Category Archives: Historical

Anne, classical singing, and Google Translate

I Google-translate old Italian arias and not-quite-as-old German lieder.

For the past 8 years I’ve been taking lessons in classical singing. Part of being able to perform a song is to first ‘get inside’ it – understand what kind of message is involved and what the emotion and atmosphere surrounding it are like. Since the songs are often in languages I know only a little (German) or not at all (old Italian), I’ve found it helpful to start the process by putting the lyrics into Google Translate and translating them.

aria lyrics

When we start work with a new song, my teacher often plays the melody for me and sometimes also tells me a little bit about what it’s about. After that I try to find out more. First I check to see if I can find some kind of translation. Many of the songs have been used in teaching classical singing for a long time, so once in a while there is some kind of translation online. However, they can be surprisingly hard to find.

At first I tried looking up everything word-by-word in a dictionary, but that was slow and not that successful. You would have to know the modern equivalents of words in an old language to be able to look them up, and I couldn’t do that. When I put the whole thing into Google Translate, there were words here and there that didn’t get translated, but I still got a general idea of what it was saying.

I’ve used this method since then with new songs. If I can capture some key words – maybe a word is repeated throughout the song, or it’s emphasized somehow or used at a high point in the melody – and get an understanding of those, that is a good start, even if I don’t get a full understanding of the entire song.

I also often translate into both English and my native language of Finnish. Doing it into 2 languages is a good method – you might get something in one of the languages that the other didn’t succeed at, so between the 2 you end up covering mostly everything. And given the options of no understanding or some understanding, I’ll take the latter.

Also, of course Google Translate does not produce a beautiful narrative like the original, but I don’t expect it to. The original is old poetry, after all, and that is difficult even for human translators. What I need is to understand a bit of what it’s about and capture some keywords.

Once a pretty funny thing happened where I found out I had completely misunderstood the translated lyrics of a song I was working on. It was in German and there was just one word I didn’t understand so I put it into Google Translate. Something about the translation of that one word gave me the impression that the speaker in the poem was a woman who was just content to be dancing around with other country folk. It turned out that it was actually about a man who was somewhat lewd and disgusting!

Finally one day I was singing it with my teacher and a pianist. We got to talking about what it’s about, what the feeling and atmosphere should be during each part of the song. I told my teacher what I thought it was and she corrected me. Whereas my earlier rendering had been slightly solemn and serious, after the explanation I understood the humor and playfulness of the poem and was able to match the slightly mischievous tone much better. It became much more fun.

Dr. C. Koby, Physicist*

Oak Ridge National Laboratory, Tennessee, USA
October 15, 1971

Part of my work as a physicist is to keep abreast of research being carried out in my field. I try to follow activities in all parts of the world, but I am particularly interested in the research being conducted in the Soviet Union. I would guess that something like 15% of the articles and books I need to read in my field are written in Russian. I have no competence in Russian so have to rely on translations to access those materials.

A very positive development in the past few years is a service that my laboratory offers which machine translates articles from Russian to English. They don’t have anyone correct the articles,  you get what the machine puts out. I have personally used the service 10 or so times over the past 3½ years.

I have one simple reason for using the MT service: it is fast. After a request, I can expect to get a machine-translated article back in 1 month, much more quickly than the 9-10 weeks some colleagues report waiting for articles translated by humans.

More time and effort is required to understand machine-translated texts as compared to those written in English – I’d say about double the time. One reason for that is that formulas and diagrams from the original articles do not get translated, so I usually have both the machine-translated text and the original Russian article, and I go back and forth between them. In general, however, I’d say that the quality of the translations is acceptable. In a funny way, over time you get used to the style. You find yourself mentally correcting awkward passages.

I think it’s important to bring up the fact that the machine translation is understandable to myself and my colleagues mostly because we are familiar with the contexts being discussed. We know our own fields well and that allows us to accommodate texts that are not perfect. It is also important that technical terms are translated correctly or at least understandably, and the majority of the time, they are. The most common problem is words that are left untranslated. I don’t know why that happens but it does, and it can affect understanding.

I’ve been asked if it would be possible for someone to completely misunderstand the machine translations and arrive at the opposite meaning from the original. Yes, sometimes the meaning is a bit distorted, but since I know the context of what I’m reading so well, I do not think that it is possible to arrive at the opposite meaning. I have never heard of it happening anyway.

Overall, I’d say that I’m happy to use the service. It allows me to read things I would normally have to wait far longer to get access to. I would, and actually have been known to, recommend the use of machine translation to my colleagues.


*       Instead of being about 1 real person, the persona in this story is a composite that is based on a group of 58 real people who were surveyed by Bozena Henisz-Dostert in the early 1970s (see reference below). The group consisted of some of the first users of one of the first machine translation systems: the Georgetown Automated Translation system, developed in the 1950s and early 1960s at Georgetown University. Although its development was discontinued in 1964 due to lack of financial support, the Georgetown system continued to be used, virtually in its 1964 state, for a decade or so by 2 groups of scientists in Oak Ridge, Tennessee and Ispra, Italy.

The persona in the story was compiled in a straightforward way. I used 23 of the survey’s 51 questions. For each question, I took the most popular answer, or the mean or average of the figures given in answers, to describe Dr. Koby. The name ‘Koby’ came from an internet name generator and does not reflect any one true person.

The story is based entirely on information from Bozena Henisz-Dostert’s article ‘Users’ evaluation of machine translation’ in the book Machine Translation by Henisz-Dostert et al. 1979.