Communications of the ACM
Google Translate Does Not Understand the Content of the Texts
How well do translation programs translate? Some selected examples are intended to give a first impression of the quality of translations. Of course, this contribution does not claim to be a comprehensive test. Today's translation programs are based on neural networks and large data sets. In addition, very powerful computers are available. The tests were mainly carried out with Google Translate and DeepL. The Microsoft and Amazon programs have similar weaknesses. Once again, there is talk of a breakthrough.
Source text
Es gibt hauptsächlich drei Arten analoger Rechenschieber: Rechenstäbe, Rechenscheiben und Rechenwalzen.
Google Translate
There are mainly three types of analogue slide rules: rulers, calculating discs and raking rollers.
DeepL
There are mainly three types of analog slide rules: slide bars, slide disks and rake rollers.
Source text
Zu den wichtigsten Formen der Rechenschieber gehören der Rechenstab, die Rechenscheibe, die Rechenwalze und die Rechenuhr.
Google Translate
Among the most important forms of slide rule are the ruler, the ruler, the rake and the ruler.
DeepL
The most important forms of slide rules are the slide bar, the disc rule, the screen roller and the screen clock.
Correct technical terms
Rechenscheibe circular slide rule
Rechenschieber slide rule
Rechenstab linear slide rule, straight slide rule
Rechenuhr pocket watch slide rule
Rechenwalze cylindrical slide rule
Source text
Gottfried Wilhelm Leibniz gilt als Erfinder des Sprossenrads und der Staffelwalze.
Google Translate
Gottfried Wilhelm Leibniz is regarded as the inventor of the sprocket wheel and the squadron roller.
DeepL
Gottfried Wilhelm Leibniz is regarded as the inventor of the sprocket wheel and the relay roller.
Correct technical terms
Sprossenrad pinwheel
Staffelwalze stepped drum
Source text
Wir sind auf der Suche nach einem verschollenen römischen Handabkus aus Elfenbein.
Google Translate
We are looking for a lost Roman ivory handcuff.
DeepL
We're looking for a lost ivory Roman handabku.
Source text
Im Unterschied zum chinesischen Rechenrahmen hat der römische Handabakus keine Stangen mit Kugeln, sondern Rillen mit Knöpfen.
Google Translate
Unlike the Chinese framework, the Roman Handabakus has no rods with balls, but grooves with buttons.
Deep L
In contrast to the Chinese calculation frame, the Roman hand tobacco has no rods with balls, but grooves with buttons.
Correct technical terms
Handabakus hand abacus
Rechenrahmen abacus
Source text
Im Altertum nutzten die Griechen für Berechnungen die Rechentafel. Die Römer verwendeten hingegen einen tragbaren Handabakus mit Rillen und Knöpfen. Das Mittelalter kannte Rechenbretter, Rechentische und Rechentücher mit Linien und Rechenmünzen. Der Zählrahmen mit Stäben und Kugeln war in China, Japan, Russland jahrhundertelang sehr beliebt.
Google Translate
In antiquity, the Greeks used the calculator for calculations. The Romans, by contrast, used a portable handabaku with grooves and buttons. The Middle Ages knew calculating boards, computer desks and books with lines and computing coins. The counting frame with bars and balls was very popular in China, Japan, Russia for centuries.
DeepL
In ancient times the Greeks used the calculation board for calculations. The Romans, on the other hand, used a portable hand tobacco with grooves and buttons. The Middle Ages knew calculation boards, calculation tables and calculation cloths with lines and calculation coins. The counting frame with rods and balls was very popular in China, Japan and Russia for centuries.
Correct technical terms
Handabakus hand abacus
Kugel bead
Rechenbrett counting board, calculating board, abacus
Rechenmünze counter, token, casting counter
Rechentafel counting board, calculating board
Rechentisch reckoning table, counting table, table abacus
Rechentuch reckoning cloth, counting cloth
Source text
Ursprünglich waren Computer Menschen, die mit mechanischen Tischrechenmaschinen Berechnungen durchführten. Später kamen elektromechanische Relaisrechner und elektronische Röhrenrechner auf, die schließlich von Transistormaschinen abgelöst wurden.
Google Translate
Originally computers were people who performed calculations with mechanical desktop calculators. Later came electromechanical relay computers and electronic tube computers, which were eventually replaced by transistor machines.
DeepL
Originally computers were people who performed calculations with mechanical table calculators. Later, electromechanical relay computers and electronic tube computers came into being, which were eventually replaced by transistor machines.
