Language (2) - Recursion Flashcards
Explain Hockett’s return
- The design features of language that seem to be most absent from animal communication seem to be productivity and duality of patterning
- why do chimpanzees appear to be unable to learn these aspects of language when they seem to be able to cope with semanticity and displacement?
Who did/coined the Nature VS Nurture debate?
Nurture = Skinner, (1957)
Nature = Chomsky, (1957)
Explain the Nurture side of the debate in relation to language(/and thought)
- Skinner’s 1957 book Verbal Behaviour suggested children learn language in exactly the same way they can learn any other behaviour
- the principles of operant conditioning governed language learning: stimuli in the environment led to language users producing linguistic responses
- Essentially children learn to associate their linguistic productions (speech) with rewards of some kind e.g. the child says “more milk” and they are rewarded with a drink
- in this approach children are either actively instructed by others or learn via imitation
What are the Pros and Cons of a Nurture only approach?
PROS:
- Children do learn language (in that they need to be exposed to it)
- certain aspects of language need to be learned i.e. irregulars such as some plurals can’t be inferred
CONS:
- Children produce their own languages (creoles) in certain circumstances (Bickerton, 1991)
- children make errors that they may never have heard before e.g. Berko, (1958)
What is meant by (a) finite state grammar?
- (In addition to Skinner’s attempts) Blomfield, (1933) and Miller, (1958) proposed a learned rule system that could explain English
- based on a NEUROLOGICAL ASSUMPTION (the brain is finite in size and its neurons have a limit on their maximum firing rate - this implies that there must be a finite number of functionally distinct states that the brain can occupy)
- LINGUISTIC ASSUMPTION: (sentences consist of strings of units - words/morphemes. These units are produced and perceived sequentially. Hearing or syaing a word changes the listener’s/speaker’s state of mind from Si to Sj - where Si and Sj are 2 of the finite states the mind can occupy)
- Essentially a finite state grammar is a chain of words
- we learn to associate that a limited range of words will follow on from the word that precedes it
- e.g. in the sentence “Please pas the ___”, then we can predict what the word is likely to be simply from having learned that the word salt follows “Please pass the…” more often than iguana would do
What are some problems of Cognitive Economy?
- If a child learned the links between words every second it would still take that child 30 years to learn the vocabulary and sentences that an average adult speaker can produce (there are estimated to be over 1 billion associative links necessary for adult speech)
- children are relatively sophisticated speakers by 4 years old and almost adult in their speech by 12 years of age
What is a problem of Infinity?
- Main problem faced by a finite state grammar is that language is infinite even though our brains are finite
- theoretically we can produce an infinitely long sentence via ‘iteration’ - repeating 1 component e.g. “My great, great, great,…insert infinite symbol…grandmother was Scottish”
- a finite state grammar can handle this by having a system for looping
What is another problem of Infinity?
- We are able to understand sentences that we have never heard before
- if we have never encountered a sentence before then how can we have possibly learnt it? In FSG terms how do we know the statistical likelihood of the words following one another in a sentence we’ve never heard before?
What is the biggest problem of Infinity?
- We are able to embed sentences 1 without the other (this is known as RECURSION), without disturbing the structural relationship between the words in the original sentence
How does Recursion work/
- Chomsk’ys career has been spent investigating the nature of language
- his attempts to explain the productivity of language have been confined to the search for a set of rules (grammar) than can describe how all language are structured: the so called ‘universal grammar’
What was Chomsky’s view of language?
- Learning language is about learning not only the meaning of words but also about learning the functions of words and a set of rules that allow you to combine those words according to the rules of language
- E.g. ‘John kissed Jane’
- means something different from:
- ‘Jane kissed John’
- even though the words are the same in each sentence the order is different
- Chomsky (1980) suggested language is best thought of as comprising 2 separate but interacting systems:
1. A Computational System: - This is concerned with syntax (structure of words/sentences etc.) and determines how words that perform different functions can be combined into sentences (this is what Chomsky means by language)
2. A Conceptual System: - This is part of a general cognition (what is stored in our long term memory). This isn’t part of language but instead contributes meaning (semantics) to language
- Chomsky suggested that children possess an innate ‘language acquisition device’
- they still have to learn language but what they are learning (in addition to the meaning of words) is the set of rules that govern their language
- also suggested that the language children hear is ‘too poor’ to allow them to learn language ‘from scratch’ (poverty of stimulus argument)
What is generative grammar?
- Various versions of Chomsky’s theories suggested that a sentence is composed of phrases, which are in turn composed of words according to what the words do in the sentence
- even in nonsense sentences we are able to identify structure
Why is simple best?
- Because the rules that underline language must be simple enough for children to learn in about 4 years (cognitive economy)
- in order to describe rules that describe the underlying structure of all languages (a universal grammar) then the more general a rule the more likely it is to describe a number of languages
Explain finite state vs phrase structure (generative) grammars
- Fine state grammars assume that language learning is a process of learning which words follow which other words (statistical regularity) - this is all that there is to language, there is no other structure apart from words
- each sentence is seen as a chain of statistical probabilities which we learn
- phrase structure or generative grammars assume that language is compositional - words can be broken down into smaller units and combinatorial words/morphemes can be built up into larger units (phrases and sentences)
- phrases and sentences can be embedded within one another to create infinitely long and structure sentences (Recursion)
- this is possible because of the compositional/combinatorial nature of language (it isn’t flat out)
- sentences are structured hierarchically
Why can’t cotton top tamarins ‘do’ recursion?
Fitch and Hauser, (2004):
Aim:
- Tested ability of cotton top tamarins to discriminate between novel sequences of sounds arranged either according to the rules of a finite state grammar of a phrase structure grammar
Procedure:
- The tamarins were divided into 2 groups, each of which were exposed to either a finite state or a phrase structure grammar
- the tamarins received 20 minutes of playback of 60 different grammar-consistent strings in a random order
- the next morning they were tested individually and looking was measured (idea that being that the monkeys will direct their gaze towards a novel stimulus)
Findings:
- Monkeys exposed to the FSG looked at speakers playing grammars that violated the rule significantly more than the speakers that didn’t
- no such significant difference was found in the case of the monkeys exposed to PSG - suggesting that the monkeys were unable to decipher the rule underlying the sound sequence (they didn’t look at violations because they couldn’t understand that the rule was being violated)