I want to rate a sentence by its complexity in the sense of:

  • Rating of 1: A very simple sentence which is just S+V+O, example: "I eat bananas."
  • Rating of 10: An uterly complex sentence with lots of subordinate clauses that is really hard to read through:
    "This very superficial grammatist, supposing empty criticism about the adoption of proper phraseology to be a show of extraordinary erudition, was displaying, in spite of ridicule, a very boastful turgid argument concerning the correction of false syntax, and about the detection of false logic in debate."

What metrics did the research found to identify the difference in complexity between sentences?

Some (naive) metrics that I mean could be:

  • Word count in a sentence.
  • Number of verbs in a sentence.
  • Number of subordinate clauses.
  • 2
    In fact, if you build a Syntax tree, you can measure it by various metrics, e.g. number of nodes and depth of the tree built. Then you may adjust it assigning various weights so that, for example, gone with the wind appeared less complex than someone who studies well, regardless of both phrases contain 4 words. – bytebuster Aug 14 '14 at 16:00
  • 1
    That's a huge topic. See this to start with: en.wikipedia.org/wiki/Readability_test – prash Aug 14 '14 at 17:42
  • 1
    All of the metrics suggested have, of course, nothing at all to do with how difficult any individual might find hearing or reading a sentence, not how likely they might be to use it. These measures are completely artificial and do not measure any human perceptive or cognitive trait -- just bits and pieces of apparatus, with different apparatus by different suppliers. – jlawler Aug 14 '14 at 20:59
  • Is your problem to actually choose/design a metric, or is it to build a tool to measure it on actual sentences. What is the context. I am asking because I was once advising a (quite successful) PhD on the same topic, but for programming languages. – babou Aug 15 '14 at 19:32

It depends on what you are trying to achieve.

If you are trying to assess the structural complexity of any given sentence, then metrics like you suggested are useful, as is measuring the complexity of the syntax tree or using something like the Flesch-Kincaid metrics.

If you are trying to assess how difficult a sentence is to comprehend (e.g. for a specific school grade/age), there are various factors involved.

One factor is whether a sentence can have multiple valid syntax trees due to the words belonging to multiple part of speech classifications, or have different meanings within the same part of speech classification. Consider:

  • The book "All About Reading" was very good. -- to read, or the town?
  • "The wind up the valley was strong." vs "The wind up toy was old."
  • "She took the lead out of the box." vs "He took the lead in the dance." -- also, the first is ambiguous: the metal, or a dog lead? You need to know the context from the previous sentence(s).
  • "The mouse was getting old, so they replaced it with a new wireless one." -- here the context for 'mouse = computer mouse' instead of 'mouse = animal' is only determined at the end of the sentence.

Another factor is whether it uses specialist words in a given domain (including archaic words), or meanings for the words. The more of these there are, and the more obscure outside the given field they are (e.g. electricity vs quantum chromodynamics), the more complex the sentence is to understand. The same goes for abbreviations, slang, accentisms ("Aye, tha' be t'one ah war lookin' fer."), or even pulling in words from other languages ("That is no bueno.").

Another factor is whether the sentence uses non-standard word ordering, either for poetic effect or for something like Yoda speak ("Go, you must.").

Another factor is if the sentence uses the incorrect form of a word (its vs it's) or the wrong homophone (their vs they're vs there) or the wrong word (e.g. via spell checkers or predictive texting for technical or uncommon terms).

Another factor is mixing or using a different spelling (British English, American English, text/SMS, leet (e.g. l33t), archaic, ...) to what the reader is used to.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.