Source text
Einer der frühesten Erfinder von Automaten war Heron von Alexandria (1. Jh. n.Chr.). Leonardo da Vinci, der vor 500 Jahren starb, entwarf äußerst komplexe Roboter, u.a. einen selber fahrenden Wagen, einen mechanischen Löwen und einen mechanischen Krieger. Im 18. Jahrhundert gab es großartige Figurenautomaten (z.B. Zeichen-, Schreib- und Musikautomaten), die heute noch voll funktionsfähig und z.T. auch programmierbar sind.
Google Translate
One of the earliest inventors of automatons was Heron of Alexandria (1st century AD). Leonardo da Vinci, who died 500 years ago, designed extremely complex robots, ia. a self-driving car, a mechanical lion and a mechanical warrior. In the 18th century, there were great figurative machines (such as drawing, writing and jukeboxes), which are still fully functional and some are still in use today, are also programmable.
DeepL
One of the earliest inventors of automatons was Heron of Alexandria (1st century AD). Leonardo da Vinci, who died 500 years ago, designed extremely complex robots, including a self-propelled carriage, a mechanical lion and a mechanical warrior. In the 18th century there were great figure automatons (e.g. drawing, writing and music automatons), which are still fully functional today and partly also programmable.
Correct technical terms
Figurenautomat automaton figure, automated figure
Musikautomat musical automaton, music automaton
Description of an astronomical clock from the 14th century
Source text (French, International clock museum, La Chaux-de-Fonds, Switzerland)
Astrarium (reconstruction) – Giovanni Dondi
Le Padouan Giovanni Dondi construisit entre 1365 et 1380 une horloge exceptionnelle reproduisant le mouvement des planètes dans le cosmos. Un astrarium est une matérialisation mécanique du système géocentrique conçu par Ptolémée au IIe siècle ap. J.-C et repris par la majorité des astronomes du Moyen Age. Elle disparut malheureusement prématurément au XVIe siècle. Luigi Pippa, horloger milanais, put la reconstruire six siècles après sa disparition grâce à sa description richement illustrée.
Google Translate (English)
Astrarium (reconstruction) - Giovanni Dondi
The Paduan Giovanni Dondi built between 1365 and 1380 an exceptional clock reproducing the movement of the planets in the cosmos. An astrarium is a mechanical materialization of the geocentric system designed by Ptolemy in the second century AD. J.-C and taken by the majority of astronomers of the Middle Ages. It disappeared unfortunately prematurely in the sixteenth century. Luigi Pippa, Milanese clockmaker, was able to rebuild it six centuries after his death thanks to his richly illustrated description.
DeepL (English)
Astrarium (reconstruction) - Giovanni Dondi
Between 1365 and 1380, Padua's Giovanni Dondi built an exceptional clock that reproduced the movement of planets in the cosmos. An astrarium is a mechanical materialization of the geocentric system designed by Ptolemy in the 2nd century AD and adopted by the majority of medieval astronomers. Unfortunately, it disappeared early in the 16th century. Luigi Pippa, a Milanese watchmaker, was able to reconstruct it six centuries after its disappearance thanks to its richly illustrated description.
Google Translate (Deutsch)
Astrarium (Wiederaufbau) - Giovanni Dondi
Das Paduan Giovanni Dondi baute zwischen 1365 und 1380 eine außergewöhnliche Uhr, die die Bewegung der Planeten im Kosmos wiedergibt. Ein Astrarium ist eine mechanische Materialisierung des geozentrischen Systems, das von Ptolemäus im zweiten Jahrhundert nach Christus entworfen wurde. J.-C und von der Mehrheit der Astronomen des Mittelalters übernommen. Im 16. Jahrhundert verschwand sie leider vorzeitig. Der Mailänder Uhrmacher Luigi Pippa konnte ihn dank seiner illustrierten Beschreibung sechs Jahrhunderte nach seinem Tod wieder aufbauen.
DeepL (Deutsch)
Astrarium (Rekonstruktion) - Giovanni Dondi
Zwischen 1365 und 1380 baute Paduas Giovanni Dondi eine außergewöhnliche Uhr, die das Uhrwerk der Planeten im Kosmos reproduzierte. Ein Astrarium ist eine mechanische Materialisierung des geozentrischen Systems, das von Ptolemäus im 2. Jahrhundert n. Chr. entworfen und von der Mehrheit der mittelalterlichen Astronomen übernommen wurde. Leider verschwand es Anfang des 16. Jahrhunderts. Luigi Pippa, ein Mailänder Uhrmacher, konnte es sechs Jahrhunderte nach seinem Verschwinden dank seiner reich illustrierten Beschreibung wieder aufbauen.
