← All workshopsRead FastText

Natural Language Processing

Build a word from shared fragments

Remove the whole-word contribution

A missing whole-word entry need not leave Botie without a representation when its spelling fragments have vectors.

BOTIE

With a zero whole-word entry, which vector remains after both known fragments contribute?
The inputs we are comparing

Word vector

A known whole word

  1. 01
  2. 10

Only the subwords are known

  1. 00
  2. 10

First subword vector

A known whole word

  1. 00
  2. 12

Only the subwords are known

  1. 00
  2. 11

Second subword vector

Same in both cases

  1. 01
  2. 11
Result for a known whole word
  1. 02
  2. 13

For a list use brackets, such as [2, 4]. For a matrix put each row in brackets. Text answers can stay as text.

Two spelling fragments pull in different directions

This check uses the pieces from “Assemble the word from its available parts”.

BOTIE

An unfamiliar word has no whole-word vector yet. Botie does recognise two of its spelling fragments, whose coordinates partly cancel. Produce the unnormalised sum used by this workshop, keeping all three coordinates.
Word vector
  1. 00
  2. 10
  3. 20
First subword vector
  1. 02
  2. 1-1
  3. 23
Second subword vector
  1. 0-2
  2. 14
  3. 21

Use a number, text, or an array matching the requested result. For a matrix, put each row in brackets. Saving your prediction runs your current construction on these inputs and checks the earlier examples too.

Build a construction in the workspace above first. This case will test those pieces.

What would be missing if neither the word nor any of its fragments had useful vectors?

Sharing spelling features provides coverage, but does not by itself establish the meaning of an unfamiliar word.

This is a small construction problem. The lesson explains the assumptions and the parts of the full method that this example leaves out.

Work through the Python challenges →
Continue the courseGloVe →
← Word2vec