What children experience, which words, containing which sounds, what they happen to hear said, what sorts of conversations they get involved in, all vary randomly.

The variation is not just in the forms, but in what defines the forms as forms. For instance, the CH sound in chew might be resolved as either a T followed by an SH sound or as a complex sound with a complete closure turning into a partial closure. Similarly, the ST in stun and sting might be resolved as either a single  sound with the closure of the vocal tract reduced in the course of articulating it or as two separate sounds, the first with a partial closure, the second with a complete closure. In both cases, the second analysis happens to work better (for reasons beyong the scope of this article). But the first language learner has to have a way of coming to this realisation on his or her own, or languages would split into sub-dialects, each defined on a different combination of the various analyses. Despite the infinite variation in what children hear, in any given speech and language community, children all end up speaking the same language. But how? This mismatch between the variable input and the common output leads to what is known as ‘the logical problem of language acquisition’.

Every language has a way of referring to the various facets of experience, woman, child, mother, water, death, night, silence, known as ‘nouns’. But the way they appear varies widely, mainly in some sort of root form, as in English, or only with sort of additional structure as in Greek and many other languages. Most European languages add an element with a very subtle role defining its relation to the discourse, traditionally known as ‘articles’, but nowadays as ‘determiners’. These additional structures vary greatly from one language to another.

Consider the seemingly simple sentence “He likes sugar”. By what was once known as ‘Do support’ the marking of the present tense is moved to the left in questions and negatives, and realised in a form of the verb do in “Does he like sugar?” or “He doesn’t like sugar”. This way of marking questions and negatives is unusual across languages, and is accordingly hard for children to learn.

There are similar problems with the sound structure. All languages allow syllables with one vowel and one consonant like key and toe, where the vowel is the nucleus. Some languages allow no other sorts of syllable. English, as well as allowing words like strength, with three sounds before the vowel or ‘nucleus’ and three after it, also allows L as the nucleus of an unstressed syllabie in words like little. Getting in the way of a correct analysis of words like little, the L is commonly pronounced with a degree of lip rounding like the vowel in hook and hood. The degree of lip rounding varies from one variety to another. The complexity here is plainly hard to learn. Many children go on confusing the tongue tip articulator of the T in little and of the D in middle until they are four or so.

In a more exteme way, at least one language, Tashlhiyt, spoken in North Africa, sometimes called ‘Berber’ to the great dislike of some speakers, allows any phoneme to constitute the nucleus of a syllable. It thus allows sentences consisting entirely of consonants, without a single vowel.

There is thus a scale from the easily learnable to the not-so-easily learnable to the vdery hard to learn. These points appear to be distributed somewhat randomly, as though languages are pulled simultaneously in opposite directions, in one direction towards simple and easy-to-learn consistency with simple sound structures,  and in the opposite direction towards precise, well-defined meanings with randomly variable forms and sound structures.

This learning happens in what is necessarily a ‘learnability space‘.