evidence for morphological decomposition

Morphological speech errors on agentive and comparative affixes
Dirk Janssen, Universität Leipzig &
Karin Humphreys, University of Illinois at Urbana-Champaign
Morphological speech errors
Intended
Produced
my anal retentiveness
→ my anive rententalness
K. Humphreys (pc)
I thought the truck was parked → I thought the park was trucked
Garrett, 1980
•• Experiment 2
2500ms
sunshine worker
reference teach
SWAP
sunshine ?
reference ?
3500ms
cue to speak
correct:
“sunshine teach, reference worker”
example morphological error:
“sunshine teacher, reference worker”
The versatile, productive English -er
comparative adjective nice+er
agentive
work+er
pseudo-morph
summer
•• Predictions
If derivational morphemes are stored independently
from their stems, more morphological speech errors
should occur when -er is a real morpheme than when
it is only a pseudo-morpheme.
comparatives = agentives > pseudo-morphs
Errors were elicited from 72 participants. The morphological category of the two critical words in the quad
varied, forming either:
comparatives
sunshine nicer
reference warmer
sunshine worker
reference teacher
sunshine summer
reference badger
agentives
pseudo-morphs
Is this derivational decomposition we observed above
taking place at a phonological, or morphological level?
If morphological, these affix errors should be sensitive
to the particular role the -er is playing. If phonological,
affix role should not matter. Experiment 2 tested sensitivity to affix role by examining whether affix errors
could occur across morphological category (eg whether
nicer/work would lead to the error nicer/worker ).
Errors were elicited from 120 participants. Affix pattern
(one vs. two affixes) varied as before. Morphological
congruity now varied, forming either:
•• Experiment 1
Same (identical to exp 1)
nicer/warmer
comparative
worker/teacher
agentive
summer/badger
pseudo-morph
Different (new)
nicer/teacher
comparative/agentive
nicer/badger
comparative/pseudo-morph
worker/badger
agentive/pseudo-morph
Addition or omission of an -er created words in all
conditions. The affix pattern of the two critical words
within the quad varied, with either:
one affix
two affixes
worker/teach, work/teacher
worker/teacher
Experiment 1
Morphological Errors
40
50
5
30
20
two affixes
one affix
40
13
36
4
2
2
10
29
10
•• Method
17
5
comparative
0
same morphological type
Two additions to this model are required: First, a lexical constructor, or LC, which merges morphemes into
a word form. Under conditions such as created in our
experiment, the LC sometimes misaligns morphemes.
Secondly, the LC is used when the speaker wants to
create a new word (eg a thing/person who structures
→ a structurer). For this, each affix needs to be represented multiple times, tagged for its role (agentive,
comparative, etc).
In everyday production, tagged affixes discourage certain speech errors: In “anive rententalness”, an is not
a good base for -ive[Latinate Noun→Adj]. In “trucked
the park”, truck can satisfy -ed[Verb, +past tense]. In
our experiment, tagged affixes can explain the lower incidence of mixed role, one-affix errors in Experiment 2.
mixed morphological type
The affix -er was more prone to errors when it was
a derivational morpheme (nicer or worker ) than when
it was only a pseudo-morpheme (summer ). Most errors occurred when combining a stem with an affixed
form (affix priming), but this does not hold between
different morphological roles (eg when combining comparative and agentive). No evidence for differing morphological status of comparatives and agentives was
found. These experiments argue for some form of decomposition of derivational morphemes, at a level that
is sensitive to morphological role.
4
0
We elicited morphological errors from speakers by presenting a series of word quads, made up of two pairs.
Speakers read these words silently, after which they
disappeared. Speakers were occasionally cued to re-
12
comparative
pseudo-morph
10
20
19
14
comparative
agentive
two affixes
one affix
pseudo-morph
pseudo-morph
30
agentive
agentive
5
comparative
comparative
number of errors
15
Secondarily, theories disagree as to whether the comparative -er is an inflection or a derivation. If an inflection, it might have more errors than the agentives.
comparatives > agentives > pseudo-morphs
The higher error rates for nicer and worker support a
model of the lexicon in which morphemes are the units
of storage. The lemma contains numbered pointers to
the morphemes. (cf. Levelt, Roelofs & Meyer 1999)
•• Summary
Experiment 2
Morphological Errors
10
40
⇒ Crucially, there were far fewer affix errors when the
affix roles were mixed, indicating that these errors are
sensitive to affix role. The number of errors for one and
two affix conditions was the same when the roles were
different: affix priming does not seem to occur here.
•• Model
read silently
auditory deadline cue
General logic: If a morpheme can move independently
during a speech error, it is a production unit, with a
separate, independent representation at some level. Inflectional morphemes (eg -ed, -ing, -s) are much more
error prone than derivational morphemes (eg -ness, able, -ion), (Garrett, 1980; Humphreys, 2002). From
this and other data, it is argued that at a syntactic,
or lemma level, inflections are decomposed but derivations are not. Can we also find experimental speech
error evidence of derivational decomposition?
dicted. Errors occurred primarily in the one-affix condition, i.e. an affix priming effect. No evidence was
seen for a difference in morphological status between
comparatives and agentives.
agentive
pseudo-morph
Are morphologically complex words stored as wholes,
or as morphemes? We present two experiments using elicited speech errors that demonstrate derivational
morphological decomposition in language production.
The data also shed light on the inflection–derivation
distinction.
spond aloud, as quickly as possible, to the immediately
preceding item. Speakers had to then either repeat the
previous phrases exactly, or had to swap words between
the phrases. The critical pairs each consisted of an initial filler word plus a critical word stem or stem+affix.
number of errors
•• Introduction
agentive
pseudo-morph
⇒ More affix errors occurred on the morphologically
complex forms than on the pseudo-morphs, as pre-
When affixes had the same role, there were more morphological errors for comparatives and agentives than
for the pseudo-morphs. More errors occurred in the
one-affix than the two-affix condition. This replicates
Experiment 1.
Email to [email protected] or [email protected]. Formated with LATEX and printed at the
Universitätsrechenzentrum Leipzig.