Imagine a Classroom

You're a teacher and 5 students are sitting in a row:

Seat 1: Rahul | Seat 2: Priya | Seat 3: Amit | Seat 4: Neha | Seat 5: Sana

Now someone asks you: "Who is sitting next to Priya?"

You can answer easily because you can see the seating arrangement. You know the order — who is first, who is second, and so on.

Now Imagine You're Blindfolded

Someone gives you 5 chits with names: Rahul, Priya, Amit, Neha, Sana.

But the chits are all mixed up in a bag. You have no idea who was sitting where.

Now can you answer "Who is next to Priya?" No! Because you lost the order information.

This is Exactly the Transformer's Problem

When a sentence like "The dog bit the man" enters a Transformer:

• The model converts each word into a number (embedding)
• But it processes all words at the same time (in parallel)
• So it's like getting all the chits in a bag — it doesn't know which word came first, second, or third

Without order, these two sentences look identical to the model:

• ✅ "The dog bit the man" (dog bites man)
• ❌ "The man bit the dog" (man bites dog)

Completely different meanings! But same words, same embeddings.