INTRODUCTION
The effects of frequency and morphological complexity
 on aphasic speech production
Three Main Arguments
results:
LESS FREQ.           4%                      10%                     21%                    15%

MORE FREQ.        15%                     6%                      18%                     10%

TOTAL                 19%                     16%                     39%                    25%
BASE FORM          13%                     2%                       18%                     10%

SUFFIXED               6%                     14%                      21%                     15%

TOTAL                 19%                     16%                      39%                     25%
frequency affects the production of irregular 
     forms but not that of regular forms

ARG 1: 
the past tense of regulars sounds
     just as unusual as their stem 
     e.g. 'infarct' ~ 'infarcted'
the past tense of low frequency irregulars 
     may sound more unusual than their stem
     e.g. 'forgo' ~ 'forwent'

ARG 2:
high-frequency irregulars produced faster than low-frequency irregulars
there is no such effect with regulars
Pinker's double dissociation model:
phonologically similar irregular verbs 
     tend to form their past tense similarly
     e.g. 'sing' ~ 'sang', 'ring' ~ 'rang'

no such effect with regulars
     e.g. 'blink' ~ 'blinked'

Rumelhart & McClelland's model cannot
     handle forms like 'rhumba' (~ 'rhumba'd')
regular processes: 
          generative rewrite rules (e.g. SPE)

irregular processes: 
          connectionist networks (e.g.
           Rumelhart & McClelland 1986)
both the base-form and the past tense form help in the recognition of regular base-forms
     
     e.g. 'walk' and 'walked' both prime 'walk'

the past tense form helps less in the recognition of irregular base-forms than the base-form does
     
     e.g. 'sang' primes 'sing' less than 'sing' does
1. Frequency effects
2. Similarity clusters
Hungarian morphology does not meet                      :
the defects of rule-based descriptions
(i) the same verb form as in target sentence

(ii) same verb stem as in target sentence, but
     with different inflection

(iii) different (though semantically or phonologically related) verb stem 

(iv) ill-formed or no verb in answer
Kaye (1995) Derivations and Interfaces
but
Rebrus et al. (1996)
sfvl
conclusions
rule-based approaches face difficulties both in theoretical and descriptive respects

our results show that frequency plays a role as much as complexity

frequency not only affects successful repetition but also the choice of a replacement - the most frequent member of the target verb's paradigm
vh
proposal:
brought to you by ZuiPrezi
anya ~ anyában (mother, inessive)
remete ~ remetében (hermit, inessive)

anya ~ anyaság (mother, motherhood)
remete ~ remeteség (hermit, hermithood)

anya ~ anyaszerü (mother, motherlike)
remete ~ remeteszerü (hermit, hermitlike)
regular morphology created by concatenation, 
irregular morphology stored in the lexicon
example: past tense of English verbs

           seep ~ seeped   [[si:p]t] 
           
           keep ~ kept       [kept]

'seeped' violates the constraint prohibiting superheavy syllables with non-coronals  
Kaye has further arguments
the experiment
MORE FREQ.    hallgat 7162            húzza 5567
                       (vs. hallgattuk1066) (vs. húz 3788)

LESS FREQ.      javít 1173                nyílnak 3586
                       (vs. javítja 6110)      (vs. nyílik 28922)
description:
1 participant (well, it was a pilot)
48 sentences to repeat
verbs from the 'Szószablya' webcorpus (machine-tagged)
the relative lemma frequency of the verbs is identical
     (around 54000)
base form featuring in test



other V form in test
(The difference between the relative frequency of the most frequent and a chosen less frequent form was always about the same, irrespective of whether the base form was the most frequent or not.)
four groups:

base form
(3rd/Sing/Present/Ind/Intr) 
most frequent

base form less frequent
FOUR CLASSES
reply sentences divided into
thank you for your attention!
IDENTICAL     CHANGED SUFFIX     CHANGED STEM     AGRAMMATIC
+unmarked        -unmarked
+frequent         +frequent

+unmarked        -unmarked
-frequent            -frequent
IDENTICAL     CHANGED SUFFIX     CHANGED STEM     AGRAMMATIC
Péter Rácz & Márton Sóskuthy

HAS-ELTE-Hungarian Zoological & Botanical Garden
BASE FORM             INFLECTED FORM
The following is a comparison of
Pinker's Double Dissociation Model
and Rumelhart and McClelland's connectionist model in an 
experimental situation.

The experiment involves 
an agrammatic aphasiac 
performing sentence 
repetition tests.

Connectionist approaches 
predict that frequency plays 
an important role in the success
 of the repetitions, while
 Pinker's model predicts 
that it does not.

Rumelhart and McClelland's 
model also has something 
to tell us about the 
patient's repair strategies.
[verb]
[past]
-ed#
vowel change
[verb]
[past]
[ing#]
spling
rhumba
3. Priming effects
regular past tense forms contain the base-form

irregular Past Tense Forms do not...
comparing (i) with (ii-iv) shows that frequent forms and base forms are repeated more successfully 

base forms win to 13:6, frequent forms to 15:4 in identical repetition (i), trends turn in class (ii) (less spectacularly in cll (iii-iv))
new aspects of classification: with respect to frequency and complexity
all suffixed verbs in class (i) are more frequent than their base forms

the base forms in class (i) are all monosyllabic (except one), (cf. the fact that non-frequent non-monosyllabic base forms were not repeated)

base forms are a minority in class (ii) (2:14) (invulnerability)

cll (iii-iv) show no clear splits (don't believe the numbers)
yet class (iii) is more populous than class (ii) (39:16) - hints at paradigms organised from the side of the suffix
some further interesting trends in classes (ii)-(iii):
class (ii) – suffix changes:

         low-freq form > high-freq form: 5 (10%)
         high-freq form > low-freq form: 1 (2%)

         suffixed form > base-form:         1 (2%)
         suffixed form > suffixed form:     5 (10%)

this matches the predictions of the connectionist model
class (iii) – stem changes:

         lemma freq (ABANDON > LEAVE):
         low-freq lem > high-freq lem:     9 (19%)
         high-freq lem > low-freq lem:     9 (19%)

         type freq (abandoned > left):
         low-freq form > high-freq form:  7 (15%)
         high-freq form > low-freq form:  11 (23%)

         morph complexity:
         suffixed form > base-form:         2 (4%)
         other:                                       16 (34%)

frequency and morph complexity both seem irrelevant here