The ATA Chronicle - September/October 2023 - 31

really ask MT to observe a
length limit or avoid letters
with diacritics. Many MT
engines also won't give you
multiple options for the same
source-language string. And
they won't typically translate
into a specific flavor of a
language, such as Colombian
Spanish, Austrian German, or
19th-century Czech. But GPT
will! And here we start seeing
the added value.
You can also consider
ChatGPT for all kinds of
editing and proofreading.
Source-text editing may be
useful if you're working with
a sloppily written source text
or a text written by a nonnative
speaker. Cleaning it up
first might very well make it a
better fit for your preferred MT
engine or any other workflow.
You can adapt an existing
translation if the target
audience changes. This
can be the case for some
informed consent forms,
for example, where you talk
to different age brackets
and use a different form of
address. You can also use
it for transcreation where
you need to produce several
options, back translations,
and rationales-all in a
table format.
MT quality estimation is
another possible use case,
and some computer-assisted
translation (CAT) tools and
translation management
systems (TMSs) are already
doing this through an API.
Also, if you realize how
helpful regular expressions
can be for your quality
assurance but can never
remember the syntax, you
can ask ChatGPT for help in
natural language.
In the realm of
terminology, you can:
www.ata-chronicle.online
y Feed a text into ChatGPT
and have any technical
terms and/or acronyms
extracted.
y Ask for definitions
in context.
y Research and produce
glossaries for various fields
as you branch out into a
new field of specialization
or review your vocabulary
ahead of an interpreting
assignment.
y Have an unformatted
glossary (such as random
notes in Word) converted
into a table format with
predefined columns.
y Or even superimpose
new terminology onto an
existing translation when
there's a terminology
change that affects legacy
translations.
Just like you can ask for
in-context definitions of
terminology, you can also ask
for context-aware synonyms.
This is great for unfamiliar
words, for additional ideas
in transcreation, or just to
build your vocabulary in
another language. You can
ask ChatGPT to rephrase,
paraphrase or just parse
convoluted sentences. This
is a fantastic help for lengthy
legal language where every
word matters and must be
accounted for.
With web browsing enabled
in the paid version, you can
get the latest information on
a certain topic, which can,
again, be a great time-saver
for interpreters. You can
ask ChatGPT to summarize
texts or, better yet, extract
information relevant to a
particular purpose. ChatGPT
is also great at multilingual
research; provide a prompt
in one language and you
can receive an output in
another language. You can
also use ChatGPT to research
a customer you would like
to market your services to, a
lead you want to engage with,
or an industry that's piquing
your interest.
As your personal office
assistant, ChatGPT can:
polish emails before you
hit send (and with that I
mean more than correcting
typos); turn bullet points into
continuous text, and vice
versa, summarize notes (e.g.,
from a meeting); highlight
and explain changes, for
instance, when a reference
document has been updated;
simulate a negotiation
to help you prepare for a
meeting with a customer or
supplier; help with travel
planning when you're going
to a conference (there are
several useful plugins for
that in the paid version); and,
very importantly, help you
get more prompts that are
relevant to your situation.
Outlook
We can expect incremental
improvements to GPT-4
as more people use it. If
there ever is a GPT-5, we
can expect it to be even
more powerful. Because
there is limited additional
data that a potential GPT-5
could be trained on, we'll
probably see a lot of smaller,
narrowly specialized, perhaps
industry-specific models.
We're going to see ever
more APIs. That is, other
apps and other services will
be connecting to the GPT
models without using the
chat interface of ChatGPT
that most of us know as
consumers. This is already
happening in a number of
CAT tools and TMSs.
We're going to see a wider
use of semi-autonomous AI
agents, such as Auto-GPT.
I would also expect to see
increasingly many onpremise
or private-cloud
solutions for big buyers
concerned about security,
privacy, and confidentiality.
And we're going to use
more multimodal models
(i.e., models that work with
image, text, audio, video,
and more; ChatGPT added
voice and vision in late
September).
What can you do to prepare
now for that future?
Experiment, if you haven't
already. If you have,
experiment more. Develop
those foundational skills. It's
like working with the
internet, with computers, or
with electricity in the early
days of those inventions.
Build your AI literacy.
Develop a feel for what's real
and what's AI. Work on your
critical thinking. And perhaps
most importantly, stay
curious and adaptable. The
pace of change will never be
this slow again.
Daniel Sebesta has been a linguist since
2003. He has a STEM MBA with a specialization
in pricing and strategy as well as a background
in linguistics. In addition to his translation
work, he consults with companies on pricing,
profitability, and language technology adoption. He is
the administrator of ATA's Language Technology Division
and identifies as an early adopter of productivity-boosting
technology. daniel@praguelinguistics.com
American Translators Association 31
http://www.ata-chronicle.online
The ATA Chronicle - September/October 2023

Table of Contents for the Digital Edition of The ATA Chronicle - September/October 2023
