Context-Vector Word Sense Disambiguation in Translation

Words like "bank", "spring", "bass" and "light" have several unrelated meanings, and a translator — human or machine — has to pick the right one from context alone. This simulator embeds each candidate meaning and each surrounding context word as a point in a 3D semantic space, then computes a weighted context vector from the words you leave switched on and measures its cosine similarity to every candidate. The closest meaning wins; a softmax over the similarities produces a confidence score, and a configurable threshold decides whether the system auto-translates or flags the segment for human review — the same trade-off real machine-translation pipelines make between speed and reliability. Shrink the context window or switch off the on-topic words and watch the decision flip, or turn ambiguous, in real time.