Long translations pose special problems for
translators. The intensive process involved in translating a
long legal document or multipage text leaves little time and freshness for the
no-less-crucial process of QA. Simply put, by the time the hard-pressed
translator gets to the editing stage, the text is already embedded in the mind as basically acceptable, limiting the translator’s ability to
identify syntactic errors. This problem is not new, with translators having
long used a set of tools, some old and some new, to overcome it to varying
effectiveness. I will discuss some traditional QA methods, some recently developed ones
and my experience with Grammarly. Whatever the individual choice of
tools, professional translators and writers must employ them to properly
check their work.
To illuminate the problem, the act of translating a
long text, whether a legal document, a set of technical instructions or
personal journal, involves multiple hours and multiple readings. Working from first
draft to non-QAed finalized version may require the linguist to read through
the original and source four or more times. Aside from the time invested, the
creation process creates a sound and vision in the mind, which become de
facto acceptable. Professionals are aware that many areas for correction
and improvement lurk in the text and strive to find them in the jungle of text.
However, the longer the text, the more difficult it is to locate them. The
moment of truth is when the customer or translation agency editor sends back a
document riddled with red marks, a truly unpleasant and often embarrassing
experience. The question that most translators and editors ask is “How did I
miss that?”. It often was not from lack of effort but due to the tools they
used.
The tools of translators and other linguists at
minimum include
Word “spellcheck” (F7), printing and reading and use of
outside editors. The Spellcheck function in Word identifies the most basic of
errors with the failure to use it bordering on professional incompetence. A
more comprehensive manner of editing is to print and read a text. For some
reason, a text appears differently on paper than on a screen and, thus, fresher
to the eyes. I personally read the text backward, i.e., from the last paragraph
to the first paragraph, in order to render the document even more different and
prevent me from going into “read a story” mode. Ideally, all linguists would
employ outside editors, a fresh pair of eyes, to review any resulting text. In
practice, the time and cost factors limit this practice from becoming standard
except in literary translation. Translation agencies employ editors, especially
those applying the various ISO standards. As I wrote, it is unpleasant to
receive red-dotted corrections even if such a result does not affect future
work. As for direct customers, the linguist is solely responsible but most
technical translators do not use outside editors as a standard practice.
Two more modern and comprehensive methods are
text-to-speech and AI. The text-to-speech function in Word is a simple manner
of reviewing a document using a completely different method. Instead of having
the eyes read a given sentence for the umpteenth time, the ears filter the
sentence. Thus, poor-sounding phrases and structure immediately hit the
linguistic warning bells and cause the translator or editor to reconsider the
wording. It is a tool that I may try in the future. Another trendier tool is AI
editing. Using one of many applications, it is possible to have AI analyze a
document, identify possible errors and suggest solutions within seconds. On the
surface level, it sounds quite magical, even ideal. My serious issues with this
method are the lack of confidentiality, the actual process and results. First,
as of today, in most cases, once a text, even without any identifying names, is
posted to AI, it enters the public realm, which may be a breach of the confidentiality
conditions. Some writers may be risking their copyright privilege if they use
AI. Furthermore, I personally find the process of writing prompts to define and
limit the range of errors as well as desired style I seek to be
overcomplicated. Even when a person overcomes that difficulty, the suggested
changes represent some collective image of the ideal such text, much of which
is not relevant objectively and/or subjectively. In other words, the benefits,
i.e., AI’s comprehensive and standardized approach, do not justify the risks,
i.e., the loss of confidentiality and hassle of writing prompts. Thus, I do not
use text-to-speech or AI.
However, recently faced with specific challenges, I
applied Grammarly to two especially long and complicated texts and found the
results positive on the balance. One text was a 20,000-word personal journal
while the other was a 5000-word contract. As time and energy were short, I
sought a tool that would identify phrasing errors and improve my translation in
my text without risking confidentiality. I used Grammarly, an
application that does involve a user fee. The process was simple, merely
uploading or dragging the text into the editing box. I then selected the type
of text, e.g. informal or formal. The result was a long list of possible errors
and suggestions for improvement. To give a perspective, the program created 500
comments for a 5,000-word text. The vast majority, around 80%, were false positive either because of context or personal choice. For example, the program was not
familiar with legal language and questioned many acceptable terms. In terms of subjective choices, I personally do not apply the Oxford
comma, the comma before the word and, nor place a comma before the word but.
Thus, I ignored those comments. On the positive side, it did identify many
passive sentences that I could render in the active voice. This comment was
of great value in the personal document but less so in the legal document even
if I try to limit use of passive structure in legal documents. Grammarly
also identified sentences that could be joined or split, suggestions that I adopted several times. Overall, review of these long texts using the program
involved several hours but produced better results in less time as compared to
rereading.
It is clear that linguists, especially translators,
must provide high quality products, meaning documents as clean of errors as
possible, especially in the light of AI-created documents. The longer the
document, the more difficulty that task is. Thus, translators and writers must
use a wide variety of tools to achieve the required quality. I intend to use
Grammarly on longer texts where a lack of time and freshness may harm the
quality of revisions. While linguists can choose their preferred tools, it is
clear that no specific method is a complete panacea.