Google Translate seems to be learning. The first translation attempt led to a completely useless, faulty mixture of French and German words:
Astrarium (Wiederaufbau) - Giovanni Dondi
Le Padouan Giovanni Dondi konstruisit entre 1365 und 1380 Eine Ausnahme, die das Mouvement des Planètes dans le cosmos reproduziert. Ein Astrologiewerk für eine Matérialisierung des Zusammenlebens mit dem Ptolémée au IIe siècle ap. J.-C und Vertreter der Astronomen von Moyen Age. Elle disparut malheureusement prématurément von XVIe siècle. Luigi pippa, horloger milanais, legte sechs rekonstruktionen vor, die eine disparition darstellen.In der ersten Reihe von Besichtigungen finden Sie hier eine unendliche Weite. Die Bewegung ist eine der schönsten Sehenswürdigkeiten der Stadt, die sich auf einer Wanderung befindet, und ist ein beliebtes Ausflugsziel.
Ambiguity
Many words are ambiguous:
In diesem Schloss gibt es ein schönes Museum. Der Schlüssel steckt im Schloss. Sie schloss die Türe. Im Park gibt es zum Ausruhen eine Bank. In der Bank kann man Geld wechseln.
Google Translate
There is a nice museum in this castle. The key is in the lock. She closed the door. In the park there is a bench to rest. In the bank you can change money.
DeepL
There's a nice museum in this castle. The key is in the lock. She locked the door. In the park there is a bench for resting. You can change money in the bank.
An example of a reverse translation
Google Translate
Source text
The spirit is willing but the flesh is weak.
Der Geist ist willig, aber das Fleisch ist schwach.
L'esprit est prêt, mais la chair est faible.
Lo spirito è disponibile ma la carne è debole.
El espíritu está dispuesto, pero la carne es débil.
Back translations
The mind is willing, but the flesh is weak (G-E).
The mind is ready, but the flesh is weak (F-E).
The spirit is available but the flesh is weak (I-E).
The spirit is willing, but the flesh is weak (S-E)
For comparison
In 1975, I organized a demonstration of the Systran automatic language translation system (Latsec, Inc.). It took place in the computer center of the Institute of Computer Science of the University of Zurich. We used the new IBM 370-155 mainframe computer (core memory of 1536 kilobytes). We translated complicated technical texts from Russian into English. The University of Grenoble provided us with machine-readable Russian documents (about 100 000 words, i.e. 300 A-4 pages). One third of this was used. The 30,000 Russian words were translated in six minutes. As expected, the results were not perfect. However, they provided information about the content of the texts. At that time, similar events took place in Bonn and Luxembourg. The Commission of the European Union used Systran for its language services for a longer period of time. The presentation in Zurich is extensively documented in:
Herbert Bruderer: Handbuch der maschinellen und maschinenunterstützten Sprachübersetzung. Automatische Übersetzung natürlicher Sprachen und mehrsprachige Terminologiedatenbanken. K. G. Saur Verlag KG, München, New York, London, Paris 1978 / Walter de Gruyter, Berlin, 880 pages (Handbook of machine translation and machine-assisted translation. Automatic translation of natural languages and multilingual terminology databases. English translation by the Commission of the European Union, Brussels).
Concluding remarks
The examples show that the quality of machine translations vary between fairly good and useless. It depends, among other things, on the language pair, the subject area and the available data set and its quality. In Wikipedia, there are also large differences between the different language versions.
Further tests have shown that automatic translations are often inconsistent and sometimes even nonsensical. Sometimes even words are missing. The decisions are sometimes difficult to understand. In many cases the results are "good enough" for some applications. For non-native translators, the programs can be a valuable help. For language pairs with very large data sets, machine translation may achieve the quality of a mediocre human translator. The raw translations usually have to be edited manually. In the case of poor raw translations, the effort required for an improvement can be higher than for a manual retranslation.
A child does not need millions of images to recognize a cat. A human does not need millions of examples to translate a text correctly. This is one of the weaknesses of this approach, which is reminiscent of brute force.
Despite significant progress, it is obvious that Google translate & Co. are not really intelligent and do not understand the meaning of the texts. The programs, however, do not know that they translate.
Herbert Bruderer is a retired lecturer in didactics of computer science at ETH Zürich. More recently, he has been an historian of technology. [email protected], [email protected]
No entries found