Experimental study of morphological priming: evidence from Russian verbal inflection

1
Experimental study of morphological priming
evidence from Russian verbal inflection

• Tatiana Svistunova
• Elizaveta Gazeeva
• Tatiana Chernigovskaya
• St. Petersburg State University

2
Structure of the mental lexicon

• Some of the major questions concerning the
  organization of the mental lexicon
• What are the representations of the words in the
  mental lexicon?
• How are the lexical items processed, stored and
  accessed in the mental lexicon?

3
Inflectional morphology

• There is another question closely related to the
  previous ones
• Are the complex / inflected words stored together
  with their stems or as separate lexical entries?
• Experimental studies of inflectional morphology
  can provide us with the answers to this question

4
Inflectional morphology

• In the present study we focus on the verbal
  inflectional morphology.
• But the implications of such studies can lead us
  to more general conclusions about the structure
  of the mental lexicon as a whole.

5
Inflectional morphology

• Thus the aim is to find out if the opposition
  regular vs. irregular inflection is reflected
  in the mental lexicon.
• In other words is there any difference between
  regularly and irregularly inflected words on the
  deep level in our mental lexicon?

6
Structure of the mental lexicon the main
approaches

• The theoretical framework for the study comes
  from research on the structure of the mental
  lexicon, which focuses predominantly on English
  regular and irregular past-tense inflection and
  on the debates between
• Dual-system approach (DS)
• Single-system approach (SS)

7
Dual-system approach

• The proponents of dual-system approach (e.g.
  Pinker and Prince (1988), Ullman (2004)) claim
  that there are two distinct systems of processing
  in our mental lexicon
• Regular forms are composed by their constituents
  (stems and affixes) by symbolic rules.

8
Dual-system approach

• Full Irregular forms are retrieved from
  associative memory.
• If we need to generate a form from a nonce-word
  we automatically use the regular (default) rule.

9
Single-system approach

• The advocates of single-system approach (e.g.
  Plunkett Marchman (1993), Bybee (1995)) argue
  that all the word forms are stored as separate
  lexical entries, both regularly and irregularly
  inflected forms are stored in our associative
  memory
• If we need to generate a form from a nonce-word
  we use analogy which is developed during the
  subjects language experience

10
The role of the frequency

• DS if we process a regular verb, the frequency
  will not influence the RTs (reaction times),
  since in this case symbolic rules are applied
  automatically. If we process an irregular verb,
  the role of the frequency will be significant,
  because all the irregular forms are retrieved
  from associative memory
• SS according to this approach, the more frequent
  the verb is the faster RT it will elicit, no
  matter what class the verb belongs to (regular or
  irregular) so the verb frequency is significant

11
Morphological priming

• The main idea of the priming method is that if
  the words / word forms are related / connected in
  the mental lexicon, then it should be much
  easier( faster) to retrieve a form / word from
  the memory if it was pre-activated by the related
  word.

12
Morphological priming

• Usually the priming stimuli are presented as
  pairs where the first word is a prime and the
  second is a target.
• For example, we have several pairs of words
• walk walk walked walk put walk

13
Morphological priming

• Obviously, the prime and the target from the
  first pair are stored within the same lexical
  entry (walk walk), the prime and target from
  the third pair seem to be unrelated
  (morphologically and semantically put walk).

14
Morphological priming

• If we compare the RTs of this 2 pairs, well
  certainly get faster RTs for the first pair,
  because the target has been already pre-activated
  by the previous presentation of the related prime.

15
Morphological priming

• Lets look at the second pair (walked walk).
• When we get the RTs for the target, we can
  compare it with the RTs from the 1st and the 3rd
  pair.
• If its closer (statistically significant) to
  the 1st, then walked walk are stored together,
  if to the 3rd, then they are unrelated and stored
  separately.

16
Morphological priming DS and SS predictions

• The proponents of these two models make different
  predictions about the results of the
  morphological priming experiments with regular
  and irregular verbs.

17
Morphological priming DS predictions

• Regular verbs since such forms as walked walk
  are stored together and the inflected form is
  computed by the symbolic rules, the presentation
  of walked will lead to significant priming
  effect (no matter how frequent the verb might be).

