Pages in topic:   < [1 2 3 4]
Your thoughts on the quality of "premium" Google Translate?
Thread poster: Samuel Murray
Siegfried Armbruster
Siegfried Armbruster  Identity Verified
Germany
Local time: 23:14
English to German
+ ...
In memoriam
complete crap Dec 21, 2016

Samuel Murray wrote:
I'm sure the first sentence's "premium" translation comes straight out of a TM and that there is nothing neural about it.


It does not come from a TM, the meaning is completly different from the German original.

My English is lousy, but the German means something along the lines

"Those who report the ones who have stolen the apples, will be rewarded".


 
Lingua 5B
Lingua 5B  Identity Verified
Bosnia and Herzegovina
Local time: 23:14
Member (2009)
English to Croatian
+ ...
Isn't this true for standard GT, too? Dec 22, 2016

Samuel Murray wrote:

Kay-Viktor Stegemann wrote:
The sample sentences you chose are rather extraordinary, long-winded literary sentences.


Yes, but that was deliberate, since one member posited that while the premium service does fairly well on short sentences, it may not do well on longer, more complicated sentences.


So, there's no difference between standard and premium then? They both have issues when sentence starts expanding beyond the most basic structure that has high rate of repetitiveness from the feeds?


 
Michael Wetzel
Michael Wetzel  Identity Verified
Germany
Local time: 23:14
German to English
Sorry to be picky Dec 22, 2016

While my source texts may tend to be closer to Nabakov than to Wikipedia, I think the new source sentences are actually unfairly difficult. Here is the first paragraph of the first article dealing with a business topic in today's "Berliner Zeitung":

"Die kleine Ausstellungsrunde von Gründern im Lichthof der TU Berlin an der Straße des 17. Juni machte deutlich, wie groß die Bandbreite der Start-ups ist, die aus den Hochschulen der Hauptstadtregion hervorgehen. Die Calimoto GmbH pr
... See more
While my source texts may tend to be closer to Nabakov than to Wikipedia, I think the new source sentences are actually unfairly difficult. Here is the first paragraph of the first article dealing with a business topic in today's "Berliner Zeitung":

"Die kleine Ausstellungsrunde von Gründern im Lichthof der TU Berlin an der Straße des 17. Juni machte deutlich, wie groß die Bandbreite der Start-ups ist, die aus den Hochschulen der Hauptstadtregion hervorgehen. Die Calimoto GmbH präsentierte hier am Montag eine Navigations-App speziell für Motorradfahrer. Die gefahrenen Strecken werden dabei anhand des Höhenprofils, der Geschwindigkeit und anderer Faktoren analysiert und bewertet." – Quelle: http://www.berliner-zeitung.de/25339688 ©2016

I then went on from there and took the first paragraph of an article dealing with a cultural topic (happens to contain some quotations that will probably be challenging):

"Im Streit um eine bessere Busanbindung der KZ-Gedenkstätte Sachsenhausen fühlt sich der Leiter Günter Morsch vom Landkreis Oberhavel und von der Verkehrsgesellschaft brüskiert. „In einem Gespräch, bei dem es um eine Lösung für unsere vielen tausend Besucher gehen sollte, wurde uns erklärt, dass man zunächst einmal den Bedarf ermitteln wolle“, sagte Morsch am Mittwoch. „Dabei kann man jeden Tag sehen, dass große Besuchergruppen zu Fuß vom Bahnhof zur Gedenkstätte marschieren müssen, weil der Bus 804 an Werktagen nur einmal stündlich und an den Wochenenden sogar nur alle zwei Stunden fährt.“ Morsch fordert eine Erhöhung des Takts für die Linie." – Quelle: http://www.berliner-zeitung.de/25356456 ©2016

I used a Google search to randomly select a German scholarly journal in the field of art. The article was not randomly selected: Looking at the table of contents, Kentridge immediately seemed appropriate. This is also the third paragraph of the article, because the first two are of a more introductory nature that does not seem representative. Additionally, there is an error in the original that I couldn't copy and paste here: "Ubu and the Truth Commission lässt er König" is all written in italics, but "lässt er König" is obviously not part of the title and should not be italicized. It would be interesting to see what GT would do with that.

