Title: Age of acquisition and frequency of occurrence: Implications for experience based models of word processing and sentence parsing
1Age of acquisition and frequency of occurrence
Implications for experience based models of word
processing and sentence parsing
2A rare phenomenon inpsychological research
- All researchers agree
- that word recognition depends on word
frequency. - High-frequency words are recognised faster than
low-frequency words.
3Frequency effectin sentence parsing?
- More controversial
- Is sentence parsing based on principles or on
experience? - Examples of experience-based accounts tuning
hypothesis, probabilistic models, neural networks - This talk is on lexical processing, but the
principles also apply to experience-based models
of sentence processing.
4How do we measureword frequemcy?
- Word frequency is the frequency (per million)
with which words are encountered in a
representative corpus of text or (more recently)
speech. - As it happens, virtually all corpora are based on
texts written and read or spoken by adults.
5Assumptions
- Either frequency of words in adulthood
frequency of words in childhood - or only words encountered in the last years matter
6Testing the assumptions
- Frequency childhood adulthood?
- ambulance adults 15 / million
- children 86 / million
- ancestor adults 15 / million
- children lt 1 / million
- It does not matter?
- Morrison Ellis (1995) words like ambulance
are named 30 ms faster than words like ancestor
(effect of 80 ms in LDT)
7Gerhand Barry (1999)
8The AoA effectdefinition
- Stimuli (words) that have been acquired early are
processed faster than stimuli that have been
acquired later, even when they are matched on
frequency. - AoA measured with rating scales or by directly
looking at the performance of children (r .75).
9The AoA effectexplanations
- Three classes of explanations
- 1. Frequency and AoA have a different origin
- frequency effect strong at input stages AoA
effect strong at verbal output stages - effect of AoA in the organisation of the semantic
system - 2. Frequency and AoA have the same origin
- cumulative frequency instead of frequency
- learning in a connectionist model with
distributed representations (loss of plasticity)
10The AoA effectexplanations
- Three classes of explanations
- 1. Frequency and AoA have a different origin
- 2. Frequency and AoA have the same origin
- 3. Effects of AoA are due to bad frequency
measures - Higher correlation with frequency when measures
take into account childhood frequencies (Zeno et
al.) - Higher correlation with frequency when measures
are based on film subtitles than on books (New et
al., 2007)
11The AoA effectexplanations
- Brysbaert Ghyselinck (Visual Cognition, 2006)
- to decide between the first two classes, we have
to look at the correlation between the frequency
effects and the AoA effects across tasks - if both effects have the same origin, there
should be strong positive correlation between the
magnitudes of the effects - if the effect of AoA is due to cumulative
frequency, then AoA lt freq
12Why the effect of AoA lt the effect of frequency
according to the cum hyp
- High frequency word gt 50 / million
- Low frequency word lt 5 / million
- This gives a ratio gt 101
- Early acquired words lt 5 yr (15 yrs ago)
- Late acquired words gt 15 yr (5 yrs ago)
- This gives a ratio of 31
- So, effect AoA 1/3 effect frequency
13Gerhand Barry (1998,1999a,b)
14Ghyselinck et al. (2004)
15Interim summary
- For many tasks, there is perfect correlation
between the frequency effect and the AoA effect. - This suggests that both effects are the result of
the same learning process (futile to look for
tasks that would show a frequency effect but no
AoA effect). - The AoA effect is too big for the cumulative
frequency hypothesis.
16The connectionist account of AoA
- Ellis Lambon-Ralph (2000)
- Simple network 100 input nodes, fully connected
to 50 hidden nodes, fully connected to 100 output
nodes - Words defined by arbitrary patterns of 0
(80) and 1 (20) - Groups of words entered at different times.
17(No Transcript)
18(No Transcript)
19Discussion with Zevin Seidenberg (2002)
- ZS an AoA effect that is larger than predicted
on the basis of the cumulative frequency
hypothesis will be found only when there is an
arbitray mapping between input and output. - Otherwise, the late-acquired patterns profit from
the connections made by the early acquired
patterns (is the case for word naming).
20Recapitulation claimBrysbaert Ghyselinck, 2006
- if the AoA and the frequency effect have the same
origin, there should be strong positive
correlation between the magnitudes of the effects
21(No Transcript)
22In search of the frequency-independent AoA effect
in picture naming
- 1. Semantic origin of AoA
- Brysbaert et al. (2000) picture naming requires
semantic mediation hence semantic origin of the
frequency-independent AoA effect. - Steyvers Tenenbaum (2005) growing network
23(No Transcript)
24In search of the frequency-independent AoA effect
- problem no evidence for a frequency-independent
AoA effect in semantic decision tasks (there is
an AoA effect in these tasks, but it looks like
it is the frequency-related AoA effect)
25In search of the frequency-independent AoA effect
- 2. The lexical-semantic competition hypothesis
- the frequency-independent effect is due to the
transition from semantics to verbal output - when we name a picture, initially many
semantically related names (lemmas) become
activated, and selecting the correct name
requires a competition between the different
candidates - hypothesis early acquired lemmas are stronger
competitors than late acquired lemmas (e.g., CAT
is a stronger competitor for the naming of LION
than vice versa)
26Belke et al. (2005)
- Made use of the semantic blocking effect.
- It is more difficult to name pictures repeatedly
if they are part of a semantically homogeneous
set than if they are part of a semantically
heterogeneous set.
27Homogeneous condition
28Heterogeneous condition
29(No Transcript)
30Belke et al. (2005)
- On the basis of the previous finding, we
predicted that the blocking effect would be
larger for late acquired words than for early
acquired words
31experiment
- 36 participants
- 16 pictures with early-acquired names and 16
pictures with late-acquired names - matched on visual similarity, frequency, name
length, - stimuli presented 6 times per block
- homogeneous and heterogeneous lists in different
blocks (counterbalanced)
32Early acquired stimuli
orange banana pear carrot lion frog
spider rabbit scissors paintbrush
ladder hammer jumper trousers dress shoe
33Late acquired stimuli
pepper cherry onion lettuce ostrich eagle beetle
camel chisel pliers spanner broom tights shaw
l mitten waistcoat
34(No Transcript)
35In search of the frequency-independent AoA effect
- 3. Picture naming is not always semantically
mediated - Funnell et al. (2006)
- There is a difference in learning experience
between young and old children. - Young children learn about things by seeing them
(either in real or in pictures). - Older children learn about things by having them
described
36In search of the frequency-independent AoA effect
- Funnell et al.
- Pictures of early acquired objects are named
faster, because there is a direct link between
the input stimulus and the name. - Pictures of late acquired objects are named
slower, because the naming requires semantic
mediation
37(No Transcript)
38Conclusions
- AoA as part of the cumulative frequency seems to
be a given. - Question whether AoA has an effect over and above
cumulative frequency looks probable (plasticity
in model) - Still uncertainty about what causes the
frequency-independent AoA effect in picture
naming
39Reading list
- Brysbaert, M. Ghyselinck, M. (2006). The effect
of age of acquisition Partly frequency related,
partly frequency independent. Visual Cognition,
13, 992-1011. - Johnston, R.A. Barry, C. (2006). Age of
acquisition and lexical processing. Visual
Cognition, 13, 789-845. - Juhasz, B.J. (2005). Age-of-acquisition effects
in word and picture identification. Psychological
Bulletin, 131, 684-712.