Introduction to Language Comprehension

Mental Representation for Production

Input > Activate (link to existing representation) > Output (speech, writing etc)

Building Blocks of Language (ordered)

-              Semantics

-              Syntax - Morphology

-              Form

-              Speech

Building Blocks of Language for Comprehension (ordered)

-              Speech

-              Form

-              Semantics - Morphology - Syntax

Mental Representation for Comprehension

Input > Activate (link to meaning) > Output (comprehension)

Challenges of Comprehension

-              Ambiguity in the speech stream (word boundaries are unclear).

-              Ambiguity at the word level (difficulty linking words that are pronounces the same to meaning).

-              Ambiguity at the phoneme level (surrounding sounds change the way we articulate).

Why there is ambiguity

Word boundaries - impossible to know where one word ends and another starts.

-              Gaps in speech sounds do not necessarily correspond to word boundaries E.g. ‘ice cream’ vs ‘I scream’.

Word level ambiguity (3 types)

• Homonyms

• Homophones

• Homographs

Homonyms

Words that sound and are spelt the same (e.g. bank/bank).

Homophones

Words that sound the same (e.g. muscle/mussel).

Homographs

Words that are spelt the same (e.g. bow tie, bow and arrow and bow down).

Ambiguity at the Phoneme Level - Coarticulation

‘Look at the thin carpet’

In this example the alveolar ridge phoneme /n/ is pronounced as an /n/ when it precedes the velar phoneme /c/.

‘Pass me the thin book’

In this example the alveolar ridge phoneme /n/ is closer to an /m/ when it precedes the bilabial phoneme /b/.

Disambiguating the Ambiguity

Categorical perception of sounds provides a way for us to work out where one sound ends and another sound starts.

 

-              Ability to distinguish between sounds based on voice onset time (VOT).

-              /b/ and /p/ are bilabial consultants articulated in the same ‘place’ but they sound different because they have different VOT.

-              VOT is defined by the point at which vocal cord vibrations start relative to the release of a closure.

Voice Onset Time

0 VOT - vocal chords vibrate as the closure for letter is released. E.g. ‘p’ in pan.

+ VOT - vocal chords vibrate after the closure for the letter is released. E.g. ‘b’ in ban.

Categorical Perception

-              Facilitates distinction between similar words e.g. ‘toll’ and ‘doll’.

-              Children appear to be hardwired to recognize different phonemes based on features such as ‘voicing’.

Regional Accents

The range of sounds allowable for a single phoneme could differ dramatically depending on the accent in which it is spoken.

Range of sounds

-              ‘y’ in city could be pronounced in many different ways depending on where someone is from.

Invariance Problem

   It is not that easy to define a range of sounds.

-              The inability to define the acoustic properties that facilitate the categorisation of phonemes in referred t as the inference problem.

Warren (1970)

Out of 20 participants - 19 said they heard no missing sounds. One participant identified a missing sound but identified the wrong one.

Our mental lexicon restores missing phonemes (sometimes incorrectly).

Disambiguating the Ambiguity

-              Acoustic properties of phonemes (some other properties can be used to define phonemes [aside from the unresolved invariance problem]).

-              Categorical perception (using voice onset time to distinguish phonemes).

-              Top-down processing (activating of existing lexical representations).

Features of Lexical Items

Top-down processing of existing mental representations of words stored in our mental lexicon can be used to work out what we are hearing.

The Mental Lexicon

Simply - all the words we know. Exists as a storage bank for all the words we have assess too. Including: Syntax, phonological, orthographic, semantics.

Spreading Activation

When words are said/thought about, automatically the idea of similar sounding words are activated e.g. Apple > apron, approve, apathy…

Lexical Access

-              Lexical access will be faster for words that are short and frequent compared to words that are long and infrequent.

-              Lexical access will be slower for words with lots of neighbours compared to words with fewer neighbours (as there will be less interference).

Define Neighbourhood Density

E.g. walk - talk walk - wall ‘neighbours.

Assessing Lexical Access - Lexical decision task

Presented with list of words and have to decide if they are real words or not - the faster a real word is discriminated, the faster the word is detected - telling us how many ‘neighbours’ there are.

Luce & Pisoni, (1998)

Found faster response with words with fewer neighbours, compared to words with more neighbours.

Other affecting factors:

-              Frequency of words used.

-              Word length

All rules interact with each other.

Marslen-Wilson & Taylor (1980) - The Impact of Context

Participants asked to monitor a speech for the word ‘motorway’ and say the word whenever they hear it.

Concluded/demonstrated - The language system predicts which words might come up next and actives them in the lexicon.

-              Sentence contexts help us activate potential words candidates, but it does not necessarily facilitate the selection of the word.

Priming Paradigm

E.g. ‘doctor’ and ‘nurse’ are semantically related - so if we were primed with one, it would be expected for us to assess the other quickly.

Zwitserlood (1989)

Cross Modal Priming

Investigated context in integrating words into a sentence.

Presented prime word - auditory and target words - visual.

Found - faster priming effects for words that are semantically related, even if only fragments of the word are showed (e.g. ‘cap’ for ‘captain’).

Biassing - can we bias our processing?

Participants are presented with a prime with some context (fragment for word within a sentence).

Found - biasing of the context does not necessarily affects access.

Biassing - What was found

-              Word monitoring evidence suggests that participants can use context to activate lexical candidates that fit the context of a sentence.

-              Cross modal priming evidence suggests that participants do not use context to only activate items that fit with the context.

The system accesses lexical items based on

-              Acoustic input

-              Top-down processing

-              Lexical characteristics (e.g. frequency, neighbourhood density).

-              Activation of related items

-              Context (in some ways).

The language system activated a huge amount of information to predict what might come up next and process the information rapidly.

Models of Speech Comprehension

Evidence of how we access lexical items has results in the development of computational models of speech comprehension that allow us to model the process of lexical access.

E.g. The Cohort Model (1987), The TRACE model (1999)