"Beinah all seine historischen Exkursionen, am Kap beginnend, führen dorthin wieder zurück. Wie eine Matrix legt er die verdrängte oder geleugnete Geschichte über die Gegenwart. In seiner 1997 uraufgeführten Tragikomödie Ubu and the Truth Commission lässt er König Ubu, Alfred Jarrys Despoten, in der Wahrheitskommission zur Aufklärung der Verbrechen der Apartheid auftreten. Dieses Schauspiel handelt von der Vergeblichkeit kollektiver Selbstvergewisserung, wird doch die Gerechtigkeit am Ende auf dem Altar der Wahrheit geopfert. Die Mörder waschen sich durch Geständnisse rein und entkommen ungestraft. Und so leben Opfer und Täter, weiße Herren und schwarze Knechte, so unversöhnt und ungleich wie je zuvor, nebeneinander her. Für Kentridge hat es in diesem Land zwar einen fundamentalen Wandel gegeben. Ihm erscheint es jedoch gelegentlich so, als sei alles unverändert geblieben. „Das neue Südafrika wirkt wie eine Übermalung des alten Südafrika“, so Kentridge." (Heinz-Norbert Jocks, "DOPPEL-DADA: Betrachtungen des ewig Gestrigen im Jetzt; Ein Gesprach von Heinz-Norbert Jocks [mit William Kentridge]," KUNSTFORUM International 243 (2016), p. 164)

As I see it, the main problem is that it is genuinely exhausting to read even 1000 or 2000 words of poorly translated (or otherwise very poorly written) text. I can still remember my first presentation after I began studying at a German university: I could see my listeners literally nodding off, in spite of their heroic efforts to politely and attentively listen to my German.
Stylistic problems and awkwardness may be merely annoying at the level of a sentence or a short paragraph, but they can quickly render a longer text unreadable. I would guess that these examples will also cause Premium GT to have significant problems at the level of content.

[Edited at 2016-12-22 08:50 GMT]
Collapse


 
Samuel Murray
Samuel Murray  Identity Verified
Netherlands
Local time: 23:14
Member (2006)
English to Afrikaans
+ ...
TOPIC STARTER
@Michael Dec 22, 2016

Michael Wetzel wrote:
Sorry to be picky


Picky is good in this thread.

While my source texts may tend to be closer to Nabakov than to Wikipedia, I think the new source sentences are actually unfairly difficult. Here is the first paragraph of the first article in today's "Berliner Zeitung" dealing with a business topic...


german difficult 1

I then went on from there and took the first paragraph of an article dealing with a cultural topic (happens to contain some quotations that will probably be challenging)...


german difficult 2

I then used a Google search to randomly selected a German scholarly journal in the field of art. Looking at the table of contents, Kentridge immediately seemed appropriate, so the article was not randomly selected. This is also the third paragraph of the article, because the first two were of a more introductory nature that is not representative...


german difficult 3

There is also an error in the original that I couldn't copy and paste here: "Ubu and the Truth Commission lässt er König" is all written in italics, but "lässt er König" should not be captialized. It would be interesting to see what GT would do with that.


Unfortunately my little translate script can handle only plain text. I think formatted text is dealt with by Google in HTML format, so Google essentially deals with tags. This raises an interesting point: how the machine would deal with formatting that is unnatural.


 
Samuel Murray
Samuel Murray  Identity Verified
Netherlands
Local time: 23:14
Member (2006)
English to Afrikaans
+ ...
TOPIC STARTER
Formatted text, oh dear, oh dear... Dec 22, 2016

Samuel Murray wrote:
Unfortunately my little translate script can handle only plain text. I think formatted text is dealt with by Google in HTML format, so Google essentially deals with tags. This raises an interesting point: how the machine would deal with formatting that is unnatural.


german with tags

I'll have to ask the Google people why Premium can't handle tags (or: how I can tell Premium to handle it).


 
Michael Wetzel
Michael Wetzel  Identity Verified
Germany
Local time: 23:14
German to English
extremely bad Dec 22, 2016

I don't have the time to go into great detail, but GT has now done even worse than I would have expected with these articles. Standard GT actually seems better in the case of the first example.

Premium GT has once again had difficulties with proper nouns (instances in first example text):
Premium wrongly treats "Lichthof" as a proper noun (i.e., it leaves it untranslated).
Premium wrongly treats "Straße des 17. Juni" as a common noun (i.e., it translates it).

... See more
I don't have the time to go into great detail, but GT has now done even worse than I would have expected with these articles. Standard GT actually seems better in the case of the first example.

Premium GT has once again had difficulties with proper nouns (instances in first example text):
Premium wrongly treats "Lichthof" as a proper noun (i.e., it leaves it untranslated).
Premium wrongly treats "Straße des 17. Juni" as a common noun (i.e., it translates it).

