For those willing to read some texts in gallurese language (‘gaddhuresu‘), a sister-language from Corsican:
There is also some poetry:
Translating properly French ‘de’ is much complicated, since there are many different grammatical cases. To cite but a few of them:
- ‘de’ as a noun complement: ‘la fille de Jean, le chapeau de la mariée’ (Jean’s daughter, the bride’s hat) = a figliola di Ghjuvanni, u cappeddu di a spusata: ‘de’ = di
- ‘de’ from ‘du’ = ‘de le’, as a partitive article: ‘je veux du vin’ (I want wine) = vogliu u vinu
- ‘de’ from ‘des’ = ‘de les’, as a partitive article: ‘je veux des cerises’ (I want cherries) = vogliu i chjarasgi: ‘de’ = <void>
- ‘de’ as a preposition: venir de , partir de, descendre de (to come from, to leave from, to go down from) = vena da, parta da, falà da: ‘de’ = da
- ‘de’ within ‘ capable de ‘ = à: il est capable de partir (he is able to leave, he may leave) = hè capaci à parta: ‘de’ = à
To sum up: ‘de’ is much versatile. Depending on its grammatical nature, French ‘de’ may be translated into di, da, <nothing>, à. But the sitution is probably much complex than that, since some other cases should be missing here.
Open test: 1 – 6/104 = 94,23%.
6 errors: vocabulary (2), disambiguation (2), ‘de’ (partitive article, 1) + ‘ixe’ unknown (is it erroneous? since it would be better written ‘IXe’ or ‘IXème’).
Now the big question is this: how hard is it to transform a single pair MT (rule-based) system into a multiple pairs ecosystem? To put it in its simplest form: how hard is it to turn a MT system related to a given pair A-B into an ecosystem with, say, 26 languages A, B, …, Z and the corrresponding full set A-B, A-C, A-D, … B-C, B-D, …, X-Z, Y-Z of pairs.
It relies on the two following questions:
The current list of pairs of Okchakko translator is the following:
- French-Corsican (cismuntincu)
- French-Corsican (sartinese)
- French-Corsican (taravese)
Posted in blog
Tagged reverse pair
The problem of dictionary coherence is notably related to synonym transitivity: are synonyms transitive one with another? Usually, the response is negative. But isn’t this established wisdow questionable? Don’t we need to distinguish between:
- quasi-synonyms which are untransitive
- true synonyms, which are transitive: for example, the verb ‘finir’ in French (to end) translates indifferently into compie or finisce in Corsican ; in the same way, ‘demander’ (to ask) translates indifferently into dumandà or chere in Corsican ; ‘envoyer’ (to send), third, translates either into mandà or into invià: these are true synonyms, which are cross-language transitive
This casts lights on the fact that our notion of synonymy is cross-languages: if words A and B of a given language L1 are synonyms, and if a word C of another language L2 is synonym with A, then C is also synonym with B.
Open test performed on june 7, 2018: 1 – (5/110) = 95,45%. Remaining errors are:
- vocabulary: ‘pur-sang’
- partitive article: ‘des étalons importés’
- translation of preposition ‘par’: ‘par l’ajout’
- translation of preposition ‘de’: ‘du milieu’
- semantic disambiguation error: ‘du milieu’ should translate into da a mità
Language pair reversal: Rule-based translation is currently disregarded, in particular for the reason that it needs a lot of work and expertise. But is that reason well-grounded? Now focussing on a given language pair, we can formulate this question more accurately as follows: from a given pair language1/language2 translator, how much work is necessary to get a working translator language2/language1? To say it otherwise: aren’t there some techniques to speed up this process?
Will progress in MT require some breakthrough in grammar conceptualization? Rule-based MT requires a detailed typology of grammatical categories. But it seems that the present state of grammar conceptualization is insufficient for MT purposes. Some enhanced typology is in order. In particular the common typology relating to:
- partitive article, i.e. the disambiguation of French ‘de’
is in need of a more accurate conceptualization.