En sincronía

Episode 37: Machine translation and the future of AVT with Yota Georgakopoulou

Damián Santilli, Blanca Arias Badia y Guillermo Parra Season 3 Episode 4

“En sincronía” is the only podcast for Spanish speakers devoted exclusively to the Audiovisual Translation (or Media Localization) field. Even though most of our content is in Spanish, we welcome international listeners to follow our interviews in English, such as this one. In episode 37, we talk to Yota Georgakopoulou, a leading audiovisual localization expert, specializing in language technologies. We’ll talk with her about machine translation and post-editing and how it’s shaping our lives in AVT and what to expect for the future. Also, in our sections in Spanish, Guillermo talks about the subtitles of the movie “The Menu”, and Blanca presents a new PhD thesis by Irene Hermosa on audio description in opera.

Interview, part 1:  00:08:38
Interview, part 2:  01:05:45

Consulta el episodio subtitulado y accede a la lista de enlaces en nuestro canal de Youtube.

En sincronía by Damián Santilli, Blanca Arias Badia & Guillermo Parra is licensed under a Creative Commons Reconocimiento-NoComercial 4.0 Internacional License: https://bit.ly/3jXTwjB

Prepara las palomitas. Está a punto de empezar En Sincronía, el podcast hispanoamericano
de traducción audiovisual con Damián Santilli, Blanca Arias Badía y Guillermo Parra.
Les damos la bienvenida al episodio número 37 de En Sincronía, el primer podcast de
traducción audiovisual del mundo. Soy Damián Santilli y estoy sincronizado desde Buenos
Aires, Argentina, con Blanca Arias Badía en Barcelona y Guillermo Parra en Menorca.
Hola Blanca.
Hola, ¿cómo estás, Damián?
Muy bien, muy bien, todo muy bien. ¿Guillermo?
Yo, bueno, yo parece que he volado antes de coger el vuelo porque sigo en Canadá pero
llego a Menorca la semana que viene, eso sí.
Ah, bueno.
Pero, oye, me encanta ya que me sitúes en Menorca porque la verdad es que tengo ganas
de un tiempo más razonable y de no estar sepultado en nieve.
Ya estamos esperando ansiosamente las vacaciones en febrero, así que después de un año con
HispaTAV y con esto y con lo otro, por fin algo de vacaciones pronto. Así que ansiosos
todo el mundo en esta casa por escapar un poco del calor sofocante de Buenos Aires.
Así que ahí estamos.
Bueno, hoy vamos a tener un episodio muy interesante donde vamos a hablar sobre traducción automática,
pero antes les quiero contar acerca de nuestro patrocinador que es Ooona, la mejor herramienta
en línea para trabajar con subtítulos.
En su suite de programas encontraremos herramientas para crear, traducir, revisar, transcribir,
convertir, codificar, pegar y comparar subtítulos.
Su interfaz web es muy intuitiva, es fácil de usar y nos permite trabajar como profesionales
de la manera más sencilla.
Además, podemos optar por distintos métodos de licencias desde una semana hasta un año
y también en su web encontrarán el programa educativo de Una y de Pool, que es una base
de datos de profesionales para que se encuentren las empresas y los subtituladores.
Pueden ingresar en ooona.net para probar el software gratis durante cuatro semanas, mejorar
su productividad y empezar a trabajar como profesionales hoy mismo.
Y en el programa de hoy vamos a entrevistar a Yota Georgakopoulou, que seguramente lo pronuncie
súper mal, pero es Yota, lo más importante que tenemos que saber, que es una traductora
experta en localización audiovisual con especialización en aplicación de tecnologías del lenguaje
en traducción audiovisual y en industrias creativas en general.
Dueña de la empresa Athena Consultancy, que es, ya vamos a ver en la entrevista, que pasó
por diferentes rubros, por diferentes empresas, por un montón de cosas y que ahora se dedica
más que nada a asesoría.
Y, bueno, ofrece servicios ahí para empresas muy importantes en áreas como estrategia
comercial, calidad, herramientas, flujo de trabajo.
Y un tema fundamental del cual vamos a hablar en todo el episodio, que es implementación
de traducción automática en los distintos, bueno, flujos de trabajo de las empresas.
Del punto de vista formal, bueno, Yota tiene un doctorado en traducción de subtitulado,
20 años de experiencia en la industria y, bueno, recientemente es fundadora, bueno,
hace un par de años ya, del capítulo griego de Women in Localization y también, como
ella misma se define, además de escritora, como influencer de la traducción audiovisual.
Así que, bueno, la entrevista va a estar muy bien porque va a estar enfocada a la actualidad
de la traducción automática, a tratar de entender más allá de las notas reverberantes
que pueden aparecer en las redes y el ruido que hace la gente o lo que puede decir.
Efectivamente, qué es lo que está pasando con la traducción audiovisual, cómo se está
aplicando y qué expectativas tenemos para la profesión en general para el futuro.
Así que, bueno, va a ser un capítulo muy interesante que seguramente dará mucho de
qué hablar.
Y, además, en el medio, después de la primera parte de la entrevista, vamos a tener las
secciones de Blanca y de Guille.
Vamos a empezar con Guille.
¿De qué va tu sección, este programa, Guille?
Pues yo voy a hablar de los subtítulos de una película que a lo mejor os suena, a lo
mejor no.
The Menu.
El menú.
Ah, sí.
Hombre, una película muy interesante, con sorpresa, no voy a añadir más, con un buen
reparto encabezado por Voldemort.
Encabezado por Anya Taylor-Joy, loco, que es argentina, eh.
Bueno, iba a decirlo también, la chica de Gambito de Dama, por supuesto, la jedrecista,
que es una crack.
Pero bueno, a mí me hacía gracia mencionar a Voldemort y sobre la traducción hablaremos
de terminología culinaria y de expresiones del mundillo y de, bueno, pues de todo esto.
Muy bien, muy bien.
¿Y Blanca?
Pues yo, aprovechando que tenemos tesis nueva, tesis doctoral nueva de traducción audiovisual
y accesibilidad, voy a hablar de la tesis nueva que hay, que en este caso se leyó hace
menos de dos meses en la Universidad Autónoma de Barcelona.
Hablé de Irene Hermosa, que estuvo en el HispaTAV, así que quizás algunos oyentes
del podcast ya la vieron presentar allí, y fue una tesis nueva.
Le mandamos un beso a Irene.
Un beso a Irene, desde aquí, que de hecho la tendremos en la sección, porque ya que
hay contacto, pues la hemos invitado para que no...
Ya sabemos que no te gusta trabajar a Blanca, ya sabemos, ya sabemos.
Y como ella es muy productiva y escribe mucho, pues venga, que hable en el podcast también.
Sí, sí, ya sabéis que yo, si puedo traerme a los autores de los trabajos de los que hablamos
en la sección, pues siempre lo hago y así yo me reduzco la carga y los oyentes tienen
la información directa, así que estupendo.
Muy bien, muy bien.
Guille, vos tendrías que intentar entrevistar a los guionistas directamente de las series
o películas que haces en tu sección, ¿no?
Sí, por qué no, les digo, sacado un ratito, no sé, tres minutos o cuatro para mi sección,
comentamos muy superficialmente la traducción que ni siquiera sabéis que se hace.
Pero lo importante es por qué usaron tal o cual término, ahí está, eso está.
Yo creo que Guillermo no puede hacerlo porque nos privaría de la voz del profesor Amor
durante unos minutos.
Claro, no podrías asurrar.
Se perdería el efecto ASMR, así que… Esto para mí saca un tema que me toca un
poco las narices y es que se empeñan en darnos plantillas, lo digo ahora porque creo que
es relevante para todo el mundo, pero cuando trabajas en su titulación digo yo que las
plantillas o las listas de diálogo las podría tocar a alguien que haya trabajado en el guión
de la película o que sepa de qué va la película porque a veces te encuentras con que pues
nadie sabe del tema y dices, bueno, ¿y los guionistas no pueden ponernos una notita en
un Excel y decir esto viene de aquí y no cuesta tanto, ¿sabes?
No, no, sobre todo en las plataformas de streaming que hay como una conexión directa con los
creadores, claro.
Con los creadores, ¿no?
Pero bueno, sí, sí, qué va a hacer, se olvidan de nosotros, pero por suerte dentro de dos
o tres meses ya va a ser todo de la máquina, así que no nos tenemos que preocupar.
¿Qué es lo que vamos a hacer?
Vamos a mirar TikTok de cosas hechas con inteligencia artificial y eso va a ser nuestra vida.
Gatitos de mentira.
Claro, claro, de sí, todo el tiempo.
Bueno, muy bien, hecha toda esta introducción muy poco útil, vamos a comenzar con…
Hemos perdido a 20 oyentes.
Lo prometemos que la entrevista es seria y vamos a trapo.
La entrevista vale mucho la pena, así que olvídense de esto de desvarío y vamos adelante
con la primera parte de la entrevista a Yota.
Hi everyone,
welcome, and we are here with Jota.
Hola Jota, how are you?
Hello Damian, hello everyone.
Thank you for having me, and happy
new year. Happy new year
to you too, and thank you so much
for being in our podcast. We know that you're
a very, very busy woman,
but we are
going to be talking a lot today
about machine translation,
and maybe the future of our profession
as well, and I think
you are the perfect person.
I'm so excited to have you here,
but for those who
don't know you, please
introduce yourself, and I don't know,
maybe if you want to mention how you
started in translation,
and studying, and all that,
so
everyone can know a little bit more about
yourself, and please also
tell us what are you
doing actually right now.
Of course.
Because I know that you do
a lot of things.
Well yes, it goes back to the previous
century, doesn't it?
It's my pleasure to be here, first of all,
and congratulations for the work
that you're doing with the podcast, I just wanted
to say it. I watched a couple of
episodes over the holidays, and
it was great work.
Thank you.
So for me,
there's nothing really unconventional
about how I started out.
