Speech perception

advertisement
Cognition and Perception: John Beech
Lecture Topics
Cognition and Perception
for PS1000
1.
2.
3.
4.
5.
6.
Dr. John Beech
(for 8 lectures)
Selective Attention (2)
Perception (2)
Pattern recognition (1)
Speech recognition (1)
Visual imagery (1)
States of consciousness (e.g. meditation, hypnosis) (1)
1
2
Selective Attention
Overview
Cherry and the shadowing task – and its problems
Broadbent‟s bottleneck theory
Treisman‟s attenuation theory
Deutch and Deutch‟s model
Visual attention
Parallel processing
Meta-attention
Topic 1: Selective Attention
3
4
Selective Attention
Selective Attention
Four functions of attention
1. Focussing: Attention can be taken to mean the ability to select
just one aspect of information (e.g. limiting the number or
items processed.)
2. Selecting: Attention can also apply to our ability to select to
attend to something (e.g. rather than daydream, or selecting an
action to do next.)
3. Perceptual enhancement: Increasing the gain of a stimulus
(e.g equivalent to turning up the car radio to hear better) –
listening more intently.
4. Sustaining attention: Persisting in concentrating on
something in the face of other distractions. (We‟ll look at this
later under “met-attention”.)
Introduction
The concept of "attention" was regarded as
important by psychologists and
philosophers.
There are at least 4 functions of attention…
5
6
1
Selective Attention
Selective Attention
• Cherry (1953) initiated first systematic work on
attention.
• Interested in “the cocktail party phenomenon”
where we selectively listen to one particular
conversation in crowded room.
• Selective attention is studied in the laboratory by
the “shadowing” technique. (Demonstration next)
Shadowing
7
Shadowing Task
Selective Attention
Cherry, 1953
Gosh,…
this…… is
……..hard
saying
“Practice
makes
perfect” .
Unattended
8
One characteristic is that people speak with a very
monotonous voice.
Also they have very little idea of what the content of
the message was - but of course they must have
recognised each individual word in order to repeat
it back.
Interesting to ask in such experiments: how much of
the rejected message is actually rejected?
Once upon
a time in a
galaxy far,
far away
Attended
Once upon a
time in a galaxy
far, far away
9
Selective Attention
10
Selective Attention
In Cherry‟s experiment participants had a
message read in right ear (passages from
newspapers) and shadowed the same ear. In
the left ear in all conditions it started off
with another normal message and then
changed to one of 4 conditions….
11
Cherry (1953)
In the left (unattended) ear:
1. No change - Normal male speech.
2. Changed to female speech.
3. Reversed male speech (same sound
spectrum)
4. A steady tone.
12
2
Selective Attention
Selective Attention
Cherry (1953) - in the left unattended ear:
1. No change - Normal male speech.
2. Changed to female speech.
3. Reversed male speech (same sound spectrum)
4. A steady tone.
Afterwards they asked questions e.g. was it human
speech? Results:
1. Conditions 1 and 2 were recognised as speech
every time - but couldn't even say if it was
English.
2. Recognised as female speech
3. Thought to be strange by some, normal by
others
4. Tone recognised by all
Moray (1959) looked at out how much
information is retained in rejected channel
or ear. He repeated certain words 35 times
into this ear but absolutely no retention of
words - even if told that they'd be tested
later on their memory for rejected ear.
13
Selective Attention
14
Selective Attention: Problems with shadowing
Monaural presentation means that two different
messages are presented to one ear. If these two
different messages are both presented to both ears
- so that both ears hear exactly the same material,
the presentation is “binaural”.
Dichotic presentation means that the two channels
are fed into the separate ears so that the left ear
hears only one message and the right ear hears the
other - as in Cherry's experiment - then the
presentation is “dichotic”.
Stereo presentation: each message is presented on
two different loudspeakers.
15
Selective Attention: Problems with shadowing
One problem is that it is far removed from the original cocktail
party phenomenon. (We don't normally repeat back a
conversation). It's repeating back that causes primary
problem.
1. Verbal Interference
a) Preventing rehearsal - using the vocal apparatus for
shadowing prevents verbal rehearsal of the unshadowed
message (Norman 1969).
b) Masks sound - the sound of own voice masks rejected ear.
Underwood & Moray (1971): males experienced more
interference when they shadowed if both messages were in
male voices compared with both in female voices. And same
effect - but in reverse - for females.
16
Selective Attention: Problems with shadowing
2. Shadowing not sensitive to changes in
attention
2. Shadowing not sensitive to changes in attention
Denis (1977) found that during the demanding semantic
target searching, comprehension performance was poorer
than under less exacting circumstances searching for
'visual' targets e.g. letter T. But, interestingly, shadowing
performance was not changed i.e. their level of uttering
words was not changed even though attention had been
affected by the semantic search task.
