A company is testing two different language modeling systems for a customer service chatbot.

• System A was trained on a large corpus of text. When it encounters a phrase like 'purchase a ticket', it performs well. However, when a user types 'acquire a ticket', the system fails, as the exact word 'acquire' was rarely seen in the context of 'ticket' in its training data.

• System B was trained on the same data. When it encounters 'acquire a ticket', it correctly understands the user's intent. The system's internal data shows that it represents the words 'purchase' and 'acquire' in a mathematically similar way.

Based on this information, analyze the fundamental difference in how these two systems represent words. Explain why this difference allows System B to generalize to the unseen phrase while System A cannot.