5

I am not a pure linguist but rather at the intersection of computational linguistics, NLP and computer science. Thus please be cautious with me and my ignorance. I am looking for definitions of sentence complexity - possibly with respect to information / entropy. For example, a matrix sentence like The washing machine which is on top of the shelf is running is clearly more complex than The washing machine is running. So one naive way of defining one sentence is more complex than another would be counting how many sentences are recursively incorporated. Another way would be counting the length of the sentence. My question would be, is there any formal work on defining this more thoroughly? I would very much appreciate any hints!

I found these things so far, writing this to make sure references don't repeat and for own documentation purpose:

StackExchange questions:

Measuring semantic complexity of a text

Measuring syntactic complexity

Semantics:

Semantic entropy

Lexical Density (Chapt. 5 and 6)

Syntax:

Dependency distance as a metric of language comprehension difficulty

Syntacic Complexity

Some syntactic determinants of sentential complexity

Some syntactic determinants of sentential complexity, II: Verb structure

On Operationalizing Syntactic Complexity

Morphology:

Measuring syntactic complexity (here, curiousdannii suggests to look at morphology. I haven't seen anything on it. I wonder if morphology is relevant considering I am only looking at English.)

Misc:

Curriculum learning for language modelling (Sec. 3)

Readability

A model for language structure

Improve this question
4
  • 1
    Some time ago I used Laurence Sterne's novel The Life and Opinions of Tristram Shandy, Gentleman to test the level of my own adequacy before going to audiences with the headmaster of the school where I worked — the sentences in the novel are often a paragraph long and are so paradoxical and mind-twisting that understanding some was a real challenge. On the other hand, Michael Moorcock in his novels could have a page-long sentence enumerating in a list the types of planes that were simultaneously bombing Berlin during WWIII. Two kinds of sentence complexity, which one do you prefer?
    – Yellow Sky
    Commented Oct 27, 2021 at 22:50
  • 3
    I suggest checking out the concept of and work on dependency distance as it is understood in dependency grammar (DG). Dependency distance is calculated by counting the number of words separating a given word from its head word. One calculates the dependency distance for each word in a sentence and then arrives at an average. The greater this number, the greater the syntactic complexity and thus the more difficult the sentence is to process. Do a google search for dependency grammar dependency distance. DG is of course quite prominent in NLP circles.
    – Tim Osborne
    Commented Oct 28, 2021 at 10:23
  • @YellowSky: I am not sure. The first one seems more complicated to me, though concepts are related sequentially. But we have a weird jump back to the first concept (the novel) in the end of the sentence. The second seems more readable. The thing is. It doesn't really matter, which I like, because I am looking for sentences which a machine considers as complex - and that is very intransparent if we consider machine learning models. So my take here is to look at various concepts and do trial-and-error.
    – Marcel Braasch
    Commented Oct 29, 2021 at 16:59
  • @TimOsborne: thanks for your suggestion, I will have a look!
    – Marcel Braasch
    Commented Oct 29, 2021 at 17:00

2 Answers 2

Reset to default
4

There are early works on the topic in Some syntactic determinants of sentential complexity and Some syntactic determinants of sentential complexity, II: Verb structure. These focus on "perceptual complexity". This much later paper (and references therein) has a very different approach but seems to equate "complexity" with "psychological difficulty". This 1974 paper on the other hand takes a mathematical linguistic stance, and aims to compute complexity, but I can't discern an actual definition of complexity. Complexity enters into the discussion of this classic 1959 paper, which measures complexity in terms of depth of postponed symbols, but also takes for granted what complexity is. All told, it seems that people in the field equate psychological difficulty with "complexity". "Complexity" is generally not a consideration in generative syntactic theories, so definitions most likely will come from another discipline.

Improve this answer
6
  • Thanks a lot, I will look at all of these!
    – Marcel Braasch
    Commented Oct 28, 2021 at 0:09
  • So it is more of a semantic and cognitive question?
    – Marcel Braasch
    Commented Oct 28, 2021 at 0:14
  • 1
    Since syntax is more of a personal matter, abstract "complexity" will vary wildly depending more on individual factors than structural ones. What's simple for you may be complex for me, and vice versa.
    – jlawler
    Commented Oct 28, 2021 at 19:21
  • @jlawler: but there must be a universal notion, no? Assuming we are both as literate and have the same knowledge. Or even if we assumed we didn't, then some sentences are harder than others, and however we order this likely coincides. At some point of the scale for someone illerate distinguishing a super complex and a pretty complex sentence might be hard, but I am somehow presuming an ordering with respect to some concept must be possible.
    – Marcel Braasch
    Commented Oct 29, 2021 at 17:03
  • 2
    My structural definition of "complexity" is purely a count of nodes and relations. Usually people equate complexity with difficulty, which raises the question "difficulty in doing what?". You seems to equate "complexity" with "hardness" (I don't): so no, there is no universal definition. I urge you to get rid of "complexity" and focus on what seems to be most important for you, namely "difficulty in doing ___" (producing? understanding? reading? pronouncing? remembering?).
    – user6726
    Commented Oct 29, 2021 at 19:01
2

In the CAF-framework (Complexity, Accuracy, Fluency) used in language teaching syntactic complexity is measured in terms of words per unit, where the unit can be the complete sentence, a clause, or a T-unit. The measure is very practical and applicable, and it seems that other measures don't add much to that.

See also this question on languagelearning.se

Improve this answer
1
  • Thank you, I will have a look at the framework!
    – Marcel Braasch
    Commented Oct 29, 2021 at 17:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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