Whats attention for?
- We are constantly presented with information from the external world
- Audition: Numerous overlapping sounds simultaneously reach your ears. The busy street; the conversation in the noisy pub
- Vision: complexity and information overload characterize most visual environments
- Your own thoughts: Even what you are thinking competes for your attention
…Our attentional capacity is severely limited: There is only so much we can attend to at one time
Effective selection of information is critical to functioning
The problem?
The Cocktail Party Problem (Cherry, 1953) – how can we follow one conversation when several people are talking at once?
- When listeners attend to one auditory input, how much processing is there of the unattended input?
- In particular, what kind of characteristics of the unattended channel are processed?
- Semantic processing: Extracting the meaning from the input. E.g., A ‘Cat’ has four legs, is furry, is a pet vs. shallow processing – ‘Cat’ rhymes with mat, has three letters, one syllable etc
Models of Auditory Selective Attention and the Bottleneck
Many argue for a bottleneck.
- Similar to a narrow part of a road, it can cause traffic congestion and a bottleneck in the processing system seriously limits our ability to process simultaneous sounds and how much information can get through.
- Theories of auditory selective attention differ in their positioning of the bottleneck
3 main types of theory proposing where is the bottleneck/filter
Early filter models (Broadbent, 1958)
Attenuation models (Treisman, 1964)
Late filter models (Deutsch & Deutsch, 1963
Differences in the theories?
All theories propose that somewhere there is a bottleneck(filter/attenuator) which allows some information through and slows down the rest.
- The main difference between the three theories discussed is the location of this bottleneck.
- Broadbent: The bottleneck occurs early in the system
- Treisman: The bottleneck occurs mid-way in the system
- Deutsch & Deutsch: The bottleneck occurs late in the system
Eperimental Methods for Assessing Auditory Selective Attention
- whats dichotic listening?
Early studies: shadowing task
- Participants hear a recording of someone speaking delivered to one ear and must repeat this speech back while they are listening to it (the attended channel)
- At the same time, a second message is played to the other ear, which is to be ignored (the unattended channel)
typical outcome from dichotic listening
typical outcome: Shadowing performance normally very good
•Participants hear remarkably little from the unattended channel (Cherry, 1953)
dichotic listening: treisman 1964
Participants shadowed coherent prose in the attended channel, but were presented a text in Czech, read with an English accent in unattended channel
•The individual sounds resemble English, but the message is gibberish
.•Results: After 1 minute of shadowing the attended channel, only 4/30 participants detected the peculiar characteristic of the unattended channel.
dichotic listening (cherry 1953)
Participants can report the physical characteristics of the voice in the unattended channel (speaker sex, voice pitch, loudness)
- Even if the semantic content is not processed to any degree.
- Moray (1959): poor memory for the unattended channel even when words were presented 35 times.
- These findings were developed by Broadbent (1958) into an early filter model of attention
Dichotic Listening: General Predictions by Model
•Broadbent:Early selection.
Predicts little or no processing of unattended auditory messages. They are filtered out early on and not subjected to semantic processing.
•Treisman’s attenuation approach:
Listeners start with processing based on physical characteristics and then process grammatical structure and meaning. The unattended channel is attenuated (turned down) and so receives less processing than the attended channel. The unattended channel could, therefore, receive some semantic processing (but less than the attended)
.•Deutsch and Deutsch’s late filter
approach argued all stimuli are fully analyzed, with the most important or relevant stimulus determining the response. Thus, they placed the bottleneck late in processing which predicts both auditory channels should receive semantic processing
Selective Auditory Attention Dichotic Listening (Broadbent, 1954):
What did they do/ find?
Participants wear headphones and are presented with different sounds in each ear.
Task was to recall as many digits as they could.
Found: Most participants chose to recall the digits ear by ear rather than pair by pair.
Selective Auditory Attention Dichotic Listening (Broadbent, 1954)
Explanation for findings
Broadbent accounted for these findings as follows:
Two stimuli/messages presented at the same time gain access in parallel to a sensory buffer.