Shadowing supposed to ensuring
participant attending to message. But
Dennis (1977): shadowing performance
can remain unchanged under different
levels of attention. His participants
shadowed message at same time
searched through word lists looking for
semantic targets (e.g. furniture). He then
tested their comprehension of the
shadowed passage.
17
18
3
Selective Attention
Selective Attention
3. Shadowing consumes resources
One way to look at this problem is to compare
between performances when participants
shadow materials and when they just listen
to materials. Typically, they demonstrate
that performance in both channels is
superior when participants just listen.
• Beech and McKeating (1980): participants
listened to radio plays. Two plays ran
simultaneously with one in each ear, and
participants had to attend to left ear. After
questioned on content in their left ear.
• During the play, actors in the plays at random
points said digits and if the participants heard a
number in either ear pressed button. Another
group of had to shadow the play in the left ear and
the two groups were compared.
19
Selective Attention
Selective Attention: Broadbent‟s bottleneck theory
The table shows the percentage of digits detected in each ear
Shadowers
Left ear
(attended)
75%
Listeners
20
Right ear
(unattended)
1%
90%
An early theory of attention was
proposed by Broadbent (1958).
The basis for his theory was a
“split span” experiment in which
he gave participants 3 pairs of
digits dichotically - in such a way
that one member of each pair
went to one ear at the same time
as the other member went to the
other ear.
26%
Thus, when people shadowed, virtually no digits could be
monitored in the unattended channel. However listeners
detected 26% of digits in their right ear suggesting that about
a quarter of the individual words streaming through their
unattended ear were actually being monitored.
21
Selective Attention: Broadbent‟s bottleneck theory
To left ear
To right ear
389
514
22
Attention Filter Theories: Early Filter Theory
Broadbent’s (1958) Bottleneck Theory
Evidence
Broadbent, 1959
741
Time
1
2
3
Digit
pair
95
81
34
382
741
382
asked to recall both messages
- Recall input to one ear then other, not order
-“switching” of attention between ears
23
24
4
Selective Attention: Broadbent‟s bottleneck theory
Selective Attention: Broadbent‟s bottleneck theory
At the slower rate participants reported digits paired together
(LRLRLR). At the faster speed of ½ sec/pair then S's
recalled spontaneously ear by ear i.e. LLLRRR.
Broadbent concluded: selection of one message via the
physical channel digit was presented - in this case the left
ear or the right ear.
Proposed: central limited capacity channel which filtered
information. But this filter takes time to switch from one
channel to next. Therefore participant couldn't alternate
attention at fast speeds, one switch requiring .75 sec.,
according to Broadbent. Meanwhile message in other ear
stored in an auditory short term store. While information
in store, representation of message decays.
25
26
Selective Attention
Selective Attention: Broadbent‟s bottleneck theory
If the information is too much
for system then filtering
mechanisms are needed to
prevent system from
becoming too overloaded.
Filters can be tuned by people
to any of several channels e.g.
the ears, eyes, smell, sense of
touch etc.
Broadbent's basic hypothesis:
Limited capacity processor can only deal with
one channel of information at a time.
Information chosen on basis of physical features
such as intensity and pitch.
Limitation of central processor is in terms of
bits of information that person can handle per
second. If information is low, system can cope
with more than one channel.
Selective Attention: Broadbent‟s bottleneck
27
theory
Selective Attention
28
Gray & Wedderburn (1960)
• Gray and Wedderburn (1960) conducted an
experiment to test Broadbent's idea that attention
was based on physical characteristics of sensory
channels. They suggested that the rejected
message might also be processed for its meaning.
• In their experiment, one word presented so that
different syllables were presented alternately to
different ears. At the same time another word is
decomposed in a similar way and presented to the
other ear.
29
Dic
Pro
cess
tion
ary
ing
30
5
Selective Attention: Treisman‟s
attenuation theory
Selective Attention
Participants could do task. Suggests: meaning
of both messages had to be extracted in
order to know which message to select.
Broadbent's system, proposes that switching
occurs early on in processing, and is based
purely on the physical characteristics of the
message.
31
32
Treisman‟s theory
Treisman‟s theory
Anne Treisman (1961): the switch really an
attenuator i.e. a mechanism that reduces
information getting past, rather than
shutting it off completely. To demonstrate:
participants are shadowing one message and
ignoring another in the other ear. Suddenly
the 2 passages crossed over!
The messages switch ears at this point
Left ear
Sitting at a mahogany three possibilities
Right ear Let us look at these
table with her head
33
34
Treisman‟s theory
Treisman‟s theory
Messages analysed by filter for crude physical properties e.g.
loudness, pitch, position, colour etc. Resulting information
is available for reporting by participant. Filter also
attenuates signal strength of output from these analysers.
