Extract only the entities defined by the task

Download the blank learner CSV. Complete your first pass before opening the answer key.

This optional lab uses synthetic practice records. Read the lab introduction and intent rules first. No paid platform or AI account is needed.

Learn

Entity annotation marks a piece of text and its type. Accuracy includes both the type and the text boundaries. In this exercise you record the exact text rather than character offsets, which avoids software-specific offset conventions. If a later project needs offsets, follow that project's definition of indexing and end boundaries.

Entity guideline version 1.0

Allowed types:

• ORDER_ID: the digits immediately following the word “order” (case-insensitive), including leading zeros. Do not include the word “order,” a trailing comma or a full stop.
• PRODUCT: exactly one of these catalogue phrases, ignoring case: “blue notebook,” “desk calendar,” “gel pen.” Copy the original case. Generic “notebook,” a colour alone or a plural “gel pens” is not a match in this deliberately strict exercise.
• DATE_TEXT: the exact explicit day-month text such as “18 November,” or one of “today,” “tomorrow,” “next Friday.” Match case-insensitively and preserve original case. Do not infer a calendar date from a relative expression.

Annotate every occurrence, in left-to-right order. If the same span text appears twice, include two entries in that order. Return an empty list when there are no defined entities. These rules are intentionally narrow so the answer can be checked consistently.

Worked example

Text: “Order 0042: the blue notebook is due tomorrow.” Entities: ORDER_ID “0042”; PRODUCT “blue notebook”; DATE_TEXT “tomorrow”.

“Tomorrow” does not become a date. The text has no reference date, and this task is extraction rather than date resolution.

Try it

Annotate N01–N08 in entities.csv. Use a list such as ORDER_ID:0042; PRODUCT:blue notebook. Include repeated mentions. Do not extract someone's name, infer a missing product or expand the allowed types.

Check your process

Can someone locate each marked span in the untouched source text? Did you preserve leading zeros? Did you mistake a word inside a larger phrase for a permitted catalogue phrase? Compare your answers with the entity key only after your first pass.

Lab sources.