Separating lexical access from decision: an MEG study

1
Separating lexical access from decision an MEG
study
KIT/MIT MEG LAB

• L. Pylkkänen1,2, A. Stringfellow1,2, M. Kelepir1,
  
• A. Marantz1,2
• 1Department of Linguistics and Philosophy,
• KIT-MIT MEG Laboratory,
• Massachusetts Institute of Technology
• 2Mind Articulation Project, International
  Cooperative Research Project, Japan Science and
  Technology Corporation

2
Introduction

• What is the timing and location of lexical
  activation in the brain?
• Evidence from electrophysiological measures
• There is a response component at 300-400ms (the
  N400 in ERPs or the M350/N400m in MEG) whose
  latency and/or amplitude are affected by some of
  the same stimulus properties that affect reaction
  times (RTs) in various word recognition tasks.
• Question Does this component reflect a lexical
  or a postlexical process?

• Goal of the present study
• To study the M350 elicited by stimuli that make
  lexical activation fast but postlexical processes
  slow. With such stimuli, does the M350 occur
  early or late?

3
What is MEG?

• Electric activity in the brain (produced by
  neurons firing) can be measured noninvasively
  outside the skull by measuring either the
  electric potential or the magnetic field produced
  by the current.
• MEG (magnetoencephalography) measures the
  magnetic field. EEG (electroencephalography)
  measures the electric potential.
• Both MEG and EEG have millisecond by millisecond
  temporal resolution. The spatial resolution of
  MEG is, however, better than the one of EEG since
  electric fields get distorted by the scalp and
  tissue while magnetic fields do not.
• Of existing brain imaging techniques, MEG has the
  best combination of spatial and temporal
  resolution.

• The brain's magnetic fields are very weak, and
  therefore sensors employing extremely sensitive
  magnetic detectors known as SQUIDs
  (Superconducting Quantum Interference Devices)
  are used to pick up the signal. Also, MEG
  measurements are carried out inside a special
  magnetically shielded metal room, to reduce the
  amount of environmental noise.
• Each MEG sensor measures changes in the magnetic
  flux that is either coming out of the skull or
  going into the skull (see picture on left).

A waveform showing the average strength of the
magnetic field in the left hemisphere sensors for
a stimulus-locked epoch of 500 ms. Each line
represents activity in one sensor.
A view of a magnetic field from outside the scalp
for one point in time. An outgoing and ingoing
magnetic flux shows up as a bipolar pattern here
red indicates outgoing and blue ingoing flux.
4
What is the M350?

• The M350 is an response component in the left
  hemisphere at 300-400 ms after the presentation
  of a word (or a word-like string) (Embick et al
  2000, Pylkkänen et al, 2000).

M350 waveform showing an average of the responses
to 70 visually presented words. The lines
represent the magnetic field strengths in 11 left
hemisphere sensors in a time period starting from
stimulus presentation (0 ms) to 500 ms post
stimulus. The cursor is pointed at 400 ms which
is the time of the M350 peak for this subject.
M350 contour map showing a view of the M350
magnetic field from outside the skull at the time
of the M350 signal maximum. The M350 has a
dipolar pattern in the left hemisphere oriented
along the anterior posterior axis with the
negative field on the left and the positive field
on the right.
M350
The M350 source is located in the left temporal
lobe below and in front of the auditory cortex.
The picture above shows four M350 localizations
(the four dipole cluster) and a localization of
the M100, an auditory evoked response, for one of
the subjects in the present study. 300-400 ms
activity post word presentation is consistently
localized in the temporal lobe across the
literature although there is variation in exactly
how much below the auditory cortex the source is
(e.g. in the present study the activity is
approx. 4 cm below the auditory cortex while
Helenius et al. 1999 report the difference to be
2 cm).
5
Stimulus properties affecting reaction times and
the latency of the M350 in lexical decision tasks

• Frequency RTs and M350 latencies are shorter for
  frequent than for infrequent words (Embick et al
  2000).
• Repetition RTs and M350 latencies are shorter
  for repeated than for nonrepeated words
  (repetition priming effect) (Pylkkänen et al
  2000).

