Problem Addressed

• Measurement of the lexical quality of the Web.
• Lexical quality the representational aspect of the textual Web content, the quality degree of words in a text (spelling errors, typos, etc.).
• Lexical quality is not presented as an accessibility metric but it is useful, since the quality of words and language impacts the readers understanding.
• WCAG principle of content being "perceivable" and "understandable".

Strategy

The output of the p(d) function changes depending on the number of elements to measure and d, which is the degree of disjunction.

• Defining a lexical quality metric based on the different kind of errors that appear in the Web:
• Kind of errors. Sample W (1,345 errors):
  • regular spelling: *toomorrow.
  • typographical: *tomorroe.
  • non-native speakers: *tomorow.
  • dyslexic: *torromow.
  • optical character recognition (OCR) errors: *tornorrow.
• Lexical quality metric:
  • We measured a lower bound of the fraction (f) of Web pages with lexical errors and the relative fraction (d) of each kind of error in the sample W.
  • The corresponding fraction of Web pages with lexical errors is then (f x d).
  • To compute LQ, we estimate (df) by searching each word in the English pages of a major search engine.
  • We use data from a leading search engine to estimate this value.
• LQ = mean_{wi ∈ WM} (df _{misspell wi}/df _{correct wi})

Outcomes

• Although the lexical quality measured will vary with the set of words chosen, the relative order of the measure will hardly change as the size of the set grows.
• LQ provides independent information about the quality of a website.
• Pearson correlation with the following measures for the top 20 sites in English of Alexa.com: Alexa unique visitors, number of pages, number of in-links, and ComScore Unique Visitors.

  Table 1: Pearson correlation for several measures in the top 20 English sites of Alexa.com in March 2011.

  Measure	Alexa	Pages	Links	ComScore
  LQ	0.4451	0.4167	0.3966	0.2356
  Alexa		0.7659	0.6897	0.6589
  Size			0.8655	0.3097
  Links				0.1319

• We applyed our methodology to several large Web domains, the major English speaking countries and the major and social media websites. Expected domains have better lexical quality (USA and UK universities and goverment) and media websites have worse lexical quality than the Web average.