I studied English lit, translation,
and I started working as a translator
from English into Greek, which is my native language,
and that was in the 90s,
but what wasn't common
that I did at the time was to do a PhD
in subtitling. Very few people were
doing that back then, and I
also started working as a subtitler,
as a freelancer, so I
know very well what it means to be a freelancer,
but fortunately for me,
when I was doing that in London
in the 90s, Greek was
an exotic language combination,
very well paid. At the time,
the work was done in-house in subtitling
companies. There were not that many Greek
subtitlers. It was
fantastic,
fantastic time for me. I really loved it,
and I actually helped the profession
mushroom,
I can say, because I was
one that set up and taught
the first modules on subtitling
in the UK at the University of Surrey
and Westminster.
I was teaching...
When was it?
From
1997,
I think,
until 2001.
I was teaching in the University
of Surrey and Westminster.
I was teaching the theory
of subtitling to MA students and also
the practice into Greek, and others were
teaching the practice into other languages.
And then I took an in-house
position at the subtitling
company, and I was asked to set up
a translation department for subtitling
into multiple languages. That was
in 2000. That was when media
localisation had just begun to be
centralised in London. Until
then, it was fragmented
in each country. You would do the Greek in Greece
and so on, but then everything started
to be centralised.
It was a very, very exciting time. I like to call it
the golden years.
The volume of work was growing
very much like it's growing today,
but the rates were fantastic.
It was a period of growth
for everyone.
That's not the case today.
There's a lot of growth
today, we'll talk about that, and there are a lot of
opportunities.
The rates are a bit different, I have to say.
That's another story.
That's a different story indeed, yes.
That company was called the European Captioning Institute
and I eventually became the managing
director until 2010
when it was sold to
Deluxe. Then I worked at Deluxe
for eight years. Everyone knows Deluxe.
I was there in a senior
role. I was leading research
in media localisation.
I was also in charge of various initiatives
to do with quality, workflows,
talent and vendor
onboarding, and I was
investigating the
implementation, the application of language
technologies in subtitling. I was
testing with all the vendors.
I was collaborating with
very high-profile universities
in computer science and machine learning
like the University of Edinburgh,
RWTH Aachen University
and the University of Zurich.
This was mostly
as part of EU-funded projects that
I was leading on behalf of the company.
It was a great
experience. I met a lot of
very nice and very smart
people and made some good
friends too. It's
funny because the decision I made
to specialise in language technologies at the time
everyone thought
I was crazy
for doing that.
They thought machine translation would
never catch on in our
field because of the creative
nature of our work, but
I always believed in
technology. Wishful thinking.
I was always
a believer in technology. I always
thought that the developments in our
field were following the developments
that we saw in the wider localisation
industry. For me, it was
a given. It would happen sooner or later.
This is why I wanted to be
there first.
Because I did that, I am
considered, because of that, a little bit of a guru
on this topic.
You are.
That's why you're here.
I'm not a scientist
though, right? It's funny sometimes
when I get to
speak instead of the scientist.
Anyway, the other reason that I was interested
in that was that we had a fantastic
archive of subtitle files, very
high-quality ones that I thought it would be
a good way to monetise them.
I actually got the opportunity to test
that
archive in the SUMAT project that
I worked on about
10 years ago or so.
That showed me really
the potential that good quality
data, subtitles in this
case, can have on machine
translation systems. I remember
we had
done an experiment and only a fraction
of this data, professionally
created, perfectly aligned
with the use of templates, would yield
much better quality than 10 times
as much freely
available data
that would come from amateurs.
This is why I'm a big believer in
customisation for
the content owners and the language service
providers that actually have
in their availability
a lot of such high
quality data.
Today, I left Deluxe 4
years ago and I've
been working as a consultant
ever since I set up my own firm.
It's called Athena Consultancy.
Now, I have the opportunity
to work for a variety of
clients, which has given me a much
wider perspective of the industry
and I'm loving it. Clearly, I work for
language service providers
but I also work for content owners
which were previously
my end clients.
I work for technology vendors
and actually, one of the
companies I consult for is UNA
who are the sponsors.
We know them!
They're sponsoring this podcast
so thank you, UNA,
for sponsoring the podcast.
Thank you, Wayne!
The projects that I
work on are very different.
They can be market research ones,
I do
a consultant software
development, I perform user acceptance
testing, I
set up, design and carry out
training on
subtitling, for instance, or dubbing,
on post-editing,
I draft guidelines, I do
strategic consultations, for instance,
on business development strategy, on
pricing, I even do marketing
work. It's really a mixed bag.
It doesn't get boring.
You even prepare PowerPoints, right?
I do that, yes.
I do. But the common
denominator in all of this is that
everything has to do with media localization.
If you're my age, you've probably been in almost
every position
in the
language service industry for media localization.
That helps.
Just
to sum this up, you always
worked with machine translation
and technology, mainly
focused on audiovisual translation,
right? Or also
in other areas as well?
Also for marketing texts.
Marketing texts, okay.
In the creative area.
Yeah, they're connected.
So you didn't
work with technical
translation? No.
The problems that you might have,
what you will come across with
machine translation in marketing is very similar
to issues that you would have
in software, I find.
Let's begin
with the current
state of machine translation.
It's difficult to
grasp the current
state from different perspectives.
Obviously, you know because you are
inside of the whole thing.
But when we see
translators commenting
on social networks or
even companies or associations,
it's not always
clear that
we actually know
how things are,
how good things are.
I see all the time
people saying,
like you mentioned before, it's not
going to replace us
or whatever, but when we
mention those things,
it's always, like I said before,
wishful thinking, right?
We don't say it because we know
it or for whatever
reason we say because
we don't want them to
replace us or
partially replace us.
What can you tell us about
companies working to improve
machine translation? We're mostly
interested also to talk about today
about the role that post
editing has in
mostly in LSPs
and how is this
affecting our profession?
I don't know. The idea of today,
the show, is to
end on a positive
side. I'm going to say this
from the very beginning. At the
end, we'll know if that's the
case. What can you
tell us?
There's a lot
actually to cover.
It is a little bit of a fact
question. Maybe it's a good
idea to start chronologically a
little bit. I don't know if
people know that there has been
experimentation with machine translation
for a lot longer
that people are aware.
I said I've been working with this
for the past decade, right?
There are studies and
early implementations, early
attempts to implement machine translation
in subtitling as far back as the
early 90s with rule
based machine translation.
There was a system in the US
actually automatically
translating
English closed captions
into Latin American Spanish ones
for the Spanish audience
with rule based machine
translation. I'd love to see those!
Probably very
literal, I would expect, but
it was a service for the
Spanish audience so they would have something
in their language instead of nothing, instead of
having to rely only on the English captions.
Then
there were also some attempts with
statistical machine translation
in the late 90s, even
by SDI.
None of these attempts really
flew, but
that was because of
the quality of machine translation
was worse in rule based.
It improved with statistical, but
it became a lot more fluent with
neural machine translation. This is
when I would say we
started seeing a lot
of implementations, about four years
ago. Can you briefly describe
and compare them?
You mean the technologies?
Yeah, the rule based, the statistical
and neural, so people know
what we're dealing with.
Rule based machine translation is
basically how the system
works, if I can put it very
bluntly, not scientifically.
Someone has to write
a lot of rules of the grammar
in one language, in the source language and in the
target, and these are mapped and you use a
lexicon, and this is how you get rule based
translation. So the old very
literal word for word
translations, this is rule based machine
translation. This is what it is.
Statistical machine translation
works on statistics,
but you use a lot of data,
so
parallel data, pairs of
subtitles and their translations.
And in a very
it's not
scientific at all to say this, but
it kind of works a little bit
like translation memories in a way, right?
The system sees the things that go together
and then comes up with the most probable
solution. Neural
is supposed to replicate
the human brain, and
the difference to statistical, it is
that it's a lot more
fluent, that is the advantage.
It can be so fluent that
it can fool you, it can have
errors in there that you don't
notice because of its fluency.
It can be a lot more creative, neural
machine translation. That's why it's
better for creative
domains like
subtitling. It doesn't
handle terminology as
well, so one of the things
that you will notice if you
use neural machine translation, if there's no
glossary integration, the proper
names will be messed up. They will not be translated
consistently most of the time.
You will see a person's
name translated one
way and then differently
further down. When
you use neural machine translation for proper
names and terms, you need to have some terminology
integration.
As I said, it is a lot more fluent, and this
is why it started being used.
The first implementation I
heard of was by iFlix
in Asian languages.
Their
systems were built by a company called
Omniscient, and then
Netflix announced that they were
using machine translation at the Media for
All conference in 2019
in Stockholm.
We saw a press release by
TransPerfect as well, advertising
that they were using
their own machine translation
and subtitling workflows. Ever
since then, I believe
more or less everyone has experimented
with it. Whether they implemented it
or not, or
whether they speak publicly about it or not,
I believe everyone has tried
it.
Is it very
recent? It is very recent,
yes. In our field, it is very recent.
I actually
remember training
some translators
that were technical translators
in subtitling, because
they needed to subtitle videos
as well as translate text.
I had a conversation
with them about subtitling editors
that was like three years ago.
I remember
being asked about
machine translation by them at the time, and I told them
there weren't
many integrations available
in subtitle editors back then.
I said, it's not integrated in this
editor, for instance.
They were like, what? How can you
translate?
I was like,
really? You'd be surprised.
You'd be surprised, yes.
In November,
I was at the
Translating Europe Forum
in Brussels. I was
chairing a panel there.
At that panel, I had Dr.
Anna Zareckaja from TransPerfect.
She's the Machine Translation Strategy
Director. I actually had
the opportunity to ask her a lot of questions
on the subject about what they do
at TransPerfect. Like I said, TransPerfect is one of
the first companies that implemented it.
The panel is actually
recorded and available on YouTube
if you want to follow it in more detail.
Yeah, we'll put a link on the
description of the show.
When I talked to her, she talked
about the production complexities
that determine where the
technology is used. Basically,
she was talking about content profiling.
Content
profiling is true
for both speech recognition and machine
translation. What it means is that
not all types of videos lend
themselves exactly in the same manner
to the use of this technology.
Even when we talk about speech recognition,
which is a more mature technology than machine
translation, and it's
a lot easier to see the benefit of
speech recognition in production, you'll
get videos with a lot of background noise
or overlapping speakers which can't really
be transcribed very well as a
result. It's the same
for machine translation.
