This is an essay I wrote in February 2017. It was published on a private blog site I used to maintain.
Ever since reading the NY times article about the amazing improvements in Google Translate, I’ve been thinking a lot more recently about artificial intelligence and machine translation.
I’ve also been thinking a lot more about Ludwig Wittgenstein, as evidenced in my last writing [not published yet], and the combination of these interests makes for today’s entry.
Wittgenstein famously summarized his early philosophical work in the short book Tractatus Logico-Philosophicus. The book has an introduction by Bertrand Russell, and a short, but very personal, preface by Wittgenstein. Earlier today I was reading proposition two in the Tractatus as conveyed in two different English translations in an attempt to better understand what Ludwig had been intending to say.
The Tractatus was originally written in German and published in 1921. The first English translation came out in 1922 and was produced by C.K. Ogden and F.P. Ramsey. The English copy of the Tractatus that I own dates to 1961, and the translation there was executed by D.F. Pears and B.F. McGuinness.
The last paragraph of Wittgenstein’s preface is a very powerful and concise bit of writing. In the following I’ve collected six transcriptions of what he wrote: the original German words, the work of the two different English translations, a Spanish translation, and two translations by Google Translate, one from the German and one from the Spanish. [The 2021 postscript provides two more relevant translations.]
Original German (1921), Ludwig Wittgenstein (54 words)
Dagegen scheint mir die Wahrheit der hier mitgeteilten Gedanken unantastbar und definitiv. Ich bin also der Meinung, die Probleme im Wesentlichen endgültig gelöst zu haben. Und wenn ich mich hierin nicht irre, so besteht nun der Wert dieser Arbeit zweitens darin, dass sie zeigt, wie wenig danmit getan ist, dass diese Probleme gelöst sind.
Original English translation (1922), Ogden/Ramsey (65 words)
On the other hand the truth of the thoughts communicated here seems to me unassailable and definitive. I am, therefore, of the opinion that the problems have in essentials been finally solved. And if I am not mistaken in this, then the value of this work secondly consists in the fact that it shows how little has been done when these problems have been solved.
Later English translation (1961), McGuinness/Pears (70 words)
On the other hand the truth of the thoughts that are here communicated seems to me unassailable and definitive. I therefore believe myself to have found, on all essential points, the final solution of the problems. And if I am not mistaken in this belief, then the second thing in which the value of this work consists is that it shows how little is achieved when these problems are solved.
February 16th, 2017, German-to-English Google Translate (54 words)
On the other hand, the truth of the ideas conveyed here seems inviolable and definite. I am therefore of the opinion that the problems have essentially been finally solved. And if I am not mistaken, the value of this work is, secondly, to show how little has been done that these problems are solved.
Spanish translation, Unknown Origin (64 words)
Por otra parte la verdad de los pensamientos aquí comunicados me parece intocable y definitiva. Soy, pues, de la opinión de que los problemas han sido, en lo esencial, finalmente resueltos. Y si no estoy equivocado en esto, el valor de este trabajo consiste, en segundo lugar, en el hecho de que muestra cuán poco se ha hecho cuando se han resuelto estos problemas.
February 16th, 2017, Spanish-to-English Google Translate (61 words)
On the other hand the truth of the thoughts communicated here seems untouchable and definitive. I am therefore of the opinion that the problems have essentially been finally solved. And if I am not mistaken in this, the value of this work consists, secondly, in the fact that it shows how little has been done when these problems have been solved.
I suspect that the Spanish translation is built from the Ogden English translation¹. I’ve provided the word counts here because I believe there’s an interesting story to be told about the original sparsity of Wittgenstein’s writing and the subsequent blooming of words as future writers rewrote/translated his work.
The next sections pick apart each of these sentences as found in the original text and five translations⁵.
Dagegen scheint mir die Wahrheit der hier mitgeteilten Gedanken unantastbar und definitiv.
