Uploaded by AlinaV

Libben and Weber2014

advertisement
See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/352165394
Semantic transparency, compounding, and the nature of independent variables
Chapter · January 2014
CITATIONS
READS
0
390
2 authors, including:
Gary Libben
Brock University
6 PUBLICATIONS 97 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Lexical Processing View project
All content following this page was uploaded by Gary Libben on 06 June 2021.
The user has requested enhancement of the downloaded file.
Libben, G. & Weber, S. (2014). Semantic transparency, compounding, and the nature of independent
variables. In Rainer, Franz, Dressler, Wolfgang U., Gardani, Francesco, Luschützky, Hans Christian,
(Eds.) Morphology and meaning. Amsterdam: Benjamins.
1
Semantic transparency, compounding, and the nature of
independent variables
Gary Libben1 and Silke Weber2
1Brock University and 2University of Calgary
We report on a psycholinguistic study of semantic transparency among
English compound words.1 We employed the P3 technique (Libben et al.
Text
2012), which combines elements of three
experimental paradigms: progressive
demasking, naming, and word typing. Both the progressive demasking data and the
word typing data showed a graded effect of semantic transparency associated with
the number and location of semantically opaque constituents in the compound.
Typing duration effects were evident at the constituent boundary, supporting
observations first made by Sahel et al. (2008). We suggest that these data point to the
value of letter typing durations in the analysis of morphological processing and the
importance of a psychocentric perspective on lexical processing – one that
emphasizes the psychological nature of morphological structures.
Keywords: Morphosemantic transparency, compounding, psychocentric approach,
P3 technique.
1. Introduction
A key reason for the success of the field of psycholinguistics over the past quarter
century has been its effective deployment of rigorous scientific techniques of
experimentation and analysis. To a large extent, this has involved treating properties
of linguistic stimuli as independent variables. As a result, studies are often
conceptualized as investigating this or that stimulus property (e.g., frequency,
abstractness, lexical category, morphological structure) on processing.
This approach has been instrumental in advancing the field. However, it also carries
with it limitations that can stand in the way of progress toward a deeper
understanding of language processing in general and a deeper understanding of
lexical representation and processing, in particular.
In this paper, we explore the consequences of a psychocentric perspective on lexical
processing that emphasizes the extent to which language structures are essentially
psychological in nature. We consider the consequences of this perspective for
matters of the representation and processing of compound words in the mind and for
issues of semantic transparency in general. Finally, we present the results of an
experimental investigation of the manner in which English compounds are processed.
The study employs the P3 technique (Libben, Weber & Miwa, 2012) that combines
aspects of word recognition and word production. The experimental paradigms that
make up the P3 technique yield data patterns in which the distinction between
2
independent variable and dependent variable is not easily discernable. We claim that
these characteristics increase the ecological validity of the experimentation and thus
facilitate the capacity of the experiments to target core properties of lexical
processing.
2. A Psychocentric approach
The approach that we have taken in this report is related to a larger perspective on
language processing that has been termed “psychocentricity” in Libben (2010). In
this approach, properties of linguistic stimuli, and words in particular, are not viewed
as external entities, but rather as properties of the state of participants as they engage
in language processing tasks. Psychocentricity affects how we think about linguistic
constructs in general and how we think about linguistic stimulus characteristics as
putative independent variables in psycholinguistic experiments.
To the best of our knowledge, the core of the psychocentric perspective was first
articulated by Derwing (1973) who claimed that:
In general, a linguistic unit at any ‘level’ exists as a unit only because
the language user treats it as a unit [and] linguistic structure is not
something that is ‘built up’ out of utterances or ‘overlaid’ upon [them],
but is [rather] something which receives its only possible empirical
realization as part of the language process of speech production and
comprehension.” (p. 305)
In making these claims in 1973, Derwing staked out a very central role for
psycholinguistics in the arbitration of linguistic constructs in general. Linguistic
structures do not belong to languages or to language units. They belong to people.
This perspective forces us to judge the ontological status of constructs in terms of
whether such constructs can be shown to make a difference in speech production and
comprehension.
The psychocentric perspective also opens the doors to an individualized notion of
language unit and linguistic structure. This individualization can make issues of
putative psychological reality much more complex. To begin with perhaps the
simplest example, we know that native speakers of a language differ in the size of
their vocabularies. For many native speakers of English, for example, the strings birm,
barm, and borm are all nonwords. Others will indeed have meanings associated with
these forms (i.e., ‘a sound barrier’, ‘a shelf-like area’, ‘to smear’ respectively). Are they
words? In one sense they are, in that conventionalized meanings already exist for the
forms. But for the many individuals who do not share in those conventionalized
meanings, they are not words.
2.1. Psychocentricity and morphological transcendence: The case of helicopter
If, as we have suggested above, the wordhood of stimuli can differ from one individual
3
to another, it is likely even more the case that subtler aspects of word morphology
will differ from one person to another. As we move to this domain, we encounter a
core component of the psychocentric perspective – namely that linguistic
characteristics might not only have differing effects on individuals, but also that they
may indeed be constituted within individuals. Thus, notions such a morpheme,
morphological head, morphological constituent, or morphological structure refer to
a set of psycholinguistic representations and processes within individuals rather than
invariant properties of words.
