Investigating phonological processing by varying pronounceability: MEG Evidence Jeremy Bronheim, Martin Hackl, David Poeppel University of Maryland, CNL Lab, Departments of Biology and Linguistics, NACS Program

1
Investigating phonological processing by varying
pronounceability MEG EvidenceJeremy Bronheim,
Martin Hackl, David PoeppelUniversity of
Maryland, CNL Lab, Departments of Biology and
Linguistics, NACS Program

• Experiment 2 MEG Study
• Materials and Methods
• 11 right-handed subjects with 2020 vision were
  run. 3 subjects data was discarded because of
  poor performance on the task or excessive noise
  in the MEG data.
• While in the MEG scanner, subjects performed a
  lexical decision task using the following items
• 30 High Frequency (HF) words
• 30 Low Frequency (LF) words
• 30 items chosen from each category of Exp. 1
  (PNW, Onset violation non-words, No vowel
  non- words, and Consonant strings)
• 30 filler words to better balance wordnon-word
  ratio
• 30 Amharic Script items (e.g. Mover) used as
  False Fonts (FF)
• Figure 2 Shows the distribution of the items
  chosen from Experiment 1.
• Figure 2 The distribution of
  items chosen from
• Experiment 1 to be stimuli in
  Experiment 2.

Results (continued) All 3 peaks were found for
all conditions. An example of a condition with
all 3 peaks is shown in Figure 6.
Figure 6 Sample data
showing M250 latency differences in two
conditions, with all 3 peaks present.
Introduction Embick and colleagues (2001)
performed an MEG study of visually presented
words during a lexical decision task. They found
that in the first 400 ms after stimulus onset,
the MEG data showed 3 primary peaks. The first,
at approximately 170ms, the M170, is thought to
correspond to visual word form recognition. The
next peak, the M250, occurs at approximately 250
ms, has not yet been found to correspond to a
specific function. The M350 was found to have a
latency that was inversely proportional to the
frequency of the word stimulus. The M350 also
shows priming effects (Pylkkänen et al. 2000) and
other properties that suggest that it is an index
of lexical access. In visual lexical decision,
the orthography of a visually presented word must
be converted into a phonological code before
lexical search is performed. If the M170 reflects
visual word form and the M350 lexical access, the
M250 may reflect aspects of phonological
processing, e.g. processes related to
pronounceability. Hypothesis 1 The M250 peak is
the activation associated with the process of
converting the orthography into a phonological
code, which is then used in the lexical decision
task. In order to test this hypothesis, we need
to show, therefore, that the time this task takes
is affected by the difficulty of converting an
items orthography into phonology. The
difficulty of this conversion can also be though
of as a measure the pronounceability of an item.
Thus Hypothesis 2 M250 latency will be related
to the pronounceability of the stimulus
presented.
Conclusions For words and pronounceable
non-words, the M350 did appear to be inversely
related to word frequency, as found in previous
studies (Embick et. al., 2001). We did, however,
find an M350 peak for the false fonts. This
would not be expected if the M350 represents
lexical access, because there should not be
anything to search the lexicon with in the false
font condition. Some other explanation for the
M350 response to false fonts must be found. The
M250 latency data (Figure 4) only partially
followed predicted trends. The words and
pronounceable non-words follow the predicted
trend (Figure 4a). Further, the fact that
pronounceable items had longer latencies than
unpronounceable items (Figure 4b) indicates that
the orthography to phonology conversion
hypothesis may, indeed, be correct. For the
non-words, we hypothesized that greater
violations would elicit earlier M250 latencies,
since the conversion process would fail. We did
not find this trend among the non-words (Figure
4a), although another possibly relevant trend was
present. If we order the expected M250 latency
for non-word categories based upon how early in
the letter string the orthography or phonotactics
of English are violated, we get the following
order (from shortest to longest latency) False
Fonts (non-English orthography), Onset violation
non-words (2 letters), consonant strings (2-4
letters), then, finally no vowel non-words (3-4
letters). The one non-word type that does not
follow this order are the no-vowel non-words.
This can be explained by a confound we introduced
into the stimuli the no-vowel non-words also
had onset violations. This feature of the
no-vowel non-words makes sense if overall
pronounceability of the non-words is what is
being tested, but not if what needs to be varied
is how early in the non-word the first violation
occurs. Because they have onset violations, the
no-vowel non-words are phonotactically bad just
as early as the onset violation non-words,
indicating that they should have the same
latency, as they do. Despite not finding the M250
latency order that we expected, our hypotheses,
that the M250 represents the orthography to
phonology conversion process and thus, that its
latency should depend on pronounceability, were
tentatively confirmed. In order to test the
validity of the interpretation leading to this
confirmation, a follow-up study varying the point
in a non-word at which it violates English
phonotactics needs to be done.

• Experiment 1 Stimulus norming
• Purpose To find four, well separated groups of
  non-word stimuli (to be used in Exp. 2) based
  upon pronounceability.
• Materials and Methods
• 4 categories of non-words
• Ratings
• Scale of 1-7
• 23 Subjects rating 225 non-words each
  (approximately 56 non-words per category)
• Results
• 4 groups were rated as significantly different (p
  lt 0.01). The separation is show in figure 1.

Non-Word Type Example
Pronounceable Non-Word blicket
Onset Violation Non-Word bficket
No Vowels Non-Word bfnckrt
Consonant String bvpxqtk
Hypothesis Peak Existence/Latency Expectations
by Category Figure 3 a (left) shows
expectation of the existence of each peak under
each condition. b (right) shows the expected
latency of M250 peaks (if existent) based upon
the hypothesis That it is effected by difficulty
of pronounceability. PNW are above LF words
because some PNW may have pronounceabilities
corresponding to HF words, and others
corresponding to LF words.
M170 M250 M350
Words HF LF ? ? ?
Pronounceable Non-Words ? ? ?
Onset Violation Non-words ? ? ?
Non-words w/o vowels, but with liquids and nasals ? ? ?
Consonant Strings ? ? X
False Fonts e.g. Qellom ? X X
M250 Latency Order (if possible peaks existent)
High Frequency Words 4
Low Frequency Words 6
Pronounceable Non-Words 5
Onset Violation Non-Words 3
Non-words w/o Vowels 2
Consonant Strings 1
Results Peak Latencies M250 Figure
4a M250 Latency by Condition
Figure 4b M250 Latency Phonotactic
Legality M350 Figure 5 M350
Latency by Condition
References Embick, D., Hackl, M., Schaeffer,
J., Kelepir, M. Marantz, A. (2001). A
magnetoencephalographic component whose latency
reflects lexical frequency. Cognitive Brain
Research, 10, 345-348. Pylkkänen, L., Flagg,
E., Stringfellow, A. Marantz, A. (2000). A
neural response sensitive to priming An MEG
study of lexical access. Poster Session presented
at the annual meeting of Cognitive Neuroscience,
San Francisco, CA.
Acknowledgments This work supported by NIH DC
05660 to DP.