In the past, I have used FOMA and TRmorph as a building block and basis for Ottoman conversion. However, I saw that writing something on top of a morphological analyzer to convert Ottoman to Turkish requires almost another morphological analyzer. (This is also true for Turkish to Ottoman conversion as well, because spelling rules of Ottoman require another layer of FSTs.)

After that initial failure, I decided to write an ad-hoc algorithm that uses surface-level representations of Turkish and Ottoman. The result has been running in dervaze for two years now.

But as it can be observed, it’s slow. I was working to replace it with a C library that will also be used in mobile applications. I have completed the dictionary part that’s used to search for words. In this version of the library, I was using trie structures.

I have decided, however, that using FSTs to translate may be more robust and formal in the long run. When you use tries, it’s almost like using half of FSTs without much thought on states. As I’m developing this solo and my main concern is expandability, converting ad-hoc rules to states seemed easier than writing a C library.

I will be writing about this experience in the coming days.

I’ve learned that there is a specialized Common Lisp implementation for mobile devices, mocl.

It should also be possible to write C-compatible shared libraries in CL and use them in iOS and Android. Although I began to prefer writing libraries in plain C (instead of any other language), I may switch to CL if there is enough motivation.

I’m looking for a one-to-one mapping between the different visual elements of a word and its representation in UTF-8. The labels should be simple to remember, yet distinctive enough to represent visual variations of words.

I’m considering creating letter+digit codes. The letter part will reflect the most similar sound, and the digit will reflect the visual variation. In this case, I need a Latin letter for each letter in the Ottoman alphabet, but there are not enough letters in the standard Turkish alphabet to denote all the letters of Ottoman. So, how can I solve this?

For example, in Ottoman, a word is spelled as mnwr (mim-nun-vav-ra), following the Arabic orthography, but in Modern Turkish, the spelling reflects the pronunciation as münevver. Since a one-to-one mapping is not possible between these two writing systems, a set of possible Ottoman spellings must be produced using a regular expression.

When the parser sees münevver, it should convert this to a pattern like mv?nh?(a1)?ww?h?r. This pattern produces a set of strings, where mvnha1vvhr might be the longest and mnwr the shortest. By performing a dictionary search, the system can verify that there is a word in Ottoman spelled as mnwr, thus selecting it as the correct spelling.

The dictionary in our study consists of word labels. The system will look up a set of handwritten word images after this label lookup and search for these images in the document. It can also create a set of candidate images from the regular expression by rendering each candidate spelling.