MT often not talked about. There are significant
issues with machine translation that inhibits adoption in its entirety. Hence,
MT’s best use-cases often involve supplementing it with human input.

And it is not rocket science to know that by
identifying these issues, LSP’s can find better ways to navigate them to get
the best out of machine translation.

In this post, we would be going deeper into
some of these challenges associated with machine translation.

What are the issues
with machine translation?

Although MT has come a long way and witnessed recent improvements, especially with the
advent of Google’s neural machine translation (NMT) system, it has not really
done much to solve some of the underlying issues.

Perhaps the main concern with machine
translation lies in the inaccuracies of the translated text, particularly when
dealing with contextual meanings or translation of words that are technical or
culturally sensitive.

At the risk of sounding critical, there are
still many ways where machine translation stalls, but we would be looking at
them from these three main areas;

1. Machine translations are still prone to errors

MT has gravitated away from the word-to-word
system of translation with the emergence of neural networking models.

And yes, this new system may have proved to be
useful in decoding some context of the most relevant translation (an inherent
problem of word-based translation), but it still leads to errors when
translating sentences with idiomatic expressions, humor, or satire.

These errors can either be in the form of
missing or incorrect words, word order at both phrase or word levels.

Also, attempting to translate certain
terminologies oftentimes could lead to inconsistent or mistranslated output,
and the same could be said for uncommon (technical) terms.

1. Poor quality output

Building on our first point, it is only
natural that inherent errors during translation result in a below-par quality
of the final output. And this goes against the assertions of big tech companies
(Microsoft), who claim that their MT engine has attained human-level parity.

(There are a lot of controversies surrounding this)

But this claim is easily put to rest when the
translated output demands it be of a high standard, which is the requirement of
most translation projects in the real world.

Holding machine translation to a high standard
quickly reveals that its quality usually gravitates to the inferior side,
especially when translating unrelated language pairs.

1. Transcreation fail

Machine translation’s lack of cognitive
thinking makes performing creative translation impossible.

From knowing when to omit or translate puns to
even going as far as creating incorporating touches of humor into translations,
which makes the final output feel original. All these little decisions that a
qualified human translator makes on a whim while translating are usually
inconceivable for even the most sophisticated machine translation engines.

And the same extends to localization. If the demands of a
project require tailoring the content to suit a particular local culture, where
the content does not stay the same but only the meaning, only human hands can
achieve such feats. As no amount of training data would be able to perform the
task to satisfactory levels.

The way forward

Our above assertions are not in any way meant
to trivialize machine translation. Make no mistake, machine translation has
transformed workflows and improved the efficiency of many LSP’s and advances in
AI and machine learning can only lead to more improvements in its
applicability.

cMost importantly, we should note that machine translation’s best performance for the foreseeable future would always be paired to post-editing (MTPE) with an experienced human translator. But it should also be said that the quality from MT must be good enough otherwise it defeats the purpose of utilizing it in the first place.

Translationsinlondon is home to very capable and experienced
translators known to get the best out of machine translation in order to
deliver high-quality projects in no time.

Send us a message today to get a free quote for your project.