One of the inputs is allowed through a filter based its physical characteristics/location, with the other input remaining in the buffer for later processing.
Only information that makes it through the filter is processed for meaning (semantically).
This filter prevents overloading the limited capacity mechanism beyond the filter which processes input for meaning.
Broadbent’s (1954) Early Filter Theory: Evaluation
Model can account for Cherry’s findings by assuming the unattended channel is rejected by the filter.
It can account for dichotic listening findings by assuming the filter selects one stream (ear) of information based on physical characteristics.
Semantic Processing in the Unattended Channel: Moray (1959)
what they did/find?
Attended Channel: Shadow a message presented in a monotone male voice.
Unattended (non-shadowed) Channel: Heard a similar prose passage, but were twice presented with their name and an instruction e.g. ‘John Smith you can stop listening now’
Found: 4/12 participants reported hearing their name in the unattended channel.
Provides evidence of semantic processing in the unattended channel.
Semantic Processing in the Unattended Channel: Corteen & Wood (1972)
what did they do/find?
Part 1: Participants presented with a list of words and each time a particular category was presented, they received an electric shock (cities for example). Aim: To form an association between shock and category.
Part 2: Dichotic listening task. As usual, participants could not remember anything from the unattended channel, but their Galvanic Skin Response (GSR) showed that each time the ignored ear was presented a ‘shocked’ word, there was a response.
Found: A GSR was even detected for words associated with shock and words of the same category that had not been presented.
Implication: This generalization of GSR strengthens the claim that the meanings of unattended words were processed even if not consciously perceived.
Semantic Processing in the Unattended Channel: Aydelott et al. (2015)
WHAT DID THEY DO
Asked listeners to perform a task on attended target words.
When unattended words related in meaning were presented shortly before the target words themselves, performance on the target words was enhanced when unattended words were presented as loudly as attended ones.
Thus, the meaning of the unattended words was processed.
Deutsch & Deutsch’s (1963) Late Filter Theory
All information (from both channels) is attended and processed semantically.
Irrelevant information filtered out at the time of short-term memory storage (it is a “late filter” model).
Both channels (sounds presented to ears) are processed to the same degree, only the most relevant channel to the task at hand is responded to.
Can account for hearing your own name in the unattended channel
Treisman’s (1964) Attenuation Model
Treisman retained an early filter that distinguishes information on physical differences (like Broadbent’s model)
But! Rather than eliminating the unattended material, Treisman’s theory “attenuates” it (turns it down).
Accounts for findings such as hearing your name in the unattended channel i.e. the cocktail party effect (Moray, 1959)
Treisman & Riley (1969)
Participants shadowed a message in one ear, but they had to tap when they heard a target word from any source (either ear)
Attenuation model: more targets should be reported from the shadowed stream than the unattended stream
Late filter model: equal numbers of words should be detected from each stream
FOUND: Many more words were detected in the attended stream relative to the unattended stream, supporting Treisman’s attenuation theory.
How do we solve the cocktail party problem? Top-down processing
Various top-down factors appear to influence our ability to solve the cocktail party problem.
McDermott (2009) found that listeners are more accurate at identifying what one speaker is saying in the context of several other voices if they have previously heard that voice before isolation.
Evans et al. (2016) compared patterns of brain activation when attended speech was presented on its own or together with competing unattended speech.
Brain areas associated with attentional control processes were more activated in the latter condition, showing dop-down processes related to attention and control are important in selective auditory processing.
In sum, listeners generally achieve the complex task of selecting one speech message from several quite effectively. There has been progress in identifying the top-down processes involved.
If listeners can identify at least one consistently distinctive feature of the target voice, this makes it easier for them to attend only to that voice.
Top-down processes often produce a ‘winner takes it al’ situation where the processing of one auditory input suppresses the brain activity associated with other inputs (Kurt et al., 2008).
The debate still continues as to the positioning of the bottleneck (as per our models discussed earlier).
