Question
Can a character-level model use word boundaries more directly through its positional representation?
Approach
The model combines absolute character position, position within a word, and the index of that word. The intended task is sequence tagging.
The implementation includes data preparation, model training, evaluation, and prediction, with a download path for Universal Dependencies data.
Evaluation
Removing each position component would help distinguish what the model learns from each signal. Saved model files alone do not settle that comparison, and this note does not report an accuracy result.