Language acquisition Flashcards
phonological development
In order to acquire a spoken language, the brain has to learn to perceive the patterns of sound combinations that constitute words. The brain also has to learn the individual sound elements that comprise particular words in a given language – and there are over 6000 world languages. This process is called phonological development. Phonology is the inventory of the sound system of a language, comprising knowledge of the sounds themselves and the specific patterns or regularities by which sounds in words can be organised. The mechanisms that the brain uses for phonological learning appear to be general cognitive mechanisms like statistical learning. Because speakers are different from each other, there is high variability in the speech signal as an acoustic signal. A word like “bat” spoken by an adult male will be physically different from the same word spoken by a five-year-old girl. Linguistics gets round some of these differences by calling the individual sound elements that make up words in a language “phonemes”. Phonemes are an abstraction from the physical acoustic stimulus, and are defined in terms of changing meaning. A word like “bat” differs from a word like “pat” by a single phoneme, /b/ versus /p/. In order to produce these two stop consonants, we vibrate our vocal chords and then obstruct the air flow of the sound that we make using the lips. At a certain point, when someone is producing these sounds, we stop hearing the phoneme /b/ and begin hearing the phoneme /p/. For adults, the point at which you stop hearing a /b/ and you begin hearing a /p/ sound is very abrupt, hence this is called categorical perception.
Eimas et al (1971)
Remarkably, in a classic study by Eimas et al (1971) infants showed exactly the same categorical perception as adults even as young as one-month old using a dishabituation of sucking measure. They created a physical continuum from /b/ to /p/ in which voice onset time incremented in 20 ms steps. For adult listeners, the category boundary was between 20 ms and 40 ms. Hence when a stimulus with a voice onset time of 20 ms was compared to one with a voice onset time of 40 ms, the perceptual experience changed from /b/ to /p/. However if a stimulus with a voice onset time of 0 ms was compared to one with an onset time of 20 ms, this was perceived by adults as the same perceptual event, namely /b/. Babies demonstrated the same.
Categorical boundaries
Given that the infant is born without any knowledge of which language they will be expected to acquire, the brain needs to be sensitive to all possible phonetic discriminations that might be needed. Categorical boundaries that are not needed for a particular language could then be discarded. This appears to be what happens during the first year of life. Again using sucking paradigms, it has been shown that whereas young infants (below around 6 - 9 months in age) are still sensitive to phoneme divisions that are not used in their language, by around a year this sensitivity has disappeared, as shown in the study by Werker and Tees (1984). Part of the perceptual learning underlying this change is the “magnet effect”. Infants learn the acoustic features that typically occur together, and construct a prototypical phoneme, eg a prototypical /b/. Non-prototypical sounds will then be perceived as more similar to the category prototype than they are to each other, even though the actual physical difference between the stimuli may be equal. The magnet effect reflects experience with your specific native language/s.
semantic development
The primary function of language is communication. Infants need to learn what words mean, so that they can be equal partners in linguistic interactions: semantic development. This requires them to make mappings between sound patterns and specific concepts and events in the everyday world. Learning word-object mapping relations is supported for infants by other adult behaviours. For example, adults spend a lot of time naming things for infants, and usually they combine naming with pointing to the novel referent. Babies learn this connection, and accordingly they give greater attention to objects that are both pointed to and labelled. With the development of intentional understanding (Lecture 3), infants’ own behaviours such as seeking joint attention events with caregivers create more opportunities for semantic development.
overextension’
Given their limited language abilities in the first two years of life, most children go through a phase of ‘overextension’. Overextension is using a single word to apply to many different referents, such as using ‘ball’ to name apples, grapes, bell clappers, and also balls. Again, this language-universal phenomenon was originally thought to show limited conceptual abilities on the part of children. For example, it seemed possible that a child who called an apple ‘a ball’ was unable to distinguish apples from balls. Systematic experimental work showed this idea to be wrong. Children never overextend in language comprehension, only in production. Overextension is a by-product of stretching a limited vocabulary to communicate as flexibly as possible during the one-word stage. In fact, the first words that are typically used by young children are very similar across languages and cultures. Young children first talk about salient objects and events in their day-to-day lives: family members, pets, words connected to routines like meals and bedtime, and words referring to the movement of people and objects.
special mechanisms of acoustic learning
As spoken language is a sound signal, there are also some special mechanisms of acoustic learning. Human language also has a unique teaching mechanism, called infant directed speech (IDS) or “Parentese”, which has a special exaggerated prosodic register. IDS is found across cultures, genders and ages. Parentese or IDS helps children solve the problem of children knowing which part of the array in front of them they should attach a word to - to learn mappings between words and their referents. For example, Fernald showed that mothers unconsciously emphasise novel words when they are reading stories to their infants. Extra word emphasis was not observed when reading the same story to another adult.
fast mapping
behaviours, caregivers report that infants can also learn a new word just by over-hearing it during adult conversation. The idea that a single experience of a new word is sufficient for learning is referred to as “fast mapping”. Some have interpreted the phenomenon as evidence that human brains are specialised to acquire language. How might we measure fast mapping and test its underlying mechanisms? One question is whether fast-mapped words are truly learned and encoded in memory – if so, they will regularly appear in the child’s language after the initial learning trial. Several studies have assessed this question, all of which adopt a similar procedure: children are presented with sets of novel objects, one of which will be named by the experimenter (“e.g. “This one is Koba”) for one set of children and the same exemplar will be introduced with an interesting “fact” – e.g. “this one is special because it was given to me by my uncle” for a second set of children. Immediately afterwards, they receive a larger set of objects containing several new examples of the training set, plus the original object either named or given a fact and assessed for whether they identify it as Koba or with the fact. This is repeated after a week delay to assess any retention of the word or fact. Far greater accuracy was observed in children in the Koba label group at both time points than in the group given a fact. These studies purport to have observed fast-mapping in the laboratory (e.g. Waxman and Booth, 2000) and concluded the presence of an innate mechanism.
