Morpheus: A Morphology-Aware Neural Tokenizer and Word Embedder for Turkish
Morpheus is a single neural model designed to be both a lossless, morphology-aware tokenizer and a structured word-embedding producer for Turkish. It uses a differentiable program to learn morpheme boundaries from character probabilities. This results in the lowest bits-per-character (1.425) and high morphological purity, making it excellent for tasks requiring accurate decoding and understanding of Turkish word structure.