Making Text Easier to Manipulate by Removing Surface Level Variations

One of the primary ways in which our textual structures make text easier to manipulate is by removing surface-level variations, including acronym, synonyms, label names and slight variations in sentence construction. Each of these variations introduce substantial differences into sentences, and mean that with the exception of some boiler-plate sentences and year-to-year within-firm re-use of material, almost no two sentences in company filings are identical.

This article explains how, by abstracting/capturing these surface-level variations, our textual structures make it feasible to examine the similarity in the conveyed relationships, rather than whether those sentences use exactly the same words to convey their messages.​

Components of surface-level variation
Our textual structures capture four qualitatively distinct forms of surface level variation, enabling you to focus on comparing meaningful differences between text. Specifically, our structures are designed to capture:

• Acronyms: Recognizing that CEO and Chief Executive Officer are the same job title
• Synonyms: For example that the term "works for" and "is employed by" convey the same meaning
• ​Differences in specific labels (e.g., people/organization names): Our textual structures capture organization and people names. These can then either be used to link to different data sources, or ignored (for example to compare the experiences of two different people)
• Slight variations in sentence construction: While we seek to that key dimensions of the sentence construction - such as the order that experiences are listed are preserved - the standardized representations also abstract slight permutations and idiosyncrasies in syntax that do not impact the conveyed meaning

Once acronyms, synonyms, and permutations in sentence constructions are removed, and label names (e.g., people names) are captured, comparisons can be made on dimensions of interest, ignoring label names as appropriate.

Illustrating the power of standardized textual-structures
The power of capturing information in standardized representation, abstracting surface-level variation, can be illustrated by considering a typical sentence from a manager's background, which may include details of the manager's employer, job title and dates of employment, for example:

• From April 1997 through March 2000 she was the Chief Finance Officer of Bridge Analysis

While this is one way in which the information can be conveyed, it is possible to write the very same information in multiple different ways, sometime using the same words, and some using different words:

• She worked as the CFO of Bridge Analysis from April 1997 through March 2000.
• From April 1997 she was the Chief Financial Officer of Bridge Analysis until March 2000
• Between April 1997 and March 2000 she worked at Bridge Analysis as the CFO

In each case, surface-level variation distorts the ability to directly compare the sentences. As such, it is difficult, even with a small number of sentences, to directly compare whether the same position are listed in subsequent filings, whether the same words are used to describe experiences, nor to directly compare the changes.

---

Gareth Keeves received a PhD in strategy from the Ross School of Business, University of Michigan. His research includes impression formation and management in the context of corporate leadership. As the CEO of Menai Insight he is developing approaches for capturing and representing meaning for corporate governance communications. ​​

---