• The M350 is earliest MEG component whose latency
  can serve as a predictor of the frequency and
  repetition priming effects on behavior.
• Two interpretations of the data between which the
  present study is designed to distinguish
• The M350 reflects automatic activation of the
  lexicon and shows the frequency and repetition
  priming effects since frequent and repeated
  stimuli activate lexical representations faster
  than infrequent and nonrepeated stimuli.
• The M350 reflects a post-lexical process, such as
  the word/nonword decision which occurs earlier in
  the frequent and repeated conditions because
  lexical activation, and hence all subsequent
  processing, is faster in these conditions. In the
  above studies, the M350 is only indirectly
  affected by the stimulus manipulations the
  frequency and priming effects really occur in
  some other, earlier, component, not discovered in
  these experiments.

6
BackgroundActivation and Competition

• We make the (standard) assumption that lexical
  processing is a combination of activation and
  competition a stimulus activates its own entry
  and a family of related representations (its
  neighbors), which then compete for selection
  (e.g. Marslen-Wilson et al, 1994).
• The more active entries there are, the more
  competition.
• The more competition, the slower the RTs in a
  lexical decision task (Vitevich and Luce 1999).

• A way to separate lexical activation from
  subsequent processes
• In a lexical decision task, present subjects with
  stimuli which
• are easy to process (because of some stimulus
  property), and hence invoke early lexical
  activation
• activate a lot of competitors, which causes late
  selection/decision.

7
Phonological density/probability

• Vitevich and Luce (1997, 1999)
• Nonwords with common sounds and sequences, i.e. a
  high phonotactic probability (MIDE, PAKE) elicit
  shorter RTs than nonwords with a low phonotactic
  probability (JIZE, YUSH) in tasks such as the
  same-different task or a reading task.
• A sublexical frequency effect The processing of
  high probability nonwords is facilitated because
  they involve feature combinations that are common
  in the language and that are, therefore,
  activated often.

• However, if the task is lexical decision, high
  probability nonwords elicit longer RTs than low
  probability nonwords.
• Because the feature combinations of high
  probability nonwords are common, they activate
  many actual lexical entries, i.e. a dense
  similarity neighborhood. The more representations
  are activated, the longer lexical decision takes
  because it requires determining which, if any,
  actual lexical entry the stimulus matches to.

• High phonotactic probability speeds activation.
• A dense similarity neighborhood slows down
  decision.

8
Hypothesis

• If the M350 reflects automatic lexical
  activation, a stimulus with a high phonotactic
  probability/density should elicit
• a fast M350 but a slow RT.
• If the M350 and RT can be pushed in opposite
  directions for the same stimulus in this way, the
  M350 must reflect a process prior to
  selection/decision.

9
Stimuli

• Materials of Vitevich and Luce 1999 converted
  into orthographic stimuli to permit direct
  comparison of the brain responses to previous
  M350 studies (Embick et al 2000, Pylkkänen et al
  2000).
• Four categories of 70 stimuli
• high probability/density words BELL, LINE
• low probability/density words PAGE, DISH
• high probability/density nonwords MIDE, PAKE
• low probability/density nonwords JIZE, YUSH
• All stimuli were monosyllabic, and high and low
  density words were matched for visual word
  frequency.
• The measures for phonotactic probability were
  positional segment frequency and biphone
  frequency and phonological neighbors were defined
  as any item that could be converted to the
  stimulus by one phoneme substitution, deletion,
  or addition in any position, as in Vitevich and
  Luce 1999 .

10
Procedure

• Task Lexical Decision
• Stimuli were presented in a randomized order in
  two blocks of 140 stimuli with an intervening
  pause allowing subjects to rest.
• Subjects made lexical decisions on the stimuli by
  pressing a button. In block 1, word-decisions
  were indicated with the index finger and non-word
  decisions with the middle finger. In block 2,
  this was reversed. Subjects were instructed to be
  as fast and as accurate as possible.

• MEG Recording
• During the experiment, subjects lay in a dimly
  lit magnetically shielded room in the KIT/MIT MEG
  laboratory while neuromagnetic fields were
  recorded using a 64-channel axial gradiometer
  whole-head system (Kanazawa Institute of
  Technology, Japan).

500 ms
time
LINE
real word?
500-1500 ms

500 ms
JIZE
real word?

