In Noam Chomsky’s framework, data (e.g., utterances, grammaticality judgments) are seen primarily as evidence for an underlying mental system: the language faculty (I-language.

“Linguistic data are not the object of inquiry; they are evidence for the nature of the underlying generative system.”
— Noam Chomsky

So:

• Data ≠ the thing to be modeled
• Data = clues to a hidden structure (Universal Grammar)

This sets Chomsky apart from empiricist models that treat data as central and sufficient for understanding language.

Chomsky famously distinguishes between:

• Competence:

  The idealized, internal knowledge of language

• Performance:

  The messy, real-world use of language (which produces data)

Data from performance (e.g., speech corpora) is considered noisy and incomplete, and not always reliable.

Linguistic data, by themselves, are not enough.

Chomsky argues we need theory guided interpretation: you cannot "learn" language just by looking at statistics of usage.

View	Chomskyan Linguistics	Statistical Language Models
What is data?	Evidence of internal, rule-based grammar	Raw input for modeling statistical patterns
Trust in data	Limited: messy, needs interpretation	High: more = better
Goal	Explain mental language competence	Predict/imitate language behavior
Theory	Necessary, innate (Universal Grammar)	Emergent from data
Famous critique	“Of course you can make a model that mimics behavior. But does it explain anything?”	(Counterpoint) Explanation isn't always needed for practical success