She was also explaining
in that panel that although
it's their default way of working
at TransPerfect, to use
machine translation that is, it's the
type of content that determines
their use. They decide which
content to apply MT on or not
and they consult with their clients.
They even recommend to their clients
to use machine translation or not
to use it. For instance, the example
she gave was informational content. It
makes sense if you have technical content
videos rather than entertainment content.
The language will be scripted,
it won't be spontaneous, it won't be
full of figurative language or slang
so it's easier to
apply machine translation there.
She also said that
apparently they don't use machine translation
in any high-profile content.
That would be of strategic importance
to their end clients.
I've heard
of another interesting
use case of machine translation
actually in our
field and that's for
as an intermediate
step when you have
non-English source audio
workflows, which as
you know, you probably worked on some
are more
tricky to work on than the straight
English into any language that we
used to do for most of the
content until now.
Now that there's a lot of other languages as
a source, you still have an English
template as an in-between,
as a pivot, right? But
when you don't have multiple
streams,
so if you have only
let's say, I don't know, Greek into
Hebrew, Korean into Brazilian or
something and there's no other target
languages, it becomes very
and there's not many translators
actually that you can find that can
translate in these
directions.
It can become
very expensive to create
an English template
in order to complete the subject.
What I've heard of people doing is
using a transcript
of the source audio, then machine translating
it into English, to
use that machine translation as the basis
for the translation into the target language
because of the lack of a
subtitler that could translate
straight from the source.
It's an
interesting approach
I find, and I think
it will happen more and more because
of the increase
in non-English
audio content
and lack of translators.
Lack of translators, yeah.
In those language combinations, right?
So I think it will happen
a lot, and I think it's
interesting to see how those workflows
go because I think with some post-editing
to fix all of the mistranslations
and with the right amount
of annotations, I think
it could work because that intermediate
step won't be published anywhere
and anyway, in
pivot templates, it's actually
better to use, in my
opinion, a more literal
translation to
help the target
language translator when
they don't understand the source
because sometimes some flair
of the source can be lost when
you translate into English.
So yes, I expect to see
more of that.
So when you're telling
that they're incorporating
machine translation in these
workflows, that means
that the task that we
do is only to post-edit
or they offer,
I mean, it's like
it's just another option like it was
with translation memories
or we only get
paid, let's say,
and the only thing we have to do is to
edit the translation
output.
I'm
not sure if I'm following your question. If
you're talking about rates, I don't know how
people are paid for it.
The thing is that
there are some companies that
for instance, I do work not in
audiovisual translation but in technical translation.
I do have some projects
with companies that
the automatic translation,
the machine translation is there
and I can use it or
not but it doesn't change
what I'm paid.
I'm paid to translate.
I'm not paid to post-edit.
I mean, it's another resource
that I can use and actually
we all can do it
even if we're not working with an LSP.
If you're working with a direct client and you're
using Trados, for instance,
you can activate the
machine translation feature and do that.
But my question is
in those cases, for instance,
if you know, in the case
of TransPerfect or other companies that
you might know, we don't need to mention the companies.
We are hired
only to post-edit
or is it that the machine
translation is just another
feature for us to use?
I understand your question.
If we're talking about, forget
about that pivot example
that I gave you.
If it's English
into Spanish, then different companies
follow a different approach.
There are some that, and I agree
with this actually, this is what I recommend,
that
make the machine translation available
to their translators for the translators
to do whatever they want with it,
to use it or not use it, to have it
as another resource.
I believe they're doing
this in order to
give translators a chance to experiment
with it, to familiarize
themselves with how to use it,
to see how it's useful.
As I said,
this is something that's very new.
This is something that you would need to do in the beginning.
It makes sense
to do that as a language service provider,
also for the language service
providers to figure out how
useful the machine translation
is to translators,
how much it helps them, whether
it actually saves them time or how much
time does it save them, to collect
metrics, to do a proper implementation.
Implementing machine translation properly
is not an easy task. It takes time
and it takes
a lot of communication.
It takes money, of course, yes,
because it doesn't come for free
and change
management is a lot
of effort on a lot of people to change
the workflow. Other companies
go follow a different
approach and basically ask
translators to post-edit
and there are some
that immediately
also apply discounts in the
rates for this post-editing
without it necessarily being very
well thought through, in my
opinion. If you don't collect a lot of data
to really understand how much
machine translation helps
people, how can you come
up with fair and equitable
rates? It's hard
to do that if you don't have the data.
I do
agree with
the companies that follow
the more relaxed, let's
say, but also slower,
approach of allowing people
to experiment with machine
translation for a while, to get
their feedback, to collect the data.
This is what I've always recommended
and I'm glad that the companies
are doing it. These are the companies that will be successful
in my opinion.
Great. Thinking about
our future as professionals,
it looks like
we need to be trained on this,
on post-editing.
If we are going to
the second scenario that you described
in which we are paid
as always, but
we need to post-edit, it's not
an option for us to do that
in the future. Do you think
we all have
to be doing training on that?
First of all,
even for the companies that are allowing
translators to experiment and so on,
clearly the goal for everyone,
it makes sense that it is to save time
to begin with, to do
the work faster, which is very
important, especially as Windows are
becoming shorter and shorter,
release Windows, and it is
eventually also to cut the cost.
You hope that the machine
translation will get better and better,
the work on the translators will become,
post-editing that is, will become easier
and easier and faster, and
eventually you would expect to
pay less
in total. But this
less in total doesn't mean less money for the
translators necessarily,
right? And I'll explain
what I mean.
Yeah, please
do explain that.
Because I'm not seeing it.
Okay, well, basically
let's
take for example this famous
30%, right?
Let's assume that
using a specific
machine translation, you
do the work
faster, you reach,
you can deliver the same
quality of a file, and you do
that 30% faster,
right? And let's also
assume that after 8
hours of this type of work, you're
not more tired than
you would have been, had you just been
translating from a template without machine translation
for 8 hours, right? In this
scenario, if you were
paid 30% less per minute
or per word or whatever
your rate is, you would
end up making the same amount of money
because your hourly rate
would be identical.
Because at the end of the
day, your
fee, your salary, the
money that you make, is how much money
you make per hour,
right? If you
are paid, I don't know, 10 cents per
minute, for example, to
translate, and you're
paid 5 cents to post-edit, just making
numbers up here, but
you can translate more than
double the amount of
content when you're
post-editing, then
you're not losing money with the 5.
It's not a lesser fee,
right? It's how much you make per
hour. You work for 8 hours
and you've made, for instance, 100
bucks, and in the other scenario,
have you made 100 bucks or have you made 120?
If you've made less, then you lose money.
And people
don't really think about it
that way at first.
I think it's
very difficult for translators to think
beyond the rate per word
or rate per minute in this case of
subtitling. There
have always been propositions of
thinking about how
much you should earn
per hour and not per
watch, whatever you want to call it.
If
you look at associations
here in Argentina, in Latin
America, that they can provide
a guidance
on rates,
they always do it per word
for instance.
I think it's a
paradigm shift that is going to
be complicated. You don't necessarily
have to be paid per hour
as long as you know
what you make per hour, which is important,
right? Yeah, I know, but in that
case, everyone is going to be saying,
the rates are now
$3
per minute, and
five years ago, they were
$4 per minute. They always
do this easy math.
Of course, and
maybe in the 90s, they were
$5 per minute or whatever
they were, but back then,
timing a subtitle
was so much more cumbersome
than it is today.
Or translating from a VHS.
Yes, it took
you a lot longer because you had to do
a lot of other things,
even the fact that you can, I don't know,
work from home. For me, when I had to
do a translation from a template
back in the 90s,
I would get a VHS
at home, I would subtitle
in my computer at
home, then I would have to go in the office.
There was traveling time involved.
It was a long distance from my home
to that particular company's office.
Go in, review the file
there so I could see my subtitles
against the
video there. I couldn't do that at home.
It wasn't a possibility back then.
Sign it off, and then I
would be paid.
These were hours that I
was spending to complete the job.
If you just think about
the per minute rate,
it doesn't really say much.
What you really should be thinking is, how much am I making
per month? How much am I making per
full working day?
Doing it per hour actually
helps.
My proposition
also
to language service providers
is to share the benefits.
If
machine translation does
indeed save people, for instance, 30% of
their time, and they do want to
apply some sort of discount to the rate
that they shouldn't actually
apply the exact same discount
as the time saving, they should
actually apply a
lesser discount so that
everyone benefits. The translators
who
do do this post-editing
work
and who offer feedback and who
help improving the machine translation system
also earn
more per hour as
a result, not in the absolute
per minute rate, but per hour
as a result, so they also
benefit.
Back to
the training
question that you asked, because
you asked me, should we all
be trained?
Be trained on being post-editors.
I do think post-editing
is part of the future of
subtitlers. There's going to be a lot of
content that
will be asked to
post-edit more and
more of it. There's a lot of content that needs to be
localized, and
there's a lot of content that can't be
localized yet
because of cost reasons,
which could be made available
in other languages if
the cost threshold was lower,
and maybe
it can't currently be... So there's even more
work there. There's even more
work, yes, and
maybe machine translation can actually
help make this
content localized with
some post-editing.
There's all the major streamers that people work
for, but there's a lot of ad-supported
streamers that don't have
the same budgets for localization
that major studios do, so they don't
localize, but maybe they have
some budget
for localization, which could
be serviced with a specific type
of workflow that could involve
machine translation.
Of course, people
should be trained on post-editing, because
if you want to
be able to take on such work,
I'm not saying you should,
if you're in the privileged position to just
if you can afford to be
picky about your work
and about the tasks you perform,
by all means, you should be doing that.
You should be taking on the
tasks that you, I don't know,
enjoy the most, or that you make the most
money out of. That's only
common sense in any
profession, but there is going to be a large
volume of work out there that will
be available for post-editing,
so if you want to work on such
content, then it is a good idea to be trained
and to train new translators
at universities,
because post-editing
is just a different task,
so you need tips, you need to
understand how to do the work, you need
guidelines,
you need to learn
some tricks on how to perform the tasks
so that you can do it better and
faster, and you need practice.
