This blogpost is the first in a series that explores some of the ideas about language acquisition put forward by cognitive psychologist Steven Pinker in his 1979 article Formal Models of Language Learning, published in the journal Cognition, where he persuaded his peers in the field of developmental psycholinguistics to consider the value of formal language models.
Pinker began by addressing the ‘big question’ of developmental psycholinguistics: How do children acquire language?
This is a particular case of a more general and fundamental question in cognitive science, the puzzle of induction: How do humans form (valid) generalisations about the world despite only limited evidence?
Understanding how humans reason by induction is of great importance because we do it all the time. I would even hazard to claim that most human knowledge is gathered by induction.
In the case of language acquisition, the generalisations that we make are about our native language. They are based on our exposure to that language during early childhood, and they allow us to produce and comprehend it.
To help us answer the question of language acquisition, Pinker highlighted six empirical conditions which an ideal theory of language acquisition must fulfil. Though some of these conditions may seem trivial, they are nonetheless worth stating.
The Learnability Condition
The theory must propose learning mechanisms by which any child can acquire language, which indeed all normal children manage, despite its sophistication. Pinker contrasts language with other complex skills, such as chess and calculus, which not all children learn.
The Equipotentiality Condition
The learning mechanisms proposed should not be restricted to any single language. All human languages are learnable.
The Time Condition
The child must be able to use these mechanisms to acquire at least the basics of their language within an acceptable time limit – about three years.
The Input Condition
The mechanisms can only require the child to use kinds and quantities of information that we know they can use.
The Developmental Condition
The theory should agree with the stages we observe in the acquisition process and should explain them – at least generally.
The Cognitive Condition
The theory must agree with what we know about a child’s cognitive abilities. Reasonable limits must be placed on their perception, memory, and attention.
In summary, an ideal theory should explain how any child can learn the basics of any language in only a few years and with only the evidence and cognitive faculties available to them, and it should predict the course of this acquisition process. In his paper, Pinker pointed out that no theory of language acquisition did this (nor does any, even to the present day). Most accounts instead focused on only one of these conditions.
At the time of the article’s writing, most psychological research on language acquisition was focused on the Input, Developmental, and Cognitive Conditions. Psychologists were investigating and trying to construct theories about:
- The language that children witness and what use they make of it (Input Condition)
- The language that children produce and what principles it abides by (Developmental Condition)
- How children’s cognitive development affects their linguistic development (Cognitive Condition)
A lot of linguistic research at the time was seeking to identify linguistic universals, a pursuit relating to the Equipotentiality Condition. Of particular note is the work of eminent linguist Noam Chomsky.
The psycholinguistic attempts to address the Learnability Condition, Pinker criticised for their vague jargon, their unhelpful descriptions of language acquisition as a simple consequence of ordinary cognitive development, and their failure to describe adequate learning mechanisms.
He instead reported on three approaches that use formal language models to address the Learnability Condition, and which give us insight into what mechanisms can be used to learn what kinds of language, and using what input.
The first approach is that of mathematical linguistics, in which we treat languages as formal objects and try to prove theorems about when it is possible to learn a language from a sample thereof, through what is called ‘grammatical induction’ or ‘grammatical inference’.
In the second approach, computer programs are designed to simulate human language acquisition, using many of the formalisms laid down in mathematical linguistics. This is the domain of artificial intelligence and cognitive simulation.
In the third approach, which builds on the first two, a group of theoretical linguists extended a learning model to acquire what are known as ‘transformational grammars’.
The benefits of formal language models
Formal models such as those used in these approaches, which focus on learnability – even if they do not satisfy all six criteria – generally show great potential because (1) the Learnability Condition is strict and (2) formal models are explicit.
The Learnability Condition is strict
Although developmental psycholinguistics had been focused on describing the stages involved in the acquisition process (Developmental Condition), Pinker said that merely demonstrating that language can be acquired in the first place (Learnability Condition) might put us in better stead to ultimately come up with a complete theory of language acquisition.
A child’s linguistic output does not give us enough insight into the mechanisms involved in their acquisition of language, because the child’s linguistic abilities are developing in parallel with their general cognitive abilities. However, according to Pinker, determining whether a theory satisfies the Learnability Condition is relatively straightforward because of the condition’s strictness.
Formal models are explicit
Formal models force us to state things clearly. They would – said Pinker – help to clarify issues and answer questions like:
- What prior knowledge is required to learn a human language?
- Do different sorts of input make acquisition easier or more difficult?
- How does semantic information affect the acquisition of syntax?
Answers to these questions had evaded experts because little was known about the mechanisms involved in language acquisition. Additionally, terms like ‘semantic information’ were not formally defined.
Pinker pointed to other areas of cognitive psychology, where mechanistic theories had advanced our collective understanding of long-term memory, visual imagery, and problem solving.
So, what are these formal models that Pinker so espoused? That, my friends, is for the next blogpost, which will provide an introduction to mathematical linguistics.