The three human translations seem to disagree with Google on a good word choice for the German word unantastbar. Google would prefer the very nice word “inviolable”; while the two English translators prefer “unassailable” and the Spanish translator prefers intocable or “untouchable.”
Interestingly, the online Harper-Collins German-English dictionary prefers an equally nice word not produced by any translator: “sacrosanct.” This would be a very fitting word, likely reflecting Wittgenstein’s intentions, but the religious overtones of the word are strongly out of character for how Wittgenstein perceives himself in this introduction, and I think it is a wise choice that no human or machine translator picked sacrosanct as the winning adjective.
The wordiness of the McGuinness/Pears translation is already evident in this first sentence. Ogden wrote “thoughts communicated here.” McGuinness and Pears required “thoughts that are here communicated.”
However, surveying all five translations, the clean simplicity of the German-English Google translation stands out. The words counts for this sentence are as follows:
The Google German-English translation is able to find a more parsimonious translation because it seems to recognize that the phrase “seems to me” is redundant in personal writing and Google is able to remove the superfluous “to me.” That the two English translators felt compelled to include the “to me” clause seems like a failure of the translation process. Clearly to them they were not the original writers of this sentence and the “to me” helps indicate just who is speaking in this sentence. But for Wittgenstein’s original intent, he needed no such clarification and for a moment it seems as if the two English translators stayed too long in their own heads and forgot about who the writer was. [see the 2021 postscript regarding Google’s change of mind]
A way in which the original German strongly wins in minimal words count is in that starting clause Dagegen, translated into English as “On the other hand” and Spanish as “Por otra parte.” Clearly German wins here on raw succinctness. I could think of the single English word Alternatively to convey a similar idea with less words. Whether this would work in a new translation would require understanding the context of the sentences just before this last paragraph in order to inspect what previous words Wittgenstein was trying to counter.
Instead of word count we can inspect syllable count on the observation that Wittgenstein’s clean writing can read like poetry and perhaps a preservation of his syllabic utterances would be a better measure of whether or not the translators have preserved his voice.
The syllable counts for the translations are as follows:
German has a reputation for being a language with large words containing high syllable counts, and the syllable counts reflected here do suggest that the relatively short 12 words of the first sentence might hide the richness of the expressivity if we note that the German required 2.2 syllables per word and the two English translators used 1.7 and 1.6 syllables per word.
But the beauty of the Google translation stands out here too. Google presented one of the sharpest translations by both word count and low syllable number, clocking in at 1.67 syllables per word and beating the number of syllables in the original German.
Wittgenstein’s writing is notably succinct and austere, and I suspect that he would have approved of the simple beauty of Google Translate’s proposal for this particular sentence.
One parting note: Google Translate chose to translate Gedanken as “ideas’’ and not the more usual word “thoughts.” We could save two syllables by jumping back to “thoughts”; I also suspect this word is truer to Wittgenstein’s voice.
Ich bin also der Meinung, die Probleme im Wesentlichen endgültig gelöst zu haben.
To remind non-German speaking readers (myself included!) of what this sentence reads like in English, I’ll repeat the Google translation: “I am therefore of the opinion that the problems have essentially been finally solved.”
Again without even counting words, I can anticipate that the two English translations founds themselves adding more words. The word and syllable counts follow:
An interesting aspect of this particular sentence is that the German-English and Spanish-English Google translations arrive at the exact same sentence even though the starting phrases appear to be different.
The key question of the translation of this sentence is what to do with the German phrase im Wesentlichen. The Harper-Collins dictionary translates this phrase as “basically” or “in the main.” The English and Spanish translators adapted this to “in essentials”, “on all essential points” or “en lo esencial.” Google interprets both the original German phrase and the Spanish phrase as a very concise “essentially.”