But as I said, you can
find your own value and you can
decide
which tasks to work on.
If you don't want to work on post-editing,
you don't have to. It's exactly the
same thing like
proofreading, and I don't mean that post-editing
is proofreading, because it's not. It's a different
task as a task, but it's the same
logic. Some people
like to proofread
other translators' work.
Most translators I know don't
enjoy these tasks. They
enjoy to translate from scratch
and most translators at university
when we all
went to university, we thought we were
going to be authors, right?
Right, text.
We didn't become translators because we were
dreaming we were going to correct other people's
work. For most people,
I think that is
true. So
if you can afford to pick the
jobs and only pick translation
tasks, by all means, do that.
It's the same thing.
The question is mainly because
it looks like
more challenging to
be doing that
than just translating, because
here, I'm
talking just on Latin America, but
you constantly see
translations on TV and
those are made by people.
So if...
We see them here as well.
Yeah.
I don't mean to intrude.
So
if you're going to be a good post-editor,
you need to be
a better translator than
you would have been if you would have been
translated. I don't know if that makes sense,
but if
some people are translating
now and their translations
are worse
than what Google
or whatever can offer,
then to correct that,
to edit that, you need
to be a much better translator.
So it's
a huge challenge and I hope everyone
can see that.
When people do say that machine translation
is going to replace translators, I don't believe that
first of all. If it replaces
anyone, it would be the bad translators
actually.
In my opinion.
And yes, I do agree you need to be a very
good translator to be an editor.
Even when it comes to proofreading and
review tasks, I always say
you can't use a new person
to perform
these tasks. You need an experienced editor.
Yet many companies do that, right?
Wrongly, in my opinion.
They should not be doing this.
Because that has a result in a lot
of errors in the end.
You do need an experienced person.
And with post-editing, you do need training.
And it's true
that the more experienced
you are, the more
the better you will be able to pick
up on the errors, especially when
there are many of them.
Machine translation will improve. The errors will be reduced
further and further and further.
This will keep happening.
But when there's many,
it's easy to miss
some, isn't it? So the more experienced a translator,
the more
easy for that translator to
identify these errors
and fix them.
And studies that have
taken place have actually shown
in machine translation and post-editing
in general, not necessarily just in subtitling,
have shown that people who are
inexperienced tend to make more
superficial corrections,
whereas the experienced translators
would go deeper into the text and
basically correct everything. Actually, over-correct
is what studies have
shown.
But I'm also thinking that you need to be
very creative
because most
of the easy tasks, the mechanical
translation part of
subtitling in the future
will be just for the
computer. So you'll be
always dealing with
the most
difficult parts of
a translation. So if you
are a new translator or
a VAT translator,
you won't be able to deal with those
texts, in particular those parts
of the text.
So, Gisela?
Yeah, sorry, it's always the same.
Like, I
was about to say, like, Jora, thank you for
your presentation. I feel like I'm attending a
conference here.
It's
really interesting.
Yeah, I want to mention, you were
talking about errors
and this is something that
I don't really know the answer to this
question, so it's also
the reason why I'm asking it.
But also for the audience, because
one may think, if at some point
we stop
translating from scratch or
we almost stop doing that
and everyone becomes a
post-editor and we feed
the translation machine models
with post-edited
material, how do you
think
that would impact the results we get?
So, if you say that we
will get increasingly less and less
errors,
how does that
correlate to the fact that we're going
to be post-editing
more and more and if perhaps
there are different quality levels
that we can feed
to the system? I don't
know if that makes sense.
Okay. When I
said post-editing more and more, I didn't mean
post-editing having to fix more errors.
I mean more volume of work to be
post-edited, right?
But what I mean is that there will be less
errors in the translation.
Exactly, yes.
You can see that. Check machine translation
from two years ago and check
machine translation systems from today
and they are better.
Check machine translation systems
that have been trained with more and better
data, they produce better quality
than machine translation systems that have
been trained with less
data or less
in-domain data. So this
will keep happening.
I don't know if you're
trying... Yeah, but I'm referring
to translation not just
measured as
the number of errors per
word or per subtitle
or in a
positive sense, so
the ability to
find creative solutions if those
solutions...
It's not just there will be less
errors, which kind of makes
sense, but also
if
nobody comes up with a certain
solution because everybody's post-editing, how
will that ever be used
by a translation machine
model? That's what I mean. I think you're talking
about impoverishing
language. Is that... Yeah,
yes. Okay, there are
discussions on this. First of all, there's two aspects,
right? One aspect is a lot of people
say that machine translation
post-editing results in more errors
in the end file,
in the deliverable.
I believe this can be
avoided by using
a proofreader afterwards. As I said,
if we are at the point when there's a lot
of errors in the post-editing text and
there have been studies, Jan Pedersen did one
that he presented at Languages in the Media
recently, where he compared
the work of a translator
translating from a template and
translating with the help of machine
translation, so basically post-editing, and then
they compared
how many errors did these translators have
in the files that they delivered, and they
found that they had more errors in the post-edited
texts.
This could have been for many reasons,
including the translators not having
any training in post-editing and not
knowing what to do exactly.
But in my opinion, you shouldn't look
at that. You should look at what gets delivered.
So there should be a second pair of
eyes. There should be a proofreader
there. And it is a good idea
to actually measure what work the
proofreader does. Do they end up
having to do more work on a post-
edited translation versus a translation
that wasn't post-edited?
And what are the
errors that are left in the final
file, if any?
I actually asked this question again
to Anna Transperfect because I was curious
if they had any metrics on that.
They didn't, but what she told
me is that they couldn't
find more errors in the final files
because there were no
client complaints about them.
So there were no rejections. There were not
more rejections from the clients.
So there were no more errors. I think
it is worth checking
more
the work of the
proofreader.
But having said this with the right
training, I do believe that the quality
can be the same if you have a second
person in the loop, right? As you
should always have because
as Damian said, there's a lot of
just by when a translation
goes through just the translator and not a
second pair of eyes, it's definitely not the same
quality. Now
the other angle on this
is habituation.
This is something I don't necessarily have an answer
to. So people say that
translation limits your thinking,
right?
Some people complain
about that,
about how the machine translation
saying that it won't allow you to
come up with the creative translation that you would have already
come up. This can be
solved with
training and with the right user interface.
So this is something
that I discuss a lot with translators at
Austerity is not having the machine
translation in your face and
over typing it to correct it
but having it on the side
kind of a different column
and I know a couple of companies have actually
implemented this in subtitling and those
translators that work with those particular
systems, I don't want to name any
names, are actually much happier
using the machine translation because it
is on the side as a maid
even psychologically it makes a difference
and also
it's not in your face, it also gives
a translator a chance to think
before looking at the translation.
When I was translating, and granted that was a
very long time ago,
I remember I was reading the source
text and listening to the video and
I was already thinking how I would say
this in Greek, right?
This is what I advise people
to do when I train them to try
to think of the translation before
they look
at the machine translation
output, before they think about whether to use
it or not, and I believe it does take
some practice so that
it doesn't impede your
creative thinking.
Now, the bigger problem though
that is a problem, I believe,
is the variety
in the language.
So, let's say you're post-editing,
if you have a translator, and there's a lot
of synonyms in language, right? And when
we speak, generally
we will use different words,
we will use a lot of variety, we will
say, I don't know, hello, in a lot of
different ways
in a language, and we won't say the same word
over and over and over again, even if
good morning is a correct translation of
a given phrase of Kalimera, you won't
necessarily say always
good morning, maybe you will say morning, maybe you will
say something else, right?
It's just off the top of my head.
Now, however,
when you post-edit, if you have a translation
that is correct, you won't
touch it. You don't touch it, right?
So, if the translation always comes up
as those two, you know, good
morning, and not of any of the other
variations, then you won't
change it, because it is correct
stylistically as well,
if the style is right.
So, there has been another study that has
shown, basically, that in post-edited texts
there's less variety
in the language,
in the use of the synonyms,
than you would find if text
was created from scratch.
This is something that I
don't have an answer to. I do want to
point it out.
So, I don't know
if it will become
an issue in
the future. But
on the other hand, I keep thinking,
you know, it's not as if
all translators have this great
artistic flair.
That's true, yeah.
If you just look at translation memories, right,
which are basically a collection of
human translations,
there's a lot of junk in there.
People have to clean out translation
memories all the time because of the errors
that they find in those translations
that were created by
translators.
So, a lot of
the subtitles out there
are average
in terms of the quality and
the language use.
And
post-editing also has to do with
the ability of the translator. I believe
someone who is an excellent translator
is likely
to do excellent work
as a post-editor as well.
So, that's
the idea. You need to be better.
You need to be better at what you do
anyway. Better than in the past.
If you're not a good
translator, you will have more errors
even in work
that is not
post-edited.
Some are adequate.
I think
machine translation post-editing
is a good
option for
workflows
or content where
adequate is enough for the
end client. And I think it is an
opportunity for language
service providers and for translators
to offer different quality levels
that Guillermo mentioned for different
content that is priced differently.
Because in the past
I remember
every single LSP would say,
I have the best translators and my
translations are the best. It's impossible
for everyone in the market to have
the best translators. Above average.
Everybody is above average.
Someone has to be the average one.
All of my students
got hired if they wanted
a job. Not all of them were
brilliant. Some had just
passed their MA and
others got a distinction.
I'm sure they didn't produce the same
translations when they worked as
translators.
That's a different
argument.
That's another show. A different episode.
Completely.
Translators and their idea of how
they translate.
And I also,
in my opinion, should be
reflected in the pay.
If you are translating
English into Greek, you get paid
whatever it is these days, I don't know,
three dollars, four dollars,
five dollars per word. It's
not a blanket rate. I don't
translate the same way as someone else.
My quality is not the same after
20 years of experience as a person
that just graduated. We should not be paid
the same. We should be paid according
to the quality that we offer, just
like in any other
profession.
On that note, we are going
to take a small break
here and we're going to
listen to Guillermo and Blanca
and their sections
and then we'll be back
with more with Yota.

Quédate y descúbrelo con Guillermo Parra.