E.g. participant has to select only a female voice; filter will
attenuate any signals that don't have this quality.
Unweakened signal plus weakened signal(s) processed further
via pattern recogniser (composed of large no. dictionary
units). By travelling up a logical tree the components of
the messages reach the end of their respective trees and
each fire a dictionary unit.
Units have 2 properties:
(a) Their thresholds differ e.g Unit firing own
name has permanently lower threshold so that
even if one's name is on attenuated message (i.e.
the weakened signal), unit would still fire.
(b) The thresholds are variable. Threshold
differences between word units may be brought
about by e.g. instructions to the participant
('listen out for articles of clothing') or by
context.
35
36
6
Treisman‟s theory
Deutch & Deutch‟s (1963) late filter theory
They proposed a 'response selection' theory of selective
attention. If one alters properties of dictionary, renders
lower level filter irrelevant. At dictionary each signal is
analysed and recognised.
From there: to output mechanism which transmits to a level
that participant becomes aware it. When signal input to
dictionary it will fire according to how important stimulus
is (changes in importance can be produced by context etc.)
and extent of output activity from output mechanism is
proportional to importance of the signal.
The feature of this system is that messages analysed at earlier
stage before the person is conscious of content.
Figure
37
Evidence against early selection (and for
Deutch & Deutch) by MacKay (1973)
Right ear
(half the
group)
“Money”
“River”
Left ear
Person says
They threw stones
toward the bank
yesterday
Sentences with
ambiguous meaning
They threw stones
toward the bank
yesterday
38
Evidence against early selection (and for
Deutch & Deutch) by MacKay (1973)
Phase 1– Sentences. They attended to their left ear and heard
28 ambiguous sentences and in the unattended ear one
group heard e.g. “money” while the other heard e.g.
“river”.
Phase 2 – Recognition. All participants had to choose
sentences that were the best match to those they‟d heard:
(a) “They threw stones toward the side of the river
yesterday”
(b) “They threw stones toward the savings and loan
association yesterday”
39
Evidence against early selection (and for
Deutch & Deutch) by MacKay (1973)
Result
Those who heard “money”in their unattended ear chose
“financial institution” and those who heard “river” chose
“river bank”
They were asked about the word in the unattended ear and
they were completely unaware of it.
40
Selective Attention: parallel processing
Visual attention
Other techniques are being used to study attention.
E.g.: Neisser & Becklen (1975) studied attention in
the visual mode. Participants examined 2 optically
superimposed video screens on which 2 different
kinds of thing were happening. They had to follow
the action in one episode - by pressing keys when
significant events occurred e.g. in hand game the
person pressed key every time an 'attacking stroke'
made (whether or not it produced a hit) - and
ignore the other.
Conclusion
The word in the unattended ear was processed for its
meaning.
41
42
7
Neissser & Becklen
(1975)
Selective Attention: parallel processing
Visual attention
• x
People could do this without difficulty. However (as
in shadowing task) very little information from the
unattended film was processed.
Neisser & Becklen (1975) proposed that it is highly
improbable that special filters 'block out' irrelevant
material. Also: people find it very difficult (or are
unable) to process two things simultaneously in the
same modality.
43
Selective Attention
44
Selective Attention: parallel processing
Broadbent's Model (dealt with before)
• In essence in the single channel hypothesis we deal only
with one channel at a time.
Information enters short term memory.
The filter selects information on the basis of the physical
features of the message e.g. intensity and pitch. The filter
can be tuned to any of several channels e.g. the ears, eyes,
smell, sense of touch etc.
There is a limited capacity channel that can handle a certain
number of bits of information/sec. If information load is too
high, then the filtering mechanism operated to try to reduce
the information. If information is low the p system (limited
capacity channel) can handle several channels at once.
45
Selective Attention: parallel processing
In other words: Broadbent's point was that as long as the
two tasks that one is performing are low in information
content (i.e. not very demanding on resources) one
can do the two tasks simultaneously.
Soon afterwards, as we've seen, Broadbent challenged
on his ideas concerning a filter (e.g. Ann Treisman). It
was argued that a certain amount of processing of
rejected messages also takes place, and Treisman
demonstrated this with some rather ingenious
experiments.
The part of Broadbent's model suggesting that the system
cannot handle information if information load is too
46
high challenged by Allport et al. (1972).
Selective Attention: parallel processing
Allport et al. (1972) found that people can
do things simultaneously in different
modalities. For example, they could
look at and remember pictures and
shadow (ie. repeat back) at the same
time without any problem. However,
they couldn‟t listen to single words in
one ear and shadow the other ear at the
same time – (similar to Cherry) – it is
very difficult to do things simultaneously
in the same modality.
Allport et al. gave stimuli (1
every 3 sec) for 45 sec, then
a one-minute rest, then test
period showing originals +
new, similar set.