Yet here I am opting to vote against concision. I feel that the adverb “essentially” is thrown out a little too casually in modern English usage. While its root origin might have meant “pertaining to the essence of”, its modern usage is much more relaxed: something like “used for emphasizing what is the most important aspect of something or fact about something.” In this sense to replace the German phrase im Wesentlichen with “essentially” is a weakening of the language, and I rather like something like what McGuinness and Pears wrote — “on all essential points” — if they just didn’t have to use so many words to get there.
On the syllable counts we again see Google Translate staying most faithful to the counts in the original clean Germanic diphthongs, and both German and Spanish score higher than English on syllables per word.
There is another aspect of these sentences which I believe Google and the Spanish translations get wrong. It concerns the translation of the German word endgültig. This translates into English as “finally” but there are at least two different meanings for the word “finally,” and I believe only the Pears and McGuinness translation find the right meaning. What I believe Wittgenstein intends is that these problems have been solved for good, and without need to be addressed again, and the examples in the Harper-Collins German-English dictionary track with this: “that settles the matter once and for all” and “they’ve separated for good.”
Of the translations, only the McGuinness and Pears English translation preserves this meaning of the word finally and does so with their characteristic verbosity: “the final solution of”. Ogden and Google Translate leave some ambiguity in the translation as to whether Wittgenstein had finally managed to solve these problems after he or others had been working on them for quite some time, and the Spanish direct translation finalmente says the same thing. In fact in Spanish to say that these problems were solved with a degree of finality, a preferred term might be to say de manera definitiva.
This brings us to the final meaty sentence of that final paragraph…
Und wenn ich mich hierin nicht irre, so besteht nun der Wert dieser Arbeit zweitens darin, dass sie zeigt, wie wenig danmit getan ist, dass diese Probleme gelöst sind.
which is translated by Google Translate as “And if I am not mistaken, the value of this work is, secondly, to show how little has been done that these problems are solved.”
This last sentence is packed full with clauses and is clearly where the wheat could be separated from the chaff on this toy translation problem. We can start with the word counts:
I’ll make three observations from this table: 1) this sentence has the lowest syllable/word ratios of all three sentences across all three languages and I think this suggests a certain directness in the tone of the sentence. 2) the Spanish > German > English ordering for syllable/word ratios is preserved again and we might speculate that this represents a global ordering on these languages² and 3) This is where Google Translate managed a smaller word solution than the original German and pulled back the total word count for all three sentences to exactly match the German fifty-four words.
We’ll start with whether or not the beginning phrase should include the last words “in this” or “in this belief” or equivalently “en esto.” Google Translate did not add these two words and I suspect that the original German did not merit it.
However, the story gets interesting here.
If I enter the German phrase Und wenn ich mich hierin nicht irre alone into Google Translate and translate into Spanish, I get the following Y si no me equivoco, en el presente documento. This clearly says that the German words must contains a self-reference since the Spanish translation went so far as to say “in the present document.” In fact, the German/English cognate hierin is translated in the Harper-Collins dictionary as “in this,” directly confirming that the translation should have a self-reference, as found in the English and Spanish translations.
Pop quiz: what’s different about the German words in the following two translations produced by Google Translate?
Und wenn ich mich hierin nicht irre so besteht nun der Wert dieser Arbeit zweitens darin = “And if I am not mistaken in this, the value of this work is, secondly, the same”
and
Und wenn ich mich hierin nicht irre, so besteht nun der Wert dieser Arbeit zweitens darin=“And if I am not mistaken, the value of this work is the second thing”
Answer: nothing is different about the words!
The second set of words is punctuated with a comma as Wittgenstein wrote, and the first set of words is missing that comma. And that’s all that it takes to convince Google to drop the hierin self-reference from its translation in the second case! Wow, that’s some crazy secret sauce they have there at Google. On one hand the self-reference is redundant in this sentence and a mindfulness for word economy would allow us to drop the phrase anyways. Which would be all fine and good except the frugal Wittgenstein had chosen to place a hierin into his sentence and with his original comma also there, Google doesn’t want to respect his intent.