Bienvenido a Subtítulos con Carácter.
Saludos, curiosos diamantes de los subtítulos.
Hoy voy a hablaros de la película The Menu, el menú,
que llegó durante el mes de enero a Disney+,
y que básicamente trata de una pareja que viaja a una isla exclusiva
para disfrutar de un menú de degustación que solo los millonarios se pueden permitir,
preparado por el chef Slovic.
Este misterioso chef, encarnado por Ralph Fiennes,
dirige a todo su equipo de cocina casi como si de una secta se tratara.
Y esto tiene mucho que ver con la primera cuestión traductológica que quería comentar,
y es que hace que todos sus empleados lo llamen chef, pero chef como nombre propio.
Esto se mantuvo en la traducción al español de España, por ejemplo,
donde la pareja está hablando de él y Nicholas Hoult dice
¿Habré cabreado a chef?
A lo que Anna Taylor-Joy responde
No hace falta que lo llames chef, porque no es su nombre, se llama Slovic.
Sin embargo, en la versión latinoamericana se optó por utilizar el artículo,
el chef, porque al fin y al cabo es su profesión, no es su nombre.
Lo que pasa es que al llegar a esta conversación la solución no tiene sentido,
porque Nicholas Hoult dice ¿No crees que el chef esté enojado conmigo?
Y ella le responde, no tienes que llamarlo chef, Tyler.
Pero claro, él lo ha llamado el chef, se ha referido a él por su profesión,
cosa que en este contexto tiene todo el sentido del mundo,
así que la objeción no se entiende,
o al menos no se entiende sin tener en cuenta el texto de partida.
Y cambiando de tema, como estamos hablando de la película sobre cocina,
sobre gastronomía, me gustaría mencionar, aunque sea por encima,
la cuestión de la terminología gastronómica.
No soy ningún experto sobre la materia.
Lo que sí he notado, por ejemplo, es que en la versión de España
iban más allá e introducían términos que no estaban en la versión original en inglés.
Esto, dependiendo del contexto, puede funcionar o no.
Creo que en este caso funciona muy bien,
sobre todo porque le da un toque de sofisticación, de autenticidad.
Os pongo varios ejemplos.
Cuando en inglés hablan de twelve customers,
que son los clientes que hay en el restaurante,
porque realmente es así de exclusivo,
en la versión latinoamericana lo traducen como son doce clientes en total.
Doce clientes, que es traducción perfectamente correcta de customers.
¿Qué pasa? Que la versión española va más allá y dice doce comensales en total,
y es el término específico de las personas que se sientan en una mesa a comer.
Otro ejemplo.
Cuando hablan del vino que toman con la cena,
el somelieris lo describe como just a wonderful match para un plato en particular.
Esto en Latinoamérica se tradujo como que combinaba maravillosamente con el plato.
Y en España dice que marida muy bien.
Y esta elección del marida me ha encantado precisamente porque se usa mucho en este contexto.
No hablamos solo de catas, sino de maridajes también.
Entonces, creo que he sabido utilizar muy bien la terminología
para recrear el lenguaje de alguien del mundillo.
Porque al fin y al cabo está en un restaurante donde sale a 1200 dólares el tenedor.
Otra cuestión que me ha llamado la atención al comparar ambas versiones
es el uso del tratamiento formal o informal en segunda persona.
En la versión latinoamericana hablan principalmente de usted.
Es un contexto muy formal y es lógico que sea así.
Mientras que en la española hay alternancia en algunos casos entre usted y tú,
que yo creo que está justificada por el tipo de película que es y por cómo son esas escenas.
Se trata de un thriller psicológico donde las cosas no son lo que aparentan,
entonces aparentemente se usa el usted, es formal,
pero hay momentos en los que esa barrera de las apariencias se rompe
y los personajes son más bruscos y más cortantes de lo que uno esperaría en el contexto.
Entonces ahí creo que el uso de tú funciona muy bien.
Especialmente cuando se dirigen a alguien por su nombre de pila.
Pero de todos modos creo que es algo relativamente subjetivo
porque de hecho en las versiones dobladas tampoco siguen el mismo criterio.
Y hasta aquí el tema de hoy.
Recordad que si queréis seguir al tanto de novedades en subtitulación,
podéis seguirme tanto en Twitter como en Instagram
y estar atentos al hashtag subtítulos con carácter.
Gracias por escuchar y hasta el próximo episodio.
Subtítulos realizados por la comunidad de Amara.org
Hola, ¿cómo estáis?
Ya he dicho en la introducción que tenemos tesis doctoral nueva
en el ámbito de la traducción audiovisual y la accesibilidad
y no podíamos dejar de escapar la ocasión
de invitar a su autora a participar en estos minutos divulgativos
para saber un poquito más de este trabajo nuevo.
Es una tesis que ha hecho Irene Hermosa
en la Universidad Autónoma de Barcelona,
en el Grupo Transmedia Catalonia, bajo la dirección de Miquel Edo.
El trabajo consiste en un estudio de corpus.
Los estudios de corpus son una metodología
que ya tiene un recorrido de más de 30 años
en los estudios de traducción.
Se han hecho aportaciones muy interesantes para nuestra disciplina.
En este caso, la gran aportación es que todavía no había ningún estudio
en profundidad de la audiodescripción de la ópera
como un texto, digamos, de pleno derecho.
Los guiones de audiodescripción, ¿cómo son lingüísticamente?
Esa es la gran pregunta de investigación que ha respondido Irene,
con la que, sin más dilación, os dejo.
Le he pedido que nos mandara un audio
y que nos contara absolutamente lo que ella quisiera de su tesis.
Sabemos que el doblaje y hasta cierto punto los subtítulos
tienen rasgos de oralidad, pero ¿y la audiodescripción?
¿Y la audiodescripción óperística?
Estas fueron algunas de las preguntas que me planteé
al comenzar mi tesis doctoral en la Universidad Autónoma de Barcelona,
que defendí el pasado 20 de diciembre.
Mi objetivo más general era definir el lenguaje de la audiodescripción
y la audointroducción, que es el fragmento de 5 a 15 minutos
que se oculta antes de que empiece la representación
y que suele incluir el resumen de la obra,
sus elementos visuales más relevantes, el elenco,
todo ello mediante un estudio de corpus lingüístico
con una muestra de guiones del Liceo de Barcelona
y del Teatro Real de Madrid.
Un segundo objetivo consistía en definir
la priorización de los signos en la ópera,
es decir, la escenografía, los audiosubtítulos,
la danza, la música instrumental, el vestuario,
también en los guiones de audiodescripción.
Lo que observamos en las audiodescripciones propiamente dichas
son características más cercanas al registro oral
que en el caso de las audointroducciones.
Algunos ejemplos son una mayor proporción de verbos,
es decir, un mayor énfasis en la acción,
oraciones y palabras más cortas que en las audointroducciones
y también menos palabras infrecuentes que en las audointroducciones.
Además, estas últimas son más difíciles de entender
que las audiodescripciones propiamente dichas,
tanto desde un punto de vista léxico como textual.
Hay que aclarar que la audiodescripción operística
representa un reto particular,
diferente de la audiodescripción fílmica, por ejemplo,
por la necesidad de incorporar constantemente los audiosubtítulos,
que recogen los diálogos que se proyectan sobre el proscenio,
en lo que llamaríamos los sobretítulos.
Esto, además del resto de todos los signos teatrales
y de la música instrumental y vocal,
que han de respetarse sobre todo en las áreas
de los pasajes más conocidos y reconocibles.
Si os interesa conocer los resultados completos de mi investigación,
pronto se publicará mi tesis en acceso abierto
y el título es
La audiodescripción para ópera en España,
estudio desde la lingüística de corpus y la semiótica.
Pues le agradecemos muchísimo a Irene
que haya participado en este episodio
y os voy a dejar a todos en la cajita de información del podcast
la web donde muy pronto estará disponible su tesis
para quien quiera consultarla.
Muchas gracias y nos vemos en los próximos minutos divulgativos.

this interview so
the question now is
why do you think that professional
associations
I mean, I like to say
most of the community of professionals
as well, always have
such a strong opposition
to machine translation
and, well, what is
currently happening with LSPs
and their workflows and all that.
Well, I think that's absolutely normal
actually. It's exactly what happened
in the 90s when
translation memories were introduced in
text localization workflows.
Translators
were against the use of
TMs for a while, and now you have
translators... Some are still against
still today.
Are they? Because today I actually
hear people, where is my translation
memory? I know a few
I know a few that don't like to
translate with the translation tool
but overall
overall, though, you actually
hear people asking about
translation memories. I mean, even
Atre has asked for translation memory
integration in subnet editors, and they think
this will be very useful, and I agree.
I think the reason
maybe
is that a technology, when it's introduced to
begin with, is not as
good as it is today. I'm sure translation memories
were not as good in the 90s as they
are today, right?
And machine translation will be
better in three years than it is now.
So, that's that
to begin with.
The technology needs to be fine-tuned.
But I think the
main reason is that
oftentimes, I mean, clearly
companies do want to use technology
to make things, as we said, faster
but also to lower costs.
The problem is when this is done
not in a fair and equitable
manner.
I always say that
half the story in machine
translation, the use of machine
translation, is how it's implemented.
It doesn't matter how good your
machine translation is.
If you mess up the implementation,
you're not going to be successful.
And what is implementation?
It is, first of all, the user interface.
Building it properly
so that it's
user-centric, so that people are really
assisted in the tasks.
If you are
to use machine translation in some
very cumbersome way where you have to
copy and paste and do a lot of
keyboard commands to get the
text there so as then to then
correct it, then, of course, it's a nightmare
for the translator. Who likes that?
You also need to be...
Exactly. So, you also need
to be trained in the task. You need to be
given clear guidelines about
what to do
to begin with. What do people expect of you?
How much do they expect you
to correct? One of the main
things in these guidelines
that I've worked on, actually, is
to provide guidelines as to
what are the types of errors. This is
a core part of training
on post-editing. What are the types of errors
that people are likely
to encounter in machine-translated
text? And how...
And examples of these
errors. And how are
these errors fixed? And how
much should you fix them? A lot of people
talk about post-editing, actually.
