Work — Pandamtl

Machines often struggle with gendered pronouns (he/she/it). If a character’s gender seems to flip mid-sentence, just go with the flow. Learn the "Terms":

| Auxiliary Task | Purpose | Example Output | |----------------|---------|----------------| | Part-of-Speech (POS) Tagging | Helps disambiguate word senses | The_DET cat_NOUN sat_VERB | | Named Entity Recognition (NER) | Improves proper noun translation | [PER: John Smith] | | Word Alignment | Aligns source-target words (for attention guidance) | IBM Model 2 style | | Language Model (LM) | Predicts next token in source language (denoising) | Masked LM objective | | Sentence Similarity | Keeps embedding space consistent across languages | Cosine similarity loss | pandamtl