Content:
"Bandbreite" (scope, range, etc.) disappears in the Premium GT translation, completely distorting the meaning of the first sentence. Worse yet, it presents a plausible translation, which would presumably be overlooked by most post-editors.
"gefahrenen Strecken": Premium's "distances traveled" is a possible translation, but clearly wrong in this context, and it makes nonsense out of the third sentence. Standard correctly selects "routes taken" ... Standard's "velocity" is a very poor choice of words relative to Premium's "speed," but is not immediately confusing.

One interesting thing is that Premium GT has dealt more or less sensibly with the "zwar" in the third-to-last and next-to-last sentences of the Kentridge text. This construction is often spread across two sentences in German, but in English it almost has to be either combined into one sentence or the qualifier has to be moved to the second sentence. That is the kind of thing that Premium GT ought to better at, and it is better here.

On the whole, however, the results here are genuinely very bad. Many of the sentences are reduced to gibberish, there are consistently major problems at the level of content and only a handful of sentences are even theoretically adequate. Now I'm interested in why these results are so much worse.

And does Google offer a way to choose between US and UK English?
Collapse


 
Samuel Murray
Samuel Murray  Identity Verified
Netherlands
Local time: 23:14
Member (2006)
English to Afrikaans
+ ...
TOPIC STARTER
@Michael Dec 22, 2016

Michael Wetzel wrote:
Worse yet, it presents a plausible translation, which would presumably be overlooked by most post-editors.


This is probably the biggest danger of the premium translation service. For me, who doesn't speak German, all three examples of the standard translation felt like a machine translation, but the premium translations felt like texts that made sense. In fact, the premium translations make perfect sense to me, but that's not relevant to us as translators if the "sense" is the wrong sense.

This also shows why one can't evaluate the quality of a machine translation unless one can speak the source language. It would seem that the premium service's language quality is very good, but the translation quality is either just as bad or not much better/worse than the standard service. The danger that that represents is that readers will be less vigilant about the fact that the translation may not say what the original text meant.


 
Michael Wetzel
Michael Wetzel  Identity Verified
Germany
Local time: 23:14
German to English
first sentence of the second example interesting for that issue Dec 22, 2016

Looking only at the English translations of the first sentence, it would seem very reasonable to suppose that Günter Morsch is some sort of official for the county/district (Landkreis Oberhavelland) and the transportation company and that Google MT simply had difficulty finding the right phrasing. However, Morsch is actually the director of the Sachsenhausen museum.

I also don't know what someone who can't speak German would do with "Morsch calls for an increase in/of the clock for
... See more
Looking only at the English translations of the first sentence, it would seem very reasonable to suppose that Günter Morsch is some sort of official for the county/district (Landkreis Oberhavelland) and the transportation company and that Google MT simply had difficulty finding the right phrasing. However, Morsch is actually the director of the Sachsenhausen museum.

I also don't know what someone who can't speak German would do with "Morsch calls for an increase in/of the clock for the line." The sentence means that busses need to stop there at shorter intervals, and I doubt that existing context will be enough for most post-editors to very quickly and reliably make these kinds of leaps from machine translations to intended meanings.

And judging from the prior examples, I would guess that the improvements to language quality (or plausibility) might be dramatic for English > German, but again, that's only one small part of the real issue.

PS: It's also interesting that both GTs chose against having the visitors "march" to the camp, as they do in the German source text. It's an unusual choice of words in German, but obviously deliberate and significant. If it weren't a quotation and a human translator decided to get rid of it as a pun that is somehow more offensive in English than in German, it could arguably be an appropriate decision. However, I'm curious why GT opts against it.
Collapse


 
Pages in topic:   < [1 2 3 4]


To report site rules violations or get help, contact a site moderator:


You can also contact site staff by submitting a support request »

Your thoughts on the quality of "premium" Google Translate?






Trados Business Manager Lite
Create customer quotes and invoices from within Trados Studio

Trados Business Manager Lite helps to simplify and speed up some of the daily tasks, such as invoicing and reporting, associated with running your freelance translation business.

More info »
TM-Town
Manage your TMs and Terms ... and boost your translation business

Are you ready for something fresh in the industry? TM-Town is a unique new site for you -- the freelance translator -- to store, manage and share translation memories (TMs) and glossaries...and potentially meet new clients on the basis of your prior work.

More info »