I haven't posted in a while, but I still do take the time to think about the problem and I do have Voynich manuscript as something I get alerts about. I also read the thread and was interested by how it parallels some thoughts I've been having.
My approach to VMS assumes it's a real natural language, but not any specific language. So I'll agree it's a language, reserve judgement on whether it's forgotten or just not yet identified, and offer no comment on it's"true origins". I'm sure all of us know at least a few words that have different meanings in different languages as well as multiple meanings in out own. I don't mind people offering identifications for certain words, but until someone offers a framework all the words fit into, the words are a starting point for discussion, not the beginnings of a solution.
Finding lots of words one believes to be Turkish or derived from Turkish is helpful in that it allows one to generalize about what the other written words might sound like in speech. But allowing what those words sound like to drive a search for similar Turkish words is likely the wrong approach. Albanian has a high percentage of loan words from Turkish. Over 20%, maybe as high as 30. I'm not suggesting the text is Albanian, just pointing out that finding Turkish words wouldn't conclusively prove the script encoded only Turkish. It wouldn't even prove VMS was authored by someone who knew Turkish grammar or syntax.
While labeled images seem to provide a good starting point they require you to properly interpret both the text and the image. if you make assumptions about the image that are wrong and arrive at an interpretation of the text you find acceptable, it's likely to also be wrong. Even worse, since you're happy with it you're unlikely to consider other options. I'd suggest an approach that relies less on the images and more on only what the text might sound like.
Begin by selecting 200 common words that are likely to appear in the text. Nouns like fire, water, earth, stone, gold, silver, star, leaf, stem, root, etc. Adjectives like wet, dry, black, hot, old, small, etc. I put a list like that together and decided against adding verbs because conjugation changes them enough they aren't always easy to spot. Pronouns, conjunctions, articles, and prepositions tend to be very language specific. No one borrows pronouns, and what we're looking for here, at least in part, is loan words. If we can find enough loan words we may be able to limit the geography of the language and definitively eliminate all the Nuatal, Ancient Babylonian, and "written by aliens" discussions.
The next step is to get phonetic spellings and alphabetic spellings of all these words in as many target languages as possible (focusing on Indo-European languages in use between 1200 and 1600 AD). Since there seems to be agreement that the approach has some success with Turkish, the abbreviated dictionaries you had could then be used to sweep VMS page by page. The phonetic encoding derived from the Turkish analysis or others identified through a NN analysis could then be used to build Voynichese specific dictionaries for various phonetic encodings. Best dictionary is the winner.
The phonology would be confirmed by finding a high percentage of the target words in the text using as few of the target dictionaries as possible. You'd probably have to read the patchwork text yourself to see if it made sense since (I hope) people still have more relevant experience with language than machines. The NN could be trained with existing historical texts that are written in one or more known languages. Once the phonology was established with a reasonable degree of certainty you still wouldn't (and might never) have a full solution. But it seems like a reasonable approach to at least creating some gaps to fill in where now there aren't even gaps..
My approach to VMS assumes it's a real natural language, but not any specific language. So I'll agree it's a language, reserve judgement on whether it's forgotten or just not yet identified, and offer no comment on it's"true origins". I'm sure all of us know at least a few words that have different meanings in different languages as well as multiple meanings in out own. I don't mind people offering identifications for certain words, but until someone offers a framework all the words fit into, the words are a starting point for discussion, not the beginnings of a solution.
Finding lots of words one believes to be Turkish or derived from Turkish is helpful in that it allows one to generalize about what the other written words might sound like in speech. But allowing what those words sound like to drive a search for similar Turkish words is likely the wrong approach. Albanian has a high percentage of loan words from Turkish. Over 20%, maybe as high as 30. I'm not suggesting the text is Albanian, just pointing out that finding Turkish words wouldn't conclusively prove the script encoded only Turkish. It wouldn't even prove VMS was authored by someone who knew Turkish grammar or syntax.
While labeled images seem to provide a good starting point they require you to properly interpret both the text and the image. if you make assumptions about the image that are wrong and arrive at an interpretation of the text you find acceptable, it's likely to also be wrong. Even worse, since you're happy with it you're unlikely to consider other options. I'd suggest an approach that relies less on the images and more on only what the text might sound like.
Begin by selecting 200 common words that are likely to appear in the text. Nouns like fire, water, earth, stone, gold, silver, star, leaf, stem, root, etc. Adjectives like wet, dry, black, hot, old, small, etc. I put a list like that together and decided against adding verbs because conjugation changes them enough they aren't always easy to spot. Pronouns, conjunctions, articles, and prepositions tend to be very language specific. No one borrows pronouns, and what we're looking for here, at least in part, is loan words. If we can find enough loan words we may be able to limit the geography of the language and definitively eliminate all the Nuatal, Ancient Babylonian, and "written by aliens" discussions.
The next step is to get phonetic spellings and alphabetic spellings of all these words in as many target languages as possible (focusing on Indo-European languages in use between 1200 and 1600 AD). Since there seems to be agreement that the approach has some success with Turkish, the abbreviated dictionaries you had could then be used to sweep VMS page by page. The phonetic encoding derived from the Turkish analysis or others identified through a NN analysis could then be used to build Voynichese specific dictionaries for various phonetic encodings. Best dictionary is the winner.
The phonology would be confirmed by finding a high percentage of the target words in the text using as few of the target dictionaries as possible. You'd probably have to read the patchwork text yourself to see if it made sense since (I hope) people still have more relevant experience with language than machines. The NN could be trained with existing historical texts that are written in one or more known languages. Once the phonology was established with a reasonable degree of certainty you still wouldn't (and might never) have a full solution. But it seems like a reasonable approach to at least creating some gaps to fill in where now there aren't even gaps..