Moving on and speaking of word economy, McGuinness/Pears and Google both score high marks versus Ogden in the next phrase, “then the value of this work secondly is that it shows…” Where Ogden used the words “then the value of this work […] consists in the fact”, Google says the exact same thing with “the value of this work is […] to show.” That’s a very nice rewriting.
However when we get to the final clause of this sentence, we can finally ding the Google German-English translation on correctness! I would argue that the English translation produced by Google is generally using excellent English, perhaps with the exception of the split auxiliary in sentence two (sticking “essentially” between “have” and “been”.) However this last phrase finally contains a translation error that would suggest to an English speaker that the translation had been performed by a non-native speaker.
Google produced “to show how little has been done that these problems are solved.” It would be far preferable to Wittgenstein’s intent and correct English to write instead “when these problems are solved” or, alternatively, to write “now that these problems are solved.”
Lastly, there should be a question of verb tense in the translations of this last phrase. Ogden uses the perfect tense “have been solved.” McGuinness and Pears use the present tense “are solved.” The Spanish translator also used the perfect tense, presumably following Ogden [footnote], “han resuelto.” Like McGuinness and Pears, Google Translate chose to use the present tense: “are solved.”
Which is it? I don’t speak German but I already suspect that the present tense is the correct answer. I do recall from a hazy learning of elementary German that ist means “is” and sind means “are” — both in the present tense — and this would seem to correlate to Google’s translation. In fact, it is an interesting general question if the more literal translations made by Google never permit the changing of tense. For that matter, it is an interesting question if such manipulation is considered permissible in the world of translation³.
In any respect, Ogden and the Spanish translator changed the tense of the sentence and McGuinness/Pears and Google stayed true to the original.
Summary
Is there a winner here? I would like to know who Wittgenstein would have picked as his translator.
McGuinness and Pears write that they embarked on their translation of the Tractatus using letters from Wittgenstein about what he would have done differently with Ogden’s translation. And they clearly improved upon Ogden’s translation in at least two respects in the second and third sentences. Yet of the five translations they also seem most unable to stick to the parsimonious and succinct style that is so characteristic of Wittgenstein’s writing.
Ogden’s and Ramsey’s translation wins for being the definitive English translation to start all other translations, but I’m finding that it lacks three essential qualities that we might expect in a true victor: 1) too many words 2) use of ambiguous words in English where Wittgenstein was not ambiguous in German and 3) not fully respecting Wittgenstein’s choice of tense⁴.
The Spanish translation appears based on Ogden’s translation, so I don’t need to review it for quality.
Finally the Google Translate German-to-English translation is clearly a work of beauty. That first sentence from Google blew me away. I was specifically impressed by the use of “conveyed here” instead of the “communicated here” (or worse) phrase in the other three human translations. This is such a prettier set of words and meaning. I would pick the Google translation hands down if it weren’t that it contains the only translation error that I was able to detect in these readings.
Overall, my pick for a winner is Google although I’m happy to write in early 2017 that Google hasn’t completely yet dominated the art of natural language translation. But the evidence gathered here suggests that, true to the NY Times article, the end is that much closer for a large number of individuals in that profession.
After this much time spent with only a few starting words, I can’t but help but take my best stab at how I’d translate the German words, trying to maintain Wittgenstein’s original voice but also lifting his words into modern American usage. I’ll leave with this:
In contrast the truth of the thoughts conveyed here seems inviolable and definitive. I am therefore of the opinion that, in the main, the problems have been solved with finality. And if I am not mistaken in this, the secondary value of this work is to show how little is achieved now that these problems are solved.
That’s 57 words.
Postscript (added in 2021)
One of the advantages of choosing to translate a paragraph of a certain inviolable nature is that the musings generated here regarding machine and human translation have a timeless quality.
Point of fact, after I had written this entire essay I found out that a computer scientist named John Sowa had investigated Google Translate with my very same paragraph in 2010(!) and posted his results to an e-mail list.