And if you look at the TAO's guidelines
and some of the other guidelines that are
around, actually advise that you
should not correct style errors, which
in my opinion is wrong
when it comes to creative texts.
You absolutely must correct
style errors
when you are working in a creative
field like marketing
or subtitling. That's what the text is
all about. Style, right?
If you get that wrong,
if it was an informational
video, it's a different story.
Maybe there you
don't need to or the client doesn't want to pay
for that, whatever. But if it's entertainment,
it's not going to be entertaining if it's not the
right style.
So you need training. You need
clear guidelines.
You need communication
about this change. Any
change management, any
change in the workflow requires change management
and communication is
a core part of it.
And of course, most
importantly, it has to do with
pay. You really need,
as I was explaining before,
you really need to pay people appropriately.
If you're going to be,
if you end up earning
less per day, per
hour, per month, you're not
going to be happy. You won't want to do this. You will
feel you're ripped off.
If you end up earning the same
and even better, more,
then I think translators
would be more willing to
embrace the technology because
they would make more money
in the end.
I don't always see this, but
what I sometimes
see, for instance, in Twitter
or whatever you want to call it
in other social networks,
is that
most of the,
it's not that they're saying
oh, these companies, they are
the worst because they're using
this bad and that's
why we
get paid less or the quality
is that. They always are
these, I don't know,
blog posts or whatever
talking about machine
translation itself.
Not
that the problem actually
is
maybe what we are seeing now.
For instance, the third
scenario that you described before
where companies use
whatever machine translation for whatever
kind of text.
But
the problem is the technology itself
and not how
companies are using the technology.
I mean, that's why you read.
It's not the problem
actually that. I disagree
there. I disagree there and I think
it's easy
to argue against the technology
and say it's not good because you don't want it
to be used. Why don't you want it to be used?
Because
there's two factors in machine
translation. There's the money factor.
Are you making the same or more
or less amount of money?
Nobody wants to make less money
for the same amount of work
including tiredness.
I'm not talking about only working
the same amount of hours but being
not more tired
than that. Interpreters only can afford
to work a few hours per day and not a full
eight-hour day because the cognitive
load is too much
and their work is very toxic.
So if post-editing
leaves you half dead after eight hours
of work, clearly it's not the same
as if you translate for eight
hours, you could keep going for another four
hours perhaps because
you don't feel tired. So there's a difference there.
If you feel just as tired,
the question is how much do you earn
at the end of that? And this
is part of what people don't like.
The other part is maybe
they don't enjoy the actual
activity. And I get that
because translating
film is fun,
right? I mean, I used to
teach at university subtitling
and I used to teach legal translation
as well. My subtitling class
is packed. It was the most
popular class in the whole course.
My legal translation
class was the least
popular class in the whole
course. I wonder why is that?
Like five students there, right?
So of course it's fun
and this
is why people go into it.
A lot of people go into it. And if you think
about it, it's not really a well
paid translation type.
It's actually the second
worst paid in my opinion.
The only thing that's worse paid than
subtitling is literary
translation.
More or less any other type of translation
is paid better if you do the math.
How much do you earn per
hour?
But it's
fun and this is why
people want to do it.
I always like to see your name on the screen and all that.
Yes, and it's fun.
There's another discussion there.
Maybe we don't go there.
So
basically with post-editing
maybe
for some people
what you're doing to them is you're taking the fun
away, right? So if you're
taking the fun away, at least they should
be making more money. So this
is my argument as well in
sharing
the benefits
from the post-editing
work.
I do want to say something more
in terms of rates.
Please do.
It is good to think about
the problem with the rates in our industry
was a problem long before
machine translation was there.
Around the
turn of the century,
up until the time that I kind of stopped being a
freelancer, the rates were great
at least in London. They were fantastic.
I was earning a lot of money
and then they started going
down
both for language service providers
and for the individual translators and they
kept going down and actually last year
is the first time I ever
heard in my career of rates
going up again. Not for everyone,
but I've heard that from a lot of
language service providers and a lot
of translators that they managed to increase
their rates because of the
demand and the lack of
supply. So rates
was an issue long before machine
translation came about.
So you should think
the companies that people are
complaining about, were they paying well
before machine translation
for non-post-editing jobs?
Were those rates good? Is it
really that the machine translation rates
are the bad ones or are all the
rates bad
in that
specific vendor
might be offering?
Those vendors are probably worse now.
Guille? Yeah,
you mentioned that about
translators being more tired
or less tired. We have
to do some research on that
and
to me that's closely linked to
the concept of
burnout and how present
that is today and
how it actually has an
impact on translators'
lives. And
it also reminds me of
languages in the media conference
where in one of the talks
I don't exactly recall which one
but one of the executives said
like, well, you work in this
in subtitling for
a few years and then go somewhere
else, follow your dreams. So how
do you feel, well, if
post-editing is
less fun, we don't know if it's more
tiring or not, if it's more demanding
or not, cognitively speaking.
But
do you know of any studies about
how that could have an
impact on burnout, on people leaving the field?
And that
also has a consequence like
then it's harder to find
people to do the job or to do the job
properly or have enough
experience. So how
do you feel about that? So first of all, let me just correct
something. I didn't say that
post-editing could be fun for some
people. I do know people that like
to edit, that like being editors,
that prefer editing and proofreading
to translation.
I do know such people. I just know fewer
of them than
the other way around, right?
So post-editing
for many translators could be easy.
Let me say something
else first and then I'll get back to your question.
So
it is very interesting and I think
it's very important for translators to know
how they work,
to be more self-aware about
how they work and how they make their living.
I don't think translators are.
I've had
a few cases when
I was conducting experiments on
machine translation post-editing
when people were surprised
by how much
faster, for instance, they were
when they were post-editing
according to my metrics
than what they expected,
than what they thought.
I'm not saying they were lying.
It's just that that's what
their expectation was. I mean, even
Damian, when we worked together
a few years ago and we worked on
a comedy and a documentary,
I remember, Damian, you
said you expected machine translation
to help you more in the documentary and in the end
the metrics showed that
the machine translation
improved your speed further, more
in the comedy
instead of the documentary.
And you were surprised by that.
You were not the only one. That happened with others
as well. I don't think people
are as aware as
they should be.
I think if your focus
is to make more money because you need to make a living
because you have a family to support
and so on, which we all do,
you need to be more aware as to
where and what tasks do you
make more money out of.
Go do legal translation.
Not necessarily.
Not necessarily.
What is it that you do? Because you just
enjoy it, right? And sometimes people
do come in the industry because they enjoy
it a lot. It's not all about money.
Exactly.
They don't get paid necessarily a lot
but if you don't get paid
well at all
or hardly enough to make a living,
then it's not a job anymore. It's a hobby, right?
So
one important thing, I think it's
important for people to know what their speed is
in the various tasks that
they undertake so they can
make informed decisions about their life.
Of course
everyone has different priorities
but it's good to be conscious
about it. I want this job
because I like it or because I'm going to
have fun or because it's going to be easy
or because I'm going to earn more.
You need to be conscious about the
decisions that you're making
and what happens to you as a
translator. And translation
speed, I'm not talking even about post
editing, I'm talking about plain translation speed
varies a lot.
And very slow, yeah.
Very slow.
I'm a very slowish translator, yeah.
It varies a lot
with a lot of different
people and it's good to be aware
because you
need to make choices sometimes and you
should make the choices that are
better for you.
I did have another point here.
Ah, yes, okay.
So the other point I wanted to make
is
subtitling is a very creative job
and a lot of people
and I love those translators
who really get into the text,
who are like authors,
who really agonize over
every word that
they translate and produce
a fantastic translation that you really
enjoy reading afterwards
and you say, ah, that was a great
film, I really enjoyed that.
So these people are probably spending
hours thinking about
how to translate a specific
turn of phrase.
It's the kind of feeling
that you have and it has happened to me as well
when I was translating. When you wake up in the middle
of the night or when you're in, I don't know,
when you're doing something completely irrelevant and then
suddenly the translation that
you were trying to think of a specific turn
of phrase comes to you and you interrupt everything
else that you're doing to go and actually
correct a single
word that you have thought
of how to translate better.
Certainly,
this is not good from
a productivity point of view, right?
You probably end up spending a lot
more time in that
translation than you would
otherwise. Now,
but you enjoy it
more. You feel like the author.
Maybe you're not tired at all when you
do that because of the enjoyment factor.
Now, when you are
post-editing, you are no longer
the author. It's some
other person, not persons,
a machine's text in this case.
When you're editing, it's someone else's
translation, so you don't
get so
involved. You're not the author
anymore. It's not your own text, so you can progress
through the text faster.
Instead of thinking for
half an hour, is this translation
really the best one for this sentence?
You probably find a good
translation instantly. You say, yeah, that works.
Move on to the next one.
You end up working faster.
I can imagine
situations where post-editing
can help people
work faster and
produce more work and therefore,
since you're paid by the minute,
end up earning more money
as well.
They have to ask themselves
if they like that.
They have to ask themselves if they like
that or they have to ask themselves if they
just are okay with that
necessarily, because at the end of the day, they have to make
the rent or whatever.
The concept of burnout that you
asked is different.
Everyone is burning out
from what I hear. There's so
much work that I hear
people talking about burnout on the
translation front, but also in language
service providers, in content
owners, everywhere.
This has to do
with the volume of
work that there is. From a
translator's point of view, I've been told
that it has to do also with
the constant
updates, because the turnaround
times are so much
shorter. People don't
always work with final materials or
there are corrections and amendments and
a ton more of emails than I had to
deal with, certainly, when I was
a freelancer.
The constant interruption
to the focused work can also
of course, lead you
to burnout.
You all, someone,
Jeremy or you, Damian, I don't know,
one of you mentioned about
the volume
and a lot of
new people in the profession. There is a lot
of new people in the profession. There's a lot of
language service providers that
are not traditionally media localizers
that are entering our space
because
of the growth that media...
Right now, I'm in the
Lint group for
the European Commission where we
talk about what is happening in the language
industry in general.
Last few months, pretty much, I'm the only
one that is talking about growth
because I'm the only one talking about
the audiovisual sector. More or less, all
the other sectors, there's problems.
