Some “linguistic formulas” to translate French texts into English?

Question

I am not sure is it correct to ask my question here or not!

I've asked this question here (in MathStackExchange) before! Maybe it is better to see the question there, because it was written colorful there.

---

---

---

I have to read some french texts. I think maybe there is some linguistic table (~LINGUISTIC Formulas) which can help me to translate the word from French to English.

I am looking for such LINGUISTIC TABLES to translate from French to English.

But What do I mean by a (~LINGUISTIC Formula)?

---

1. Consider the following formula: {duced ⟷ duit (duite)}

I mean that: wherever we see "duit (duite)" at the end of a French adjective (derivated from a verb), we can replace it by "duced" to translate it to English:

photoinduced, superinduced, cotransduced, reintroduced, overproduced, outproduced, oversauced, reproduced, introduced, transduced, reinduced, subduced, traduced, conduced, produced, spruced, adduced, deduced, reduced, abduced, seduced, induced, deuced, sauced, educed,

for instance, we have: introduced = introduit (introduite).

---

2. Also, let's consider the following one-sided changes: {ô→os} and {ê→es}

hôpital=hospital, hôtel=hostel, forêt=forest, côte=coast;

conquête=conquest, tempête=tempest; ancêtre=ancestor, bête=beast;

pâté=paste, île=isle,

3. Also, consider this one-sided change: {é at the beginning of the French word→s}

épice=spice, éponge=sponge, étrange=strange, école=school, étudier=study, état=state.

---

---

---

Please let me know if still there are some ambiguities about the LINGUISTIC Formulas.

I am looking for such LINGUISTIC FORMULAS to translate from French to English.

Answer 1

Unfortunately, it can't be done.

Translation is an immensely complicated process, and nobody's ever made a mathematical procedure that can do it with any reliability. The best we have right now are systems like Google Translate, which sometimes work and sometimes fail hilariously, but are nowhere near as reliable as a good human translator.

What you've found here are some cognates between French and English, inherited from a common ancestor. And with these words, it is sometimes possible to figure out the changes that happened from that ancestor to Modern French, and from that ancestor to Modern English, and use them to convert one to another.

But the vast majority of words don't have this nice relationship. Any rule that converts apple to pomme is going to be a special case that doesn't generalize at all, since the two words are unrelated.

Once you've converted all the stems, you then need to figure out the differences in morphology, the way words are constructed: should English "speak" correspond to French parle, or parles, or parlons, or parlez, or parlent, or parler, or…?

And even after all that, you need to figure out the differences in syntax, the way words fit together to create larger units like phrases and sentences…it's simply too complicated of an undertaking for straightforward formulae to capture.

Answer 2

Your examples are of cognates and loanwords (mostly loanwords), which means words that have the same origin in English and French, or that English borrowed from French. English does have many words that come from langues d'oïl, i.e. French dialects from various places and point in time. A major contributor to this has been the Norman conquest.

For this reason, you can definitely find a lot of patterns between some groups of French words and English words (but no guarantee that the pattern will always be present, because words may have been borrowed from entirely different eras, when both languages were different, or not from the same French dialect).

On the other hand, the absolutely isn't a "formula", nevermind a mathematical one, that just lets you simply convert French to English: translation in general is best accomplished by heuristics, and of course, in the case of English and French, while there are many loanwords, most words simply aren't related in any straightforward way, and as such no unifying "table" or "formula" could connect them.

Answer 3

I would call what you're describing character-level rules-based machine translation (RBMT).

RBMT lost out to SMT in the early 2000s, and SMT evolved into NMT in the 2010s.

But character mappings are still used in production systems like Google Translate for "translation" between orthographic variants, so for example between Cyrillic and Latin Serbo-Croatian, or between Simplified and Traditional Chinese, where the overall semantics and syntax are the same. It's not really better than NMT though.

Char-level RBMT could maybe work between Hindi and Urdu or Tajik and Persian or British and American English, although still the results would be a sort of translationese. Production MT systems are starting to supported translation between European and Brazilian Portuguese, and my bet would be they'll use NMT for that, not RBMT.

But RBMT does not work well for translation between actual distinct languages or even dialects, because both semantics and structure are not 1:1. In the early days of the internet, someone created a Schwobifying-Proxy to convert standard German to Swabian dialect - a very similar language - but even that doesn't work, it's more of a joke.

We could find examples of cognates shared between Japanese and English - mostly direct loans - and maybe define some rules that mostly correctly predict the conversion. But could we then translate Japanese to English? No. The coverage isn't full anyway, the meanings are shifted, and of course the syntax doesn't match. (As other answers wrote.)

The set of cognates shared between English and French happens to be far larger, but the overall equation is the same.