Detailed results from low-pass-filter-plus-speech-compression stimuli transcription
To reiterate from the body of the article:
The transcriber was a graduate student at the University of Freiburg
visiting the Center for Research in Language for two months. He was told
that he should transcribe each sentence he hears as accurately as
possible. He was also told that the sentences always describe an animal
or animalss doing bad action to another animal/s; these instructions
were in keeping with what experimental subjects were told about the
task. The transcriber first heard the 8 practice sentences in normal
presentation, then the same 8 sentences in LPC presentation.
All sentences were presented in pseudo-randomized order. For each
sentence, experimenter (F.D.) read the names of agent and patient
animals in randomized order. Each sentence was played as many times as
needed for the transcriber to feel comfortable with his transcription.
The transcriber typed the responses into a text file; he was told not to
worry about punctuation or capitalization, but to correct for spelling.
The transcriber was blind to the purpose of the experiment from which
the stimuli were drawn, or the purpose of the transcription itself.
Results, reported by type of error:
I) Interestingly, the vast majority of errors involved exchanging one
verb stem for another.
Total exemplars of each verb:
beissen 13 fressen 10 jagen 12 kneifen 12 schlagen 12
stossen 12 toten 13 verletzen 12
1) kneifen - substituted: a) toeten (10) b) verletzen (1) c) treten (1)
2) fressen - substituted: a) stossen (1)
3) beissen - substituted a) fressen (6) b) stossen (1)
4) jagen - substituted a) moegen (3) b) toeten (6) c) schlagen (1) d) laemen (1) e) rufen (1)
5) schlagen - substituted a) toeten (2)
6) verletzen a) fressen (4) b) toeten (1)
II) A much smaller error category was errors involving number on the
noun stem (3 errors total). Because nouns were named before the sentence
presentation, these were either typos, memory errors, or true
transcription errors. Errors involving nouns are surrounded by astericks
(*x*), while other errors are surrounded by crosshatches (#x#). All noun
errors were of omission - the correct noun declension in indicated in
parentheses
1) 'Die Ratte wird von de#r# Schildkroete*(n)* gefressen'
2) 'Die Spinne*(n)* toete*t* die Henne
3) 'Es ist die Fliege die die Spinne*(n)* frisst
III) An equally small number of errors in verb morphology were made;
some of these involved the same sentences as the noun errors. Errors in
verb morphology are indicated in astericks, while other errors are in
crosshatches.
1) 'Die Ratte wird von de*r* Schildkroete#(n)# gefressen (incorrect
inflection of 'die' because of number error in noun - same as (1) above)
2) 'die Spinne#(n)# toete*t* die Henne' ('n' was deleted, 't' should
have been 'n')
3) 'Es ist die Biene die die Ziegen toete*t* ('t' should have been 'n')
4) 'Die Schlange wird von den Spinnen getoete*t*' (because of the error
in verb stem, the wrong inflection was used)
The possible number of morphological errors per sentence type are
indicated here:
Active: Die Ziege beisst die Schlange (3) Subject Cleft : Es ist die
Ziege, die die Schlange beisst (4) Object Cleft : Es ist die Ziege, die
die Schlangen beissen (4) Passive: Die Ziege wird von der Schlange
gebissen. (5)
Total possible number of errors (3x24) + (4x24) + (4x24) + (5x24) = 384