Because of this growth,
language service providers,
non-media ones,
want to come in the media space.
Also,
even if they don't want to, they
have to deal with video
for their other clients, for their institutional
clients, for their
technical clients, e-learning
clients. So, there's a
lot of people who were not...
a lot of translators
who did not traditionally work in
subtitling but worked in other translation
fields that are now coming
into subtitling.
There's a lot of new blood.
These people are already accustomed to
using machine translation and translation
memories and all of these
tools. I think they will expect
it, they will want,
they will expect to work
on such tasks. I will not say they will want,
they will expect to work on such tasks.
So, I
don't think there will be
as much resistance in the future
because of the new
blood. The translators
that are graduating university,
I have
friends of mine that are
professors at university teaching
translation, who tell me
not just subtitling,
translation in general, who tell me
that their students don't want to
translate without machine translation.
They find it easier. Of course.
Yes. I mean, it's like
it's how you grow up.
I expect the younger generation
and it would be easier to actually have
a poll on this
with demographics
and age groups
to see. But in those cases, do you think
that they can
become those
translators who are better
than machine translation?
Becoming better
has to do with practice and experience.
If you've just graduated,
clearly you're not better than someone who's
been doing the job for 20 years.
So, they need the right training
and they need to be trained
in my opinion
to translate from scratch
without any aid.
Just as they need
to learn how to spell
without having...
That came up
to my mind, you know, because
I was thinking if you
learn to translate from scratch
to just correct
Google Translate, then how
you're going to be better at some point.
Part of the
training definitely has to do
with that.
I wanted to... I had some notes
here about... There's a question
that I keep asking in conferences.
Yes.
So, I do keep asking
people...
I'm interested in asking translators
how much they use machine translation for
video localization work. Because I chair
panels a lot, I've kind of
developed a habit of asking this
question almost every time.
And I've noticed
an increase in the uptake
from the answers. I've noticed an
increase on the uptake of the technology.
Even among subtitles.
So, I asked this question at the Translating Europe
Forum in November, how much
people use machine translation
in video localization.
20% of the participants answered
that they've never used machine translation
or tried it and don't use it because it doesn't
work for them. 43%
said that they use it
for a lot or the bulk of their work
because it makes them more productive.
And 29% said that they use it
as an aid, even if it doesn't
make them faster yet.
So, the numbers are going
up
favorably towards the use
of MT as time goes by.
And I would really like to know...
I would be...
I would have loved to know what the answers
were with much larger
numbers of people than the ones
that are attending my panel only.
It would be good if
an association like ATRAE could do
some such research
and even
check the demographics
to see in which age group
these preferences
correspond to. I think that would be
very interesting.
I like to compare
also with
how I use technology
being of a certain age
with how
my son uses technology.
So, we were playing this game
on the same game
on my phone. The game gave
you some help
to go to the next level.
I don't remember the
details of it. And I
remember that me
being born in
70s and growing up
as I grew up, it was like
I can do this, I'm the best, I don't
need any help to go to the next level.
I'll just do it because I'm so good.
And this is how I would
play the game. And then my son
would play the game. And what he would do
always is, first of all,
look at the help and use all the help.
It would be impossible
to him to think to play
the game in any other way.
What I was doing for him was
simply crazy.
And I wonder whether this is maybe
how the new generation of translators
will think about machine translation.
That this is my help.
And this is how
what I will use to translate.
And especially as the machine
translation improves,
at some point it will be a no-brainer.
I was going to ask you that.
If you think
it was going to happen
like it did with CAD tools and all that.
Yeah.
I do think it will.
I do think it will.
With time, right?
The thing that I see
and I see it
from your conversation is that
maybe since
subtitlers were never
introduced in their workflows
to CAD tools,
that's why it's more difficult
for them to incorporate
machine translation. Because, like you said,
if I've been translating
engineering text
for 20 years, I've been
through this like 10 years ago.
But first I went through
travels and all that, and then I
went through machine translation
and all that.
Maybe that's why
it's more difficult for us.
Yes, and it also has
to do with age again
because of what you studied at university,
right? When I was studying
translation, there were no CAD tools
to be taught.
Things were just starting
after I graduated.
I never
experienced this at university.
People do experience all
these things at university today,
so they will be more accustomed
to all of these tools irrespective of
which field of translation
they go to.
And, of course,
if translators
are even
they know the tools and even involved
in their development
and in their implementation,
then it will be
a lot easier for them to
take this up. I think that, as I said,
the technology will keep
getting better and better, and
there won't be a point resisting
this. And I already see the first
signs of this because there are
more people that are accepting
of it now.
So,
yes, I do think that this will happen.
But, however, I don't think post-editing
will go away anytime soon, not
for creative fields like
ours. I do expect it
will be there for a while. It will just be
easier, I suppose,
with time. I can only hope that
all the LSPs in the world have
you as a consultant, so
they listen to you.
Guisse?
Yeah, we're talking about
post-editing all the time, and I'm interested in
the pre-editing
as well. So, do you
know of any studies that
research how
technical creation impacts
post-editing work and
machine translation as well?
Yes, that's a great question,
actually.
Right. So,
two ways to answer it. Clearly,
of course, when you use machine translation, you have to
have a template. You have to have a file to
translate, right?
And it
makes sense that
when you have simpler
syntactic structures,
more straightforward ones, subject,
verb, object, the machine translation
will give you better output, right?
That's common sense.
It's only logical.
So, the nice
edited templates
with low reading speeds that we
used to have, you know,
20 years ago, the ones that
I would use to work with,
which I still prefer,
those would lend
themselves better for
machine translation. I actually did this test
when I was working
with statistical machine translation
and the quality of the machine translation
output clearly was better
when you were using such
templates.
So, it is my hope
that because of machine translation,
maybe,
we will return to better
edited templates for
that reason. There is
some research in
simplifying source text
automatically and actually
Netflix has published
something on this, looked into
pre-editing of the source text
and simplifying the source text before
the machine translation in
order to improve the machine
translation output. You know, you
don't necessarily have
to do this automatically. You could just
create a more
edited template
to begin with
instead of creating something more verbatim
and then trying to automatically
simplify it. Unless you
also get the transcript automatically, then
you try to automatically simplify
it, which is something
that people have
worked on. Aptek actually,
which is
a company I consult for that
produces machine translation for subtitling
actually worked
on not simplification
but automatic editing
down of
text. So,
not summarization because summarization
is a lot more
severe condensation of text.
So, the type of
condensation that we have in subtitling,
they have actually worked on that and they
have developed something like that. It's
kind of like machine translating
into the same language.
That's how it works, more or less.
So,
my hope on this is
that
the templates will improve
because of the fact that
these more edited templates give
you better machine translation output.
We all know that
one of the cardinal rules of subtitling is
one sentence, one subtitle.
Yes.
Change the passive
voice into active. Make the syntax
simpler because that's easier
for the viewer to
read faster, to read easier, to
comprehend, right? All of these
things,
when people don't do these things
anymore as much as they used to
in the past,
templates are a lot more verbatim
these days than they used to be.
So, these things actually
work in favor of
machine translation. So, I'm hoping that because
of that we'll revert back to that
I did have a preference for these better
templates because the result is
better for the viewer,
in my opinion anyway.
I do have one final question.
What happens
when the machine, if you
know, when the machine translation doesn't
have a great
background
from previous
corpora?
I'm thinking specifically
for the case of
Argentine
variation because it's
something new.
You do know that we have neutral Spanish
in Latin America, so we do have
a lot of things
translated into neutral, but
only in Mexico
there were local
translations and now it
started to appear in Argentina, so
if we put
this, I mean, that's what I
want to know. What happens
if you say to the machine,
translate this into
not Argentinian because they do have
some text now, but if you want to do
now, I don't know, Colombian Spanish
and there are no
or very few translations.
Do you think
it would take a long time
to do it? What happens?
It can do it anyway?
Okay, I'm not a scientist. I'm not the
most qualified one to answer this, but
my gut feeling
is that
I would estimate that
no, it can't work. You need the
corpora. Basically,
you can't train neural
or statistical machine translation
systems without corpora. If you don't have the corpora,
that's it. You have to create corpora
first. And
there are
better models now,
multilingual models
that go from
many languages into many languages, so there's ways
to actually
improve the machine translation output
even for the lesser sourced
languages, but you
need some corpora. If you don't have
them, then you can't do it.
So the corpora would have to
be created first.
Okay, great. That's
what I thought too, but
I just wanted to check.
So,
I do have like
five more questions, but we are
out of time, so just
want to ask you
what's your view?
We've been talking about this
from the beginning, and I think
that we can end up on a positive
note, but
just give you the mic
to give us your
view on the future of
translators, particularly
in the case of the visual translation,
of course.
Well, I think the profession
will keep changing inevitably,
but it has always changed,
right? As I said,
there's more language service
providers, traditional ones, new players
coming into the market, more
translators outside of
media localization that are coming into the market.
I think there will continue
to be M&A.
Video localization will expand
even more to all
domains, basically, so if you are a
subtitler, you won't just be working
not for entertainment content, but for
any kind of content.
And the fact that
the people that will come into
the market that have been traditionally
using machine translation and translation
memories, I think this will
help machine translation
spread
further.
And then
the biggest
disruption, I think, in our field
will be in dubbing, actually.
I don't know at all about that.
We already see some
dubbing editors that look
a lot like
fancy, enhanced
subtitle editors. I think
they will become widespread
and a means
for streamlining dubbing.
I expect to see dubbing templates
making an emergence
in the near future. I expect
to see many subtitlers
who don't like post-editing,
being converted
into dubbing script adapters.
This is a job that's
quite niche, still.
In many countries, it's
performed by monolingual
editors. There's
a lack of dubbing script
adapters, actually, and this is why there are
initiatives to increase the pool
of translators that work on dubbing
script adaptation. And it's a lot harder
to machine translate scripts
for dubbing, because you have the lip-sync element
and, of course, there are efforts to automate
that, too, but it's a lot harder.