So with Sowa’s investigation in hand I know that in March 2010 Google produced the following translation:
On the other hand seems to me the truth of the thoughts communicated here unassailable and definitive. I am therefore of the opinion that the problems largely been finally solved. And if I am not mistaken, so now is the value of this work, secondly the fact that it shows how little has been done that the problems are solved.
(Word to the wise: don’t try to read this too hard. It is syntactic and semantic garbage. I think we all have forgotten that 2010 Google Translate was a toddler.)
Now that I have put this essay on Medium, I can also update with the current Google translation (October 23rd, 2021):
On the other hand, the truth of the thoughts communicated here seems inviolable and definitive to me. So I am of the opinion that the problems have essentially been finally resolved. And if I am not mistaken, the value of this work is, secondly, that it shows how little has been done that these problems have been solved.
Why am I not surprised that this translation is beautiful? Comparing to the five translations at the top of this essay, we can see:
- Google Translate is at no point merely copying some other human translator’s work. The first sentence now reads as a merger of the Ogden/Ramsey translation plus Google’s 2017 translation.
- More interestingly is the use of the word resolved instead of solved at the end of sentence two. I think this is the core masterstroke of the new Google translation. No previous English translator went there (myself included), and now that I see it I think it’s the perfect word choice.
- 58 words. The full analysis is here
So the new Google Translate version continues to score well for parsimony versus the original text.
Nonetheless the most difficult last sentence has barely changed in Google’s translation between 2017 and 2021. In this regard, I continue to prefer my own disentangling of that original utterance:
And if I am not mistaken in this, the secondary value of this work is to show how little is achieved now that these problems are solved.
Footnotes
[1] The Spanish translation I’m providing here can be found here. I wasn’t able to find a name of the translator nor year of translation. A close inspection of the Spanish translation for this particular paragraph suggests that it was closely derived from the Ogden translation of 1922 and not necessarily the original German words. The first sentence is fairly consistent, with one exception, throughout the five translations and doesn’t help very much in establishing this observation. However, the second sentence uses the awkward English phrase “in essentials” in the Ogden translation and this phrase seems lifted verbatim into Spanish in the phrase “en lo esencial”. The third sentence provides more definitive proof of the origin of the Spanish translation since the original German only contained 29 words and could be translated by Google into a tight 25 words. The Spanish sentence stays true to the wordiness of Ogden’s translation and contains 33 words. It translates Ogden’s phrase “And if I am not mistaken in this” with the literal translation “Y si no estoy equivocal en esto”. The extra directional clause “in this” is not present in the original German and suggests that the Spanish translator is reading over Ogden’s shoulder to produce the Spanish copy.
[2] For example the plot here suggests that Italian has more phonemes/word than German. And this article suggests that the information content of each syllable of Spanish is lower than the equivalent information content in English or Mandarin and 3) This is where Google Translate managed a smaller word solution than the original German and pulled back the total word count for all three sentences to exactly match the German fifty-four words.
[3] Translation is an entire field of study and I do not pretend or claim to have studied much in the way of translation theory. It’s fascinating and beautiful to me that there is an entire field of study called translation theory, and I’m sure I’d enjoy the works in this domain if I were less employed in other non-translation related activities most days.
[4] I didn’t consult this article while writing this post but I bet this journal article from 1999 comparing the Ogden/Ramsey and McGuinness/Pears translation is an interesting read.
[5] The three main sections make frequent use of the original texts presented in the first section. Without those sentences in constant focus, this essay likely suffers a bit for coherence. This begs the question of what would be the optimal format for writing like this? I believe there are two answers: 1) printing this entire essay on paper while ensuring that each section remains on a separate page, allowing one to have the first section and subsequent sections in view at the same time or 2) rearranging this essay to fit on a poster such that the first section is in the center and the remaining sections dangle as extensions from the center.
