It is my opinion that attempting an AI or any other online common model for writing Konkani first needs an understanding of Konkanni in at least four different scripts, not beginning in English. For example, the ergative of Konkanni, although used by all, a majority of speakers and writers are ignorant of its presence and shape. The transitive, intransitive and causative in Konkanni is not done properly. vochonk is written as vochunk, roddonk as roddunk. It is a zaunv form, it is not kor or koroi form, and its equivalent in English is: intransitive. But ti bori zali, zavunk. to boro zalo, tem borem zavunk. To uttlo, uttonk, not uttunk. Intr. in English but za in Konkanni. kor, korunk. koroi, koronvk. These forms in Romi are not uniformly used by all. There is immense variation, even in FN. I propose, first agree on a soroll form or common form of Konkanni. Don't go for bamonn (with Marathi-speaking and writing past elders) form of Konkanni. It is recent, most recent. But they do have a base Konkanni if they care for it: Chan'neache rati, maddache savllent etc. Goinchea mhojea... etc. Monglluranchea xarant ek cheddum dekilam, kazar mhojelaguim zata mhonnon tannem sangilam etc. There is a lot to explain, which I have worked out over thirty years of field research, but I did not come to put it down. The MSS were stolen at Alvas, Mudubidri and the Lexicon work of 1988-1991 was suppressed by the promoters. St Aloysius College. And its re-starting was suppressed again by SAC dtb U. There is a lot to report. Is it worth while? WRDS
On Wed, Apr 15, 2026 at 1:31 AM John de Figueiredo <[email protected]> wrote: > Thank you, Frederick, for this helpful information > John M. de Figueiredo > Sent from my iPhone > > On Apr 14, 2026, at 12:43 PM, Frederick Noronha < > [email protected]> wrote: > > > Prof John and all, > > Google Translate handles Konkani *only **moderately well.* Its > performance varies widely depending on the dialect, script (Devanagari, > Roman, Kannada, Malayalam, Perso-Arabic), and sentence complexity. It > manages simple phrases and everyday communication. But it struggles with > literary or technical texts. Especially those involving regional idioms or > culturally specific expressions. It often produces awkward or inaccurate > results. This could be because of its relatively limited digital corpus for > Konkani compared to major languages. It’s a helpful starting tool, but not > fully reliable for precise or complex translation. At the moment, it is > focussed on Nagari (Devanagari) Konkani, though Romi abilities were also > promised in the past. See Meet the Goan American who Mangified Google > Translate in Konkani | Prudent Media Goa > https://www.youtube.com/watch?v=J9Mh0VcPD20 > > <1.png> > > > Google Translate does not primarily “translate into Romi Konkani” as a > fully separate linguistic output; rather, what you often see at the bottom > (Roman script after Devanagari output) is usually generated through what > they're calling a *transliteration layer*. In simple terms, the system > first produces the Konkani meaning in Devanagari, and then applies a > rule-based + statistical mapping that converts those sounds into Latin > letters based on likely phonetic equivalents (for example, “क” → “ka”, “च” > → “cha”). This Roman output is therefore closer to a *phonetic rendering > of the Devanagari text* than a true native Romi Konkani orthography, > which has its own historical spelling conventions and variations. Because > of this, the Romi line can look inconsistent or “non-standard,” especially > with vowels, aspirated consonants, and regional pronunciation differences. > > This might make sense to those who are in the tech domain: AI translation > for Konkani—especially across *multiple scripts (Devanagari, Roman/Romi, > Kannada)*—is not a single “one-click” system working perfectly > end-to-end. Instead, most systems combine three components: (1) *neural > machine translation (NMT)* to convert meaning from source to target > language, (2) *script normalization*, and (3) *transliteration layers* > that map sounds into different writing systems. > > Techies keep calling Konkani a "*low-resource* (limited training data)" > data. I suspect much of the text produced over decades and even centuries > have not been scanned; and if they have been scanned in > government/taxpayer-funded projects, these have not been adequately shared > or utilised. Many digitisation projects have been announced in Goa since > even the 1980s and 1990s, but what's the