Participants either
shadowed, or not. Stimuli
were pictures, wordsvisual, words-auditory.
47
48
8
Selective Attention: parallel processing
Atwood‟s (1971) model
Selective Attention: parallel processing
Allport et al. (1972) in another
experiment looked at the sight
reading of music while shadowing
and found that this was achieved
relatively easily, but performance
dipped when shadowing more
difficult material (Norse poetry).
So the information content does
have a role to play.
Atwood proposed that we have 2 separately operating
modules (or systems) in which different operations can be
done simultaneously as long as they are in the separate
modules. But if we try to do two things in the same
module then we experience problems.
He proposed that the two systems were the visual system and
the verbal auditory (VA) system.
49
50
Selective Attention: parallel processing
Selective Attention:
parallel processing
Atwood‟s (1971) model
Atwood (1971) gave 35 phrases („nudist devouring
bird‟) with 4 sec gaps between each phase. Then
cued recall („nudist‟ ____?). Phrases were
concrete (told to image these) or abstract
(„Einstein was a genius‟). Three types of
interference:
1. none
2. visual (shown „1‟ said „2‟)
3. auditory (heard „1‟ said „2‟). 6 groups.
Found…
51
Selective Attention: parallel processing - Atwood
Selective Attention: parallel processing
Criticisms of Atwood
Type of interference
(% correct)
Concrete
phrases
Abstract
phrases
visual
auditory
none
58%
76%
82%
60%
44%
70%
52
Concrete phrases more bizarre than abstract ones
Atwood confounded instructional set with abstractnessconcreteness of material i.e. visual imagery for concrete
phrases, verbal processing for abstract phrases
Lack of replication - mainly in unpublished studies (e.g.
Brooks, Bower, Neisser, Janssen)
Despite these problems, the Atwood model was a forerunner
of more recent theories of modularity. Useful to take view
that different modules can operate with some measure of
independence of each other.
53
54
9
Meta-attention
Meta-attention
But we do make decisions reflecting use of subjective
judgements about attention and performance (e.g. if
academic work, might have radio on, but turn it off for
other tasks).
Reisberg and McLean (1985) examined effects of distraction
on participants in terms of performance on the task and
their subjective impressions of distraction.
Meta-cognition is our ability to tune in or reflect upon our
own mental processes. Thus meta-attention refers to our
ability to introspect about the attention process. On one
view attention is closely allied to consciousness. If we
control our attention, we should be able to reflect upon it.
55
Meta-attention
56
Meta-attention
Reisber & McLean’s findings
Participants had to add up 10 numbers shown on a
computer screen and distraction was recording of
1 minute of Joan Rivers. Their subjective
impressions were monitored by asking the
following 2 questions at the end of each of 20
trials:
1. How much of the distracting message did you feel
that you heard? (0-100%)
2. How much did you feel disrupted by the presence
of the distraction? (0-5 rating scale)
1. The presence of a distractor slowed down
responding to the main task
2. There was no association between response
latency and reports of distraction, nor an
association with how much the message was
heard. Thus, people were hopeless in predicting
their trial-by-trial performance.
Thus material had significant distracting effect.
57
Meta-attention
58
Meta-attention
One explanation might be that participants don't actually have
access to processes involved in producing performance.
An implication might be: We find ourselves in situation in
which we have to work but there are distractions, e.g. we
might be in the library trying to work and someone is
talking elsewhere. We might decide to try to deal with the
distraction, however, our judgement of what is actually a
distraction might be mistaken.
In next experiment included a financial reward for doing task
and found that distracting effect virtually disappeared.
However, there wasn't corresponding effect on self-report,
because they reported being equally distracted by comedy
tape as participants in previous group.
Again, there's no relationship between actual performance
and self-reports from trial-to-trial.
59
60
10
Meta-attention
Meta-attention: Conclusion
Hovey (1928) tried to use extreme distractors.
There were 7 electrical bells, 4 buzzers, 2 organ
pipes and 3 whistles, a circular saw, lively
music, a spotlight flashing continuously; others
entered noisily and strangely garbed, carrying
strange apparatus. Nail kegs rolled up and
down the aisles.
Finding: these distractors worked only for short
time while the participants were becoming
adjusted to the situation. (Participants were
compared with non-distracted group while both
groups took on IQ test).
A paradox about distraction effects is very difficult
to create artificially in lab., but in everyday life
seems to occur so easily. In lab. distractors have
to be quite high in distractibility such as stimuli of
great personal importance (e.g. one's name) or
inputs which conflict with a highly-learned
response (e.g. the Stroop effect). But even when
distractor found (e.g. comedy) can be overcome
by giving participants sufficient inducements to
concentrate. Where there is a will there is a way.
61
62
11
Download