L20 - Word Segmentation Flashcards
What type of phonemes can infants hear?
All phonemes up until 8 months
How do infants learn to distinguish between phonemes?
Statistical learning
The sounds they hear fall on a distribution and they pay attention to the important ones
Definitions: What is a ‘Syllable’
A set of phonemes
What are the 3 components of a syllable?
Which is most important?
Onset
Nucleus
Code
The nucleus is most important - which is the cluster of vowels in the middle
What is the ‘nucleus’ of a syllable?
The vowel/cluster of vowels in the middle
What is a ‘morpheme’?
Can convey meaning but unlike a word, they do not stand on their own
What is a ‘morpheme’ that can stand on its own considered to be?
What if it depends on another morpheme to convey an idea
It is considered to be a ‘root’ word
When it depends on another morpheme to convey an idea it is considered a ‘affix’ because it has gramattical function
(i.e. s to give it a plural function or ‘ed’ to show something that was done in the past)
This is the correct segmentation for where the different words start and finish - true or false
False
there are no silences between words
Can you distinguish words by looking at speech on a recording?
No, the pauses in words are not enough to segment words
Do parents teach children words in isolation? (or with multiple words)
With multiple words, but they pronounce it slower than they would with adults
How do we do word segmentation?
1. We use ‘constraints’
I.e. general rules that we all follow
2. Transitional probabilities
I.e. learning through statistical learning
What are the two types of ‘constraints’ that we use for word segmentation?
Phonotactic and Prosodic constraints
What are Phonotactic constraints?
Restrictions on where phoneme and syllables are allowed to be in English
What are Prosodic constraints?
Rules about stress, pitch and intonation
E.g. mothers of infants raise the pitch of their speech to highlight important words
Are language constraints universal?
No, culture-specific
What are ‘transitional probabilities’?
Language ‘rules’ regarding where we can identify where word boundaries are
Learning through statistical learning
Saffran, Aslin and Newport (1996) - explain the experiment methodology
- A group of syllables are repeated in a repetitive cycle
- Other syllable pairings are presented afterwards and they have to identify whether they remember the pairings or not
- They measured ‘listening time’ of the children to determine the unfamiliarity of the syllables. They measured how long they stared out the speaker
- If they are presented with a word that is unfamiliar they attend for longer - and are able to differentiate novel vs repetitive syllables
What were the results of the Saffran, Aslin and Newport (1996) experiment?
Children can identify new syllable combinations with great accuracy - better than adults
They can understand what is likely to be a word and what is unlikely to be a word
Explain how the sound (word) ‘bay’ uses transitional probabilities
It can be at the start of a word (baby, bacon) or at the end of a word (obey, bay)
If you saw it with other syllables that it commonly isn’t paired with (baytwo, baymun, bayme) you may assume there is a word boundary there.
We will hear sounds within a word together, rather than sounds across boundaries
We unconsciously keep track of these statistical probabilities and use this to identify which words should be segmented
Three words are repeated over and over in a random order with no cues (monotone, pauses etc.) to participants. What is this transitional probabilities test showing us?
Shows that words with pho/ment or seg-lity are more likely to be identified with a word boundary, rather than seg-ment which was found to have 100% success in identifying when listening.
We learn with what fits and what ends with the instances we are presented with.
What were the results of the Saffran, Newport, Aslin (1996) experiment with adults, where they gave them an artificial language with 6 words (with 3 english syllables) and the only cue to word segmentation was transitional probabilities?
Adults did significantly better than chance at
- Detecting words versus non words
- Detecting part words versus whole words
Adults are capable of using statistical regularities in a language learning task (we have the ability to use word boundaries fast, based on transitional learning)
Experiments on 8-month-old infants regarding transitional probabilities (where they were shown repeated words in random order for 2 minutes) showed what?
Infants had a novelty preference for ‘non-words’
They looked at new sounds longer at the new pairings of syllables, showing that they could learn quickly
In the second experiment infants were tested on “words and part words”. What were the part words?
Words that start with the end of one syllable and go onto the next
What were the infants transitional probabilities for the part words in comparison to the whole words?
.5 in comparison to .1 for whole words
They heard the part words only half the time, it is more difficult than words with a transitional probability of 0 (whole words)
Infants can tell words from non-words accurately from as young as 8 months. True or False?
True
Infants will look longer at people speaking a foreign language, true or false?
Why?
True
because infants love novelty and they do not understand the language being used.
“Transitional probabilities help infants and adults _____ words in speech”
Segment
How well can we use computational models to calculate transitional probabilities (percentage)?
52- 70%
Where do computational models tend to get transitional probabilities wrong?
They see the letter ‘d’ as its own word
It thinks that ‘lookat’ is one word
‘Isit’ its own word
Computational models get things wrong in a different way than infants when they use transitional probabilities.
True or false
False - infants tend to make transitional probability errors in the same way that computational models do
Transitional probabilities are…
The probability that one syllable will follow another
Are transitional probabilities higher between words or within words?
Within words
Because we hear sound pairings within words more frequently than across boundaries