How Babies Resolve Segmentation Problem

Discriminating Words

fluent speech doesnt provie unambiguous indications of where words begin and end

• learning a language involves figuring out which sounds clump together to form basic units and learning how these units in turn can be combined with other units

• In talking to their babies, parents rarely speak to their children in single-word utterances (about 10% of the time)

• babies are confronted with speech in which multiple words are sewn seamlessly together and they must figure out all on their own where the edges of words are.

Strategies to Overcome Segmentation Problem

• the isolated words hypothesis

• phonotactic constraints

• the prosodic bootstrapping hypothesis

the isolated words hypothesis

• Bortfeld et al 2005: at 6m familiar words help infants map out where other words start and begin

• use words like their own name or commonly heard words as anchors to segment surrounding speech

• generate hypotheses about which adjacent clumps of sound correspond to word units

phonotactic constraints

• using intuition of which sound sequences are allowed at the beginnings and end of words

• rules and patterns governing the allowed sequences of sounds in a specific language, e.g in English certain consonant clusters like "ng" can occur at the end of words (e.g., "sing") but not at the beginning

• phonotactic constraints vary between languages, have ingrained word templates and some words may not match those templates

• 7½ to 9 months, babies develop sensitivity to the permissible sound sequences of their native language

• BUT if infant doesnt know where word begins and ends difficult to learn some sequences are more frequent at beginning and end of a word

The Prosodic Bootstrapping Hypothesis

• infants use prosodic features (pitch, stress patterns) to help them learn the syntax and structure of native language

• auditory cues act as a scaffold, enabling infants to segment speech into meaningful units

• Infants learn to associate stressed syllables with the beginnings of words

• 90% of english bi-syllabic words have trochaic stress pattern (COOkie, BOTtle)

• 10% have iambic stress pattern (guiTAR, debATE)

• by 7 ½ months recognise words with trochaic stress patterns

• metrical segmentation errors: TARis as a word in a sentence like my guitar is out of tune

• MSE stop at 10m, older infants dont rely on stress patterns so much

• studies show that infants can apply metrical segmentation strategies to languages they've never heard before, highlighting the universal applicability of prosodic cues

Infant Directed Speech/Motherese

• exaggerated prosody, higher pitch, slower tempo

• easier for infants to detect patterns and cues

• exaggerated intonation and articulation highlights syllable word boundaries, easier for infants to segment speech into meaningful units

• infants exposed to IDS better at recognising and learning words

• repetitive and rhythmic nature reinforces linguistic structures

• Kaplan 2002: infants often fail to learn from their own mothers' IDS due to its reduced perceptual salience. Depressed mothers tend to use less exaggerated prosody in IDS

Statistical Learning Saffran et al 1996

• babies can segment streams of sounds from unfamiliar language after 2 mins of exposure without hearing single word on its own and stress patterns or phonotactic constraints

• calculate transitional probabilities: likelihood that one syllable follows another

• syllables that co-occur and are high probability paires become words

• low probability pairs treated as separate speech entities

• Saffran: infants could identify words from continuous, nonsensical streams of syllables after few mins of exposure