A data scientist is working on two separate natural language processing projects. Both projects use a model architecture where all input tokens are processed in parallel, rather than one after another.

• Project A: A sentiment analysis model for customer reviews. The model's first step is to identify the presence of specific positive and negative keywords from a predefined list, treating the review as an unordered collection of these keywords.
• Project B: A machine translation system that translates English sentences into French, where word order is critical for grammatical correctness and meaning.

For which project is the inclusion of position-specific information vectors (which are added to the initial token vectors) an essential design choice? Justify your answer by explaining why it is necessary for one project but not for the other.