Criticism on Koba experiment
However, a number of procedural elements to these studies challenges this idea. Real-world observations of fast mapping refers to the repetition and retention of a word after a single (or maybe two or three) exposures. In studies however, procedures allow for elaborative encoding, multiple exposures and repetition by the infant. In an experiment designed to vary encoding conditions, Vlach and Sandhofer (2012) observed that retention of a word exposed only once showed a classic Ebbinhaus curve of forgetting in both children and adults. They were also able to replicate previous studies reports of slower rates of forgetting by systematically varying levels of memory (encoding) supports. These findings suggest that real world mapping may reflect an overinterpretation on the part of caregivers of young children’s language abilities and that rapid acquisition of new words is probably supported by enriched encoding provided by caregivers (see Parentese). It is a learning phenomenon.
Syntactic or grammatical development
Syntactic or grammatical development encompasses learning the set of grammatical ‘rules’ that determine how words can be combined into sentences and phrases in a given language, and learning the set of ‘rules’ governing the internal structure of words (morphology, for example, we can say “burglary” but not “stealery”). The basic unit of morphological analysis is the morpheme. A word like “walk” is a single morpheme, but the word “walked” comprises two morphemes, since the additional /ed/ conveys that this action happened in the past. This is called inflectional morphology. Morphemes like –ing and –ed are added to verb stems to convey information about actions and timing. Morphology can also be derivational, as when we create new words by adding affixes (“un” + “happy” = “unhappy”) or suffixes (“make” + “er” = “maker”). All of these grammatical rules appear to be acquired by young children on an implicit basis, by listening to the adults around them.
Language acqusition device
Could these rules possibly be learned piece-meal through operant conditioning, rewarding correct structures and punishing incorrect ones? The famous philosopher Chomsky wrote a searing critique of BF Skinner’s book “Verbal Behaviour” (1957) to argue that these kinds of linguistic rules appear to be highly abstract, and therefore require a special innate neural architecture in order to learn them. Furthermore, he argued that caregivers do not provide the consistent and accurate feedback to their children’s emerging grammatical experiments (the “no negative evidence” argument) and that the input they received is not adequate to account for the sophistication of grammatical strings generated so early in life (the “poverty of the stimulus” argument). His theory was that because children do not exhibit “wild” grammars during early language use, there must be a constraint on the kinds of grammars that can be generated. He referred to this as the “Language Acquisition Device”, an innate mechanism that filters possible grammatical structures of the language you are exposed to. He famously said “language grows like an organ”.
phenomena in language acquisition that suggest that children do appear to learn rules about the grammar they are hearing
However, there are phenomena in language acquisition that suggest that children do appear to learn rules about the grammar they are hearing. One is the U-shaped curve seen in production of regular and irregular word endings for the past tense, suggesting that the child initially learns individual instances of past tense word ending rules irrespective of the regularity of the word, and then from those extract a rule – which is a kind of average of the set of experiences the child has had of past tense endings which on the whole will be the regular form. At that point, their production shows errors but then over time they appear to learn the rule that there are exceptions for irregular words
alternative to the innate view is “usage-based” theories of grammatical development
So the alternative to the innate view is “usage-based” theories of grammatical development. These argue that these rules can be acquired via the experiences that infants receive when they are having conversations with others. For example, Thomasello argued that infants simply copy the utterances they hear around them irrespective of whether the utterances are directed towards them or someone else. So whilst Chomsky was undoubtedly correct that a strictly behaviourist Skinnerian view of language acquisition based on systematic reward and punishment of the child’s verbal behaviours could not possibly account for the nature of syntax development, what he overlooked was the fact that quite complex abstract rule-based systems could be learned implicitly by neural networks through exposure.
connectionist networks
You will have come across connectionist networks in Dr Bozic’s lectures on language, and will remember that connections within a network are formed and adjusted based on the statistical probabilities of the input and over hundreds of trials the network will create patterns of associations between input and output that have been reliably experienced with exposure. Such a model is able to demonstrate the possibility of a U-shaped curve in development. Because of the frequency of use of regular compared to irregular forms in everyday spoken language, the network will receive greater exposure over time to regular forms than to irregular forms. Of course, at the beginning, both regular and irregular words are relatively new and so each type of form is learned at roughly the same rate. However, difference in relative exposure to regular and irregular forms will increase with time and the greater exposure to regular forms will enable rapid acquisition of the regular rule relative to the irregular rule. But once both asymptote, both sets of rules have been learned.
