Overview

The lexicon topic fixed a business word list for 10-K tone. This topic asks where that list is applied, and what happens when the representation stops treating every word type as orthogonal. A 10-K and an earnings-call transcript are different media. An embedding is a different representation: words that keep similar company can sit near each other even when the strings differ.

By the end of this note you should be able to:

• separate filing tone, call tone, and readability, and name the parsing failures that change a 10-K section;
• state why a 10-K lexicon needs an explicit modification before it is used on a call;
• define a word embedding from the distributional hypothesis and list the design choices that change what “similar” means.

1. Contents
2. The big picture