• Subjects
• Nine right-handed (2 F, 7 M), English-speaking
  adults with normal or corrected-to-normal vision

11
Results
P lt 0.05
P lt 0.05
P lt 0.05
P lt 0.05

• As predicted,
• RTs to high probability/density stimuli were
  slower than RTs to stimuli with low
  probability/density both in the word and nonword
  conditions (P lt 0.05).
• Probability/density had the opposite effect on
  M350 latencies the M350 peaked approximately 20
  ms earlier in the high probability/density
  conditions (P lt 0.05).

12
Effect of phonotactic probability on the M350
(positive and negative signal maxima for one
subject)
M350 positive maximum
M350
M350 negative maximum
M350
13
Discussion

• The latencies of the M350 reflect RTs as long as
  the stimuli are not varied in a way that affects
  postlexical processing (as shown by the frequency
  and repetition priming studies).
• The M350 and RTs can be pushed in opposite
  directions by stimuli that facilitate lexical
  activation but make subsequent processing, such
  as selection/decision, harder.
• The M350 cannot reflect the post-lexical
  word/nonword decision since in that case M350
  latencies should reflect RTs also in this type of
  manipulation.
• Further predictions of the M350 lexical
  activation hypothesis
• The M350 should be sensitive both to phonological
  and semantic priming (as lexical entries are the
  connection between sounds and meaning).
• The M350 should be the first shared component
  between auditory and visual word processing.
• Hypotheses which associate the M350 with semantic
  processing or integration (following much of the
  N400 literature) do not predict (i) or (ii) nor
  the results of the present study.

14
Summary From stimulus onset to RT
M350 The first MEG component sensitive to
manipulations of stimulus properties affecting
lexical activation. Working hypothesis this
component reflects automatic spreading activation
of the lexicon at signal maximum all the
competitors are activated.
stimulus
RT
BELL
M250 A component between the M180 and M350. Also
insensitive to variations in stimulus properties
that affect lexical access. Clearly distinct from
the M350 as these two responses have opposite
polarities. Processing of orthographic forms?
Postlexical processes including the word/nonword
decision of the lexical decision task.
M180 A visual response unaffected by stimulus
properties such as frequency (Hackl et al, 2000),
repetition (Sekiguchi et al, 2000, Pylkkänen et
al 2000) and phonotactic probability/density.
Clearly posterior dipolar pattern.
15
References

• D. Embick, M. Hackl, J. Schaeffer, M. Kelepir
  and A. Marantz. A magnetoencephalographic
  component whose latency reflects lexical
  frequency to appear in Brain Research.
• P. Helenius, R. Salmelin, E. Service, J. F.
  Connolly, Semantic Cortical Activation in
  Dyslexic Readers in Journal of Cognitive
  Neuroscience, 115, pp. 535-550. 1999.
• S. Kuriki, Y. Hirata, N. Fujimaki and T.
  Kobayashi, Magnetoencephalographic study on the
  cerebral neural activities related to the
  processing of visually presented characters in
  Cognitive Brain Research, 4, 1996, 185-199.
• L. Pylkkänen, A. Stringfellow, E. Flagg, A.
  Marantz. A Neural Response Sensitive to
  Repetition and Phonotactic Probability MEG
  Investigations of Lexical Access. Proceedings of
  Biomag 2000, 12th International Conference on
  Biomagnetism. 2000.
• T. Sekiguchi, S. Koyama and R. Kakigi, The
  effect of word repetition on evoked magnetic
  responses in the human brain in Japanese
  Psychological Research, 42, 2000, 3-14.
• M. Taft, Reading and the mental lexicon. Hove,
  England Lawrence Erlbaum Associates, 1991.
• M. S. Vitevitch and P. Luce, Probabilistic
  Phonotactics and Neighborhood Activation in
  Spoken Word Recognition in Journal of Memory and
  Language 40 374-408, 1999.
• M.S. Vitevitch, P. A. Luce, J. Charles-Luce and
  D. Kemmerer, Phonotactics and syllable stress
  Implications for the processing of spoken
  nonsense words in Language and Speech, 40,
  47-62, 1997.
• Correspondence to liina_at_mit.edu
• Poster available at http//web.mit.edu/liina/www