outcome of this is not exactly > clear. Systems like Google Translate often rely on broader > Indian-language models and then approximate output, especially for Roman > script, which is usually generated via phonetic transliteration rather than > a native Konkani orthographic model. > > Current research-grade systems like *AI4Bharat’s IndicTrans2* (used in > academic and open-source Indic NLP work) are believed to outperform > commercial tools for many Indian languages, including Konkani. These have > trained specifically on multilingual Indic corpora and supposedly handle > script mapping more consistently. They might lack user interfaces that the > commercial tools have. Microsoft Translator and Google’s system are > blaming the limited training data and uneven coverage across scripts for > their lack of accuracy. > > Another interesting tool is Konkanverter.com. It works *quite well for > basic Romi transliteration*, especially when converting from Devanagari > or Kannada into Roman script in a consistent, rule-based way. Its strength > is that it follows a *standardised orthography model*, so outputs are > fairly predictable and useful for learners or documentation. It is however > still a *rule-based transliteration engine, not a dialect-aware AI > system. * So it does not truly “understand” spoken variation or regional > pronunciation differences. > > Konkanverter.com sometimes struggles with *dialectal variation and > ambiguity in *Goan sub-dialects (Bardez, Saxtti, Antruz) and Catholic vs > Hindu usage differences. Many Konkani words change pronunciation and vowel > quality across regions, but Konkanverter generally maps one standard form > rather than adapting to those shifts. It has been called a *clean script > converter, not a dialect-sensitive transliteration AI. I*t performs well > for formal text but only moderately for capturing real-world spoken > variation. I find it rendering Romi script better than others. > Please note: I am not a technical person, but have used both *Google > Translate* and *Konkanverter.com* a fair bit for my own purposes. Those > who are into this subject could correct the above..... > > FN > > _/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/ > _/ Frederick Noronha फ्रेडरिक नोरोन्या * فريدريك نورونيا > _/ AUDIO https://archive.org/details/@fredericknoronha > _/ http://goa1556.in +91-9822122436 784 Saligao Goa > _/ Goanet :: 30 years of discussions. [email protected] > _/ http://lists.goanet.org/pipermail/goanet-goanet.org/ > _/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/ > > > On Mon, 13 Apr 2026 at 09:08, John de Figueiredo <[email protected]> > wrote: > >> Konkani is now one of the languages in Google Translate. Congratulations >> to those who made this possible. Looks like it is available in Devanagari >> only. I tested a few words and they were fine. With AI it should be easy to >> include other scripts. >> John M. de Figueiredo >> Sent from my iPhone >> >> -- >> You received this message because you are subscribed to the Google Groups >> "Goa-Research-Net" group. >> To unsubscribe from this group and stop receiving emails from it, send an >> email to [email protected]. >> To view this discussion, visit >> https://groups.google.com/d/msgid/goa-research-net/E11C911D-637A-4358-B3C8-67B9A591A33F%40sbcglobal.net >> . >> > -- > You received this message because you are subscribed to the Google Groups > "Goa-Research-Net" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > To view this discussion, visit > https://groups.google.com/d/msgid/goa-research-net/CA%2Bmqab_8%2BbxxFcai6NNwMNUi75q7fhaYaBh5sv8O9vqzkrT8sw%40mail.gmail.com > <https://groups.google.com/d/msgid/goa-research-net/CA%2Bmqab_8%2BbxxFcai6NNwMNUi75q7fhaYaBh5sv8O9vqzkrT8sw%40mail.gmail.com?utm_medium=email&utm_source=footer> > . > > -- > You received this message because you are subscribed to the Google Groups > "Goa-Research-Net" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > To view this discussion, visit > https://groups.google.com/d/msgid/goa-research-net/6BFD9DB8-0B2B-4BBE-A3C6-7FD41F832101%40sbcglobal.net > <https://groups.google.com/d/msgid/goa-research-net/6BFD9DB8-0B2B-4BBE-A3C6-7FD41F832101%40sbcglobal.net?utm_medium=email&utm_source=footer> > . > -- You received this message because you are subscribed to the Google Groups "Goa-Research-Net" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion, visit https://groups.google.com/d/msgid/goa-research-net/CA%2BvNr4LawEMb95GLNcJGNTX6XAFm_%2ByKw4j_FsmtDzjuOt%3DP7Q%40mail.gmail.com.