I expect to see
a lot of people going
into that. If you want to be
creative, that's another creative
translation job that is not likely
to be affected by machines
soon. I also
expect to see the use of
synthetic voices expand
in more types of video,
because speech synthesis, I don't know if you've
heard what's going on with speech synthesis,
it's my new favorite language technology,
and it's
improved tremendously.
Now we're able to control the pitch, the
style, the tempo, and also
the emotion
of the voice. So, I
expect that we will have
some people that will have to work
and post-edit
synthetic voices, voice
editing. And who will
these people be? Maybe they will be
translators, right?
Maybe this will be another type of job
that translators
can perform.
Write the translated text
and then voice it
synthetically and listen to it
and make sure that it's pronounced
correctly by post-editing
the synthetic voice.
And, of course, I do believe
that machine translation is here
to stay and that it will improve further.
We've already said that many times.
But, as I said,
I don't think post-editing will
disappear for a while
until I retire,
is what I'd like to say.
And I just turned 50, so that's
quite longer.
If you trust my predictions.
We have 50 more years
to go. Okay.
Not 50 more years. We're going to be working here 100 or
even 200.
I don't know.
But I think, you know, my advice
to translators is,
especially those that don't like post-editing,
is to be agile, to
become more aware of
their own values so they can
go after
the jobs that are right for them.
The ones that machines can't do
yet, like dubbing, script,
adaptation, right?
So, there's a lot of options.
It's not just
post-editing. And also,
you can also do what
I'm doing, which is
advise on what
machines can and cannot do
and help in the development of machines.
Be the one that
tells machines how to
work. Curate
content for machine translation purposes,
for instance. I do believe there
will be a lot of varied tasks
for people to do.
And I do believe also that
translators are inherently
very flexible
people and can
go into different directions.
Just look at
your students, what jobs they
end up doing and how they end up
using their language and translation
skills.
I think the future is bright for translation.
I don't think people need to be
afraid of technology.
I think language
is the last thing that
will be fully automated.
There will be other things that will be
automated first.
Let's hope so.
It's very hard. That's the reason.
That's my
five cents.
Thank you. Thank you so much.
It was an amazing interview and I have
a million things to ask you.
You'll be back.
You'll be back on the show because
we have lots to talk about.
It's always a pleasure to
listen to you.
I really enjoyed
this interview. Thank you so much
for being with us. Thank you
Damián and Guillermo and Blanca.
I really enjoyed it as well. I hope our
audience has enjoyed it. It's been my pleasure.
All the best.
I will look forward to your
questions. Take care.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.

Muy bien. Muy buena la entrevista
con Jota. ¿Qué les pareció, Guille?
Una pasada.
La verdad es que se nota que es
consultora y buena consultora.
Me gustaría que algunas empresas
incluso internacionales
tuvieran esa perspectiva sobre
todas estas tecnologías y cómo
aplicarlas y tal. Vamos, que la
contraten y que la escuchen.
Que la escuchen sobre todo.
Esa es la parte más importante.
Que la escuchen porque
es verdad que es un tema tabú
y cuesta sacarlo.
Todos somos reacios a determinadas cosas
pero hay que matizar.
Una cosa es la tecnología y otra cosa es cómo se está aplicando.
También creo que se han quedado
algunas cosas en el tintero.
Yo sé que a Jota le gusta hablar
de reconocimiento de habla.
También algún artículo suyo que he leído
en su web
hablaba también de lo útil
que era diferenciar las fuentes
de la traducción.
Saber si es traducción automática, si es memoria de traducción,
si es memoria de traducción de dónde salen
las propuestas que te hace el software,
etc.
Y todo esto, a ver,
no podemos meterlo todo en una entrevista pero
significa que habrá que volverla a invitar.
Y ya para acabar, no quiero tampoco
pero voy a
hacer la cuña publicitaria porque
Steve Alice, Cabañés
y yo sacamos un artículo
sobre la post-edición
donde hablamos de mitos y realidades de la post-edición
que si queréis
profundizar entre entrevista con Jota
y entrevista con Jota, pues ahí
tenemos el artículo en la linterna
del traductor.
También tenemos el enlace
en las redes.
Yo creo que es un tema que da
mucho del que hablar.
Me sorprendió mucho el juego que hiciste en Twitter
de que no fue tan fácil
para la gente reconocer
una cosa de la otra. Me acuerdo
que en el HispaTAV sí todo el mundo era
como que reconocía cuál era
la post-edición y cuál
era la traducción de cero, pero me sorprendió un montón.
¿Será porque Twitter vota a cualquier gente?
Ojo, sí. Yo creo que hay varias cosas.
En algunas respuestas era
como... Después entran
al Twitter de Guille y no saben
de qué estamos hablando, pero básicamente
era que Guille puso
los ejemplos de
dos traducciones
e identificar cuál era post-editada
e identificar cuál era traducida de cero
y me sorprendió un montón de que había
ejemplos en los que era 50 y 50
básicamente en la respuesta.
Ojo, que había dos de esos,
de los 11 que puse.
Pero eran muy fácil
inclusive esos dos, ¿no?
Sí, yo creo que hay varias cosas.
Mucha gente, por defecto,
mucha gente comentó que no sabía
qué era la post-edición, que entendían que la post-edición
era traducción automática arreglada
y mejorada y ya está. Entonces
estaban votando lo que ellos creían que ya estaba arreglado.
Entonces, bueno,
también culpa mía por no acabar
de explicarlo bien y también porque no se miraron
lo que era post-edición. Por eso, había mucha
gente que no sabía del tema.
O de votar sin saber de qué estás hablando.
Claro. Y otros que
intuitivamente votaban al revés. O sea,
intuitivamente iban a elegir en vez de darle
la respuesta que había pasado por traducción automática
la que se supone que era
humana porque ellos tiraban hacia lo bueno
por defecto, ¿no? Entonces
eso también afectó.
Y yo creo que, incluso
cuando lo hicimos en la Xpat App,
claro, nos quedamos con la impresión un poco en general
pero no votó todo el mundo. A lo mejor si hubiera votado
todo el mundo, el resultado habría sido
más equilibrado
o no tan claro.
Habría que hacer esa prueba con algún sistema de estos
de votación
en vivo o algo así.
Si no es algo tipo
Twitter, que por cierto, no se me ocurrió
hacerlo en su momento, pero si no es Twitter
es de pago
y no es barato. Pero sí, sí, hay que
probar todas estas cosas. Es que
influye, claro. Yo estaba preparando mi presentación y digo
a ver, me voy a suscribir todo un año a esta plataforma
para hacer una votación un día.
Pero sí, sí.
Y luego también influye creo que
en Latinoamérica hay algunas de esas fórmulas
que aún se usan.
O sea, el Cielos en España
nos parece una barbaridad, pero
en el español neutro, a lo mejor
la gente lo usa
y lo lee y no le llama la atención
en absoluto. Entonces, claro, todo esto
influye en la percepción.
Sí, sí, sin lugar a dudas. Blanca,
¿qué te parece la entrevista?
Yo estoy muy contenta porque
no lo he confesado
en la entrevista porque me daba vergüencita,
¿no? Pero yo era
muy fan de Jota durante
el doctorado y de hecho
el libro que tiene
del 2009, Reduction Levels in Subtitling,
para mí fue
importantísimo
porque
como mi tesis es del Pleistoceno, yo trabajé
con series en DVD.
Entonces, ella justamente
era una de las autoras que había hablado
de Subtitling for the DVD Industry
en
la tesis y en
ese libro y, bueno, es una
persona a la que he seguido y estoy muy contenta
de haberle
podido entrevistar en
el podcast. Y luego me ha parecido
muy inspiradora esa
casi última frase que nos ha dicho, ¿no?,
de que las lenguas
parece que van a ser lo último que se pueda
automatizar del todo. Creo que que lo diga
una persona como ella
da mucha esperanza, la verdad.
A mí la verdad que sí, es una
persona que la vengo, no tanto como vos,
pero la vengo siguiendo hace unos cuantos años
y he tenido la suerte de trabajar
para ella y con ella en varios proyectos
así que realmente
tenía muchas ganas de hacer una entrevista, pero como dice
Guille, es una persona
para la cual tendríamos que usar
dos o tres programas porque, sinceramente,
ha quedado muchas cosas
a mí. Una de las cosas que
publicó recientemente en su sitio web tenía que ver
con una encuesta que hizo entre
la preferencia entre
software en la nube o software en el
escritorio y
es un tema muy interesante también
para tener en cuenta
porque cambia mucho con las modalidades de trabajo
pero también está bueno que haga
estas cosas que es
escuchar a la gente, ¿no?, por ejemplo, ver
qué opina el profesional
en este ámbito y en otros ámbitos, en un congreso
o lo que sea y después ella tiene la suerte
de poder llevarlo a las empresas,
¿no?, que eso también es
algo que está muy bien porque muchas veces
se hacen estudios o análisis
pero después eso no llega
a quienes toman las decisiones sobre
procesos y ese tipo de cosas, así que
me gusta que pueda mezclar
esas dos cosas. Bueno, muy bien,
hasta aquí llegamos entonces con el episodio
de hoy. Recuerden visitar
a nuestro patrocinador, a Ooona,
en el sitio web ooona.net y registrarse
para las cuatro semanas de prueba.
Recuerden también que pueden encontrar
toda la información del podcast en nuestro sitio
web www.ensincroniapodcast.com
y también en nuestro canal de YouTube donde van a encontrar
la versión
con subtítulos del capítulo
y que también pueden suscribirse
a la gacetilla del programa, esto
es algo nuevo que tenemos en nuestra
gacetilla, espero que se suscriban así
y reciban información de todos los episodios.
El enlace para la suscripción a la gacetilla
lo van a encontrar en nuestros perfiles
de Instagram, así que se meten ahí
y cargan su información, nos cuentan
qué episodio les gustó más,
qué temas quieren que tratemos
en mayor profundidad o más
todavía en el podcast, así que
también conocemos un poco más
a quienes están del otro lado escuchándonos
y bueno, sobre todo para que no se pierdan
después ninguna de las novedades de los
episodios y lo pueden
encontrar entonces ahí en nuestras cuentas de Instagram.
¡Hasta la próxima!

People on this episode