Assign part-of-speech tags using each word's most frequent training tag and a deterministic fallback for unseen words.
