Neural mechanisms of spoken word recognition: Onsets and inhibition

1
Neural mechanisms of spoken word
recognition Onsets and inhibition
LSA02

• Liina Pylkkänen Alec Marantz
• KIT/MIT MEG Laboratory
• Dept. of Linguistics and Philosophy, MIT

2
Problem of recognition
LSA02

• How does a single mental representation get
  selected as the optimal match to a stimulus?

Stimulus TURN
3
Behavioral inhibition
LSA02

• Words are often harder to recognize when they are
  preceded by similar sounding words (e.g. Rastle
  et al. 2000)

slower when preceded by
SPINACH
SPIN
than when preceded by
MUFFLER
4
Behavioral inhibition
LSA02

• Words are often harder to recognize when they are
  preceded by similar sounding words (e.g. Rastle
  et al. 2000)

slower when preceded by
SPINACH
SPIN
5
Inhibited activation
LSA02
activation level
time
time
RT
SPIN
s p i n a c h
PRIME
TARGET
6
Alternative Inhibited recognition
LSA02
activation level
time
RT
SPIN
s p i n a c h
PRIME
TARGET
7
Mechanisms of recognition
LSA02

• Inhibited activation
• Mismatching candidates are suppressed below their
  resting level
• Inhibited recognition
• Mismatching candidates are rejected simply
  because they receive less excitation from the
  input
• BUT Both predict delayed RTs.
• Difficult to distinguish behaviorally

8
Timing of activation
LSA02
INHIBITED ACTIVATION
INHIBITED RECOGNITION
9
Previously
LSA02

• Magnetoencephalographic component at 350ms
  post-stimulus onset which indexes initial
  excitation of lexical entries
• ? M350
• (Embick et al. Cognitive Brain Research 2001,
  Pylkkänen et al. Brain and Language, 2001)

10
Magnetoencephalography (MEG)
LSA02
http//www.ctf.com/Pages/page33.html
http//www.ctf.com/Pages/page33.html
11
M350 a particular magnetic field distribution
at 300-400ms post onset of a visual word
LSA02
12
M350 the 1st MEG component sensitive to lexical
factors
LSA02

• Frequency (Embick et al 2001)
• Frequent word elicit shorter M350 latencies than
  infrequent words
• Repetition (Pylkkänen et al 2001a)
• Repeated words elicit shorter M350 latencies than
  nonrepeated words

13
M350 NOT modulated by factors affecting the
speed of recognition
LSA02

• Pylkkänen et al. (2001b)
• Would stimuli that invoke fast activation but
  late recognition elicit shortened or lengthened
  M350 latencies?
• M350 latencies are shortened while lexical
  decision times are delayed
• M350 indexes initial activation of lexical
  entries

14
M350 a tool for investigating inhibitory
mechanisms in word recognition
LSA02
INHIBITED ACTIVATION
INHIBITED RECOGNITION
15
Materials

• Crossmodal priming paradigm
• SOA Duration of prime
• Task Lexical decision

16
Materials

• Two types of phonological similarity (embedded in
  a larger experiment)

1. ONSET-MATCHING
2. NON-ONSET-MATCHING
17
Role of onsets in recognition?

• The Cohort model (Marslen-Wilson Welsh 1986)
• Activation purely onset-driven
• TRACE (McClelland and Elman)
• Activation is continuous
• Today
• abundant evidence for continuous activation but
  also for the special role of onsets
• How are onsets special?

18
Materials
AUDITORY PRIME
s p i n a ch
1. ONSET-MATCHING
REACH
t e a c h e r
1. NON-ONSET-MATCHING
19
Materials
AUDITORY PRIME

• Would the M350 show inhibition or priming?
• If inhibition, activation is inhibited.
• If priming, RT inhibition originates in
  competition.

s p i n a ch
1. ONSET-MATCHING
REACH
t e a c h e r
1. NON-ONSET-MATCHING
20
Materials
VISUAL TARGET
AUDITORY PRIME
SPIN
s p i n a ch
1. ONSET-MATCHING
SPIN
m u ff l e r
REACH
t e a c h e r
1. NON-ONSET-MATCHING
REACH
21
Results
22
Results
23
Results
24
Results
25
spinach-spin inhibited activation?
onset-matching competitors are suppressed
26
teacher-reach inhibited recognition?
non-onset-matching competitors NOT suppressed
27
Same behavior but different neurophysiological
effects

• Cognitive level of similarity induced inhibition
  depends on the type of phonological similarity
  between the competitor entries
• The speech recognition system has a mechanism of
  complete deactivation that targets onset-matching
  incorrect entries.
• Non-onset-matching entries are activated, but not
  suppressed

28
Asymmetry predicted by the suppression mechanism

• Isel and Bacri 1999
• trombone RIB PRIMING (via BONE)
• cargo BUS NO PRIMING (via CAR)

29
Onsets in acoustics and phonology

• Resist neutralization and are perceptually
  salient.
• If word onsets play a special role in word
  recognition, the mechanisms of the speech
  perception system can potentially shed light on
  the distribution of these phonetic and
  phonological processes (Gow, Melvold, Manuel,
  1994 Frisch, 2001)

30
Results compatible with

• Marslen-Wilson (1993)
• non-onsetmatching representations get activated,
  but only for repair purposes
• i.e. the mispronunciation shigarette activates
  cigarette but not to a degree sufficient for word
  recognition

31
Acknowledgments

• Andrew Stringfellow, MIT
• - Lab manager
• Laura Gonnerman, USC
• - Majority of crossmodal materials