18
Morphological priming DS predictions

• Irregular verbs since such forms as taught
  teach are stored separately, the presentation of
  taught will not prime teach as strongly as
  the verb itself (teach teach).
• In this case frequency is what counts if the
  verb is frequent it has more chances to prime the
  target

19
Morphological priming SS predictions

• Regular and irregular verbs there is no
  distinction in their processing.
• The most important role plays frequency the
  stronger connections between the items in the
  associative network are the faster responses we
  get (no matter if its a regular or irregular
  verb).

20
Verbal inflection data across languages

• The majority of the studies of regular vs.
  irregular inflection were conducted on English
  and German material.
• So far we have data on such languages as
  Norwegian, Spanish, Catalan, Italian, etc.

21
Verbal inflection data across languages

• But the data are still inconclusive.
• Russian could provide us with important findings,
  since its the language with complex inflectional
  morphology with more than one regular and
  irregular class.

22
Russian verbal inflection

• According to the classification by Jacobson
  (1948) and his followers there are 11 verbal
  classes in Russian without opposition regular
  vs. irregular.
• A series of experimental works (e.g.
  Chernigovskaya et al. (2008) showed that there is
  a default j-rule (adding the suffix -j- to
  the stem) which is applied during the processing
  of nonce and unfamiliar verbs.

23
Russian verbal inflection

• The authors propose that if the 2 rules compete
  (default -j- rule and non-default) then such
  factors as token and type frequency are
  important.
• In other words, the authors claim that its not
  only one factor which determines everything, but
  their complex interaction token and type
  frequency and class, productivity of the verb.

24
The present study

• The aim of this pilot study was to find out what
  is the interaction between 2 factors of frequency
  and verb class in the morphological priming study
  using Russian material.

25
The present study

• Also to compare our data on Russian with the
  findings of the experiment on German carried out
  by Sonnenstuhl et al. (1999), which we took as a
  basis for our research.
• We used the same technique and made up the
  stimuli set according to principles used in the
  study).

26
Experimental material

• Stimuli 20 experimental triplets with regular
  verbs (-aj- class) and 21 with irregular verbs
  (-a- class). Each triplet consisted of 3
  conditions
• Identical condition delat delat (to do
  to do).
• Experimental condition delaju delat (I do
  to do).
• Control condition kurit delat (to smoke
  to do).

27
Experimental material

• All the triplets were divided into 3 classes of
  frequency (high, medium, low).
• Besides the experimental triplets the stimuli set
  consisted of filler pairs
• 243 pairs with existing Russian verbs as primes
  and targets (morphologically/ semantically
  unrelated)
• 243 pairs with existing Russian verbs as primes
  and nonce-verbs (infinitives) as targets.

28
Experimental material

• Then 3 versions were constructed with only one
  pair from each experimental triplet (because no
  participant should see a verb twice).

29
Method and participants

• Participants 20 native speakers of Russian (15
  women and 5 men). Mean age from 18 to 45 y.o.
• The presentation of the stimuli set and RTs
  measurement were operated by PsyScope program
  (http//.psy.ck.sissa.it)

30
Method and participants

• Experimental task lexical decision task
  word/non-word.
• Auditory primes were immediately followed by
  visual targets. The measurement of the RTs began
  with the presentation of the target.
• Subjects had to react only to the targets on the
  screen by pressing a button.

31
Method and participants
32
Results

• ANOVA with repeated measures shows that the
  following factors influenced RTs
• condition (control, experimental or identical)
• token frequency (high, medium or low)
• verb class (-aj- vs. -a-).
• So we observed a priming effect.

33
Results

• Then two separate paired t-tests were running to
  compare three conditions for verbs of different
  classes and different frequencies.
• They show that the effect of partial priming
  (statistic difference between control and other
  conditions) were observed for verbs of the
  regular -aj- class like in Sonnenstuhl et al.
  (1999) and for verbs of medium and low frequency.

34
Discussion

• On the one hand, the fact that the different
  priming effects were observed for regular and
  irregular classes agrees with the predictions of
  the DS approach.
• On the other hand, the influence of the frequency
  factor agrees with the predictions of the SS
  approach
• Also our results differ from those obtained by
  Sonnenstuhl et al. (1999), where the data
  coincided with the predictions made by DS.

35
General discussion

• Our results show that high frequency verbs are
  processed faster than those of lower frequency
  not depending on their verb class (regular or
  irregular).
• But medium and low frequency verbs elicit faster
  RTs if they belong to the regular productive
  class.
• It means that our data conflict with both major
  models of the mental lexicon.

36

• THANK YOU FOR YOUR ATTENTION!