In the context of the stimuli that we focus on in this report, English compound words,
issues of psychocentricity often involve the extent to which particular compound
constituents are part of an individual’s mental processes, whether those constituents
are organized in particular morphological configurations, and how the constituents
relate to the compound and to the whole-word counterparts of the constituents in
terms of meaning.
We are all familiar with straightforward cases in which a word may be considered
linguistically to be morphologically complex but be monomorphemic for some native
speakers. Sometimes this is the result of lexical borrowing, as in the case of the
English word kindergarten, which is straightforwardly the compound ‘child’ + interfix
+ ‘garden’ in German.
Other cases can be more complex. In the paragraphs, below, we discuss the case of
the English word helicopter in some depth, as it reveals interactions between
psychocentric and linguistic properties of stimuli.
The word helicopter is in etymological terms composed of helico- (helix) + pter (wing).
It came into English from the French helicopetère, coined by Gustave de Ponton
d'Amecourt in 1861. Some native speakers of English who have knowledge of Greek
may perceive the compound helicopter to be composed of helico+pter. However, most
native speakers of English will have either internalized the word as having the
structure heli+copter or perhaps perceive it as not having any internal
morphosemantic structure at all.
The example of the compound helicopter allows us to consider a key matter in the
representation and processing of words in the mind from a psychocentric
perspective. From this perspective, it is not easy to say what the actual morphological
structure of helicopter is. We can expect that, for most people, it is a compound
composed of heli- and –copter. This does not negate the fact that it is etymologically
composed of helico + pter. Nor does it negate the importance of the phonotactic
constraints that likely created the pressures to reanalyze helicopter so that it does
not contain the morpheme-initial or syllable-initial consonant cluster /pt/ (Olofsson,
1990). But it is very unlikely that any of this is part of the lexical processing system
of native speakers of English. For most speakers of English, if helicopter is composed
of heli- + -copter, it has always been composed of heli- + -copter.
4
An additional matter that merits consideration is the nature of the morphemes
created by the analysis of helicopter into heli- + -copter. In our use of hyphens on the
right in heli- and on the left in -copter, it is our intention to signify that these should
be seen as positionally bound compound constituents. These positionally bound
elements take on a grammatical function as morphological elements that become the
building blocks of new compounds. It is interesting to note, that because heli- and –
copter were not free-standing morphemes in English before their creation as
compound constituents, they seem to both have as their lexical referent the meaning
of the entire compound helicopter. Thus among the more interesting uses of –copter,
we find holy-copter (a helicopter used by the Pope), bat-copter (a helicopter used by
Batman) and medi-copter (a helicopter used by medical personnel). This last
compound is likely to be interpreted as synonymous with the also existing compound
heli-ambulance. Additional members of the heli- morphological family are heli-pad
(where a helicopter lands) and the semantic cluster heli-skiing, heli-boarding, and
heli-fishing which involve the use of helicopters to transport people to remote
locations for sporting or recreational purposes. A particularly striking new usage is
heli-parent, which is based on the expression helicopter parent and describes overprotective parents who compulsively ‘hover’ over their children.
The case of helicopter highlights some of the properties of productivity in compounds
and, in particular, points to a view in which compound constituents are not simply
morphological arrangements of free standing morphemes, but rather are positionally
bound morphological structures. The positional boundedness is evident in the
productivity patterns noted above, where the elements heli- and –copter have distinct
morphological positions. Thus, whereas sea-copter can be seen as a novel compound,
copter-sea is much less interpretable. Finally, although heli- and –copter can be said
to mean roughly the same things, this equivalence is substantially affected by the fact
that heli- is a modifier and –copter is a morphological head. As a result, copter is the
preferred independent morphological shortening of helicopter.
Our treatment of compound constituents as positionally bound elements follows
from Libben (2010) in which the term morphological transcendence is used to
describe the process by which words become positionally bound compound
constituents. It is claimed that morphological transcendence addresses the
prevalence of semantic opacity in compounding. Within this framework, semantic
opacity develops as a result of lexical processing. It is often the case that, as
compounds are used in language comprehension and production, their constituents
acquire grammaticalized, positionally bound representations in the mental lexicon.
And, it is the natural result of lexical growth in the language that these compound
constituents acquire morphosemantic families. As Singh (2006) has noted,
compounds are rarely created in the manner that helicopter was created—through a
completely novel combination of two morphological elements. Much more common
is the expansion of an existing pattern to include new elements (e.g., blueberry,
blackberry, redberry, yellowberry etc.). When this happens, a certain amount of
semantic bleaching is often present, so that the compound constituent drifts from the
meaning of its monomorphemic whole-word counterpart to a typically more general
5
representation. The development of –man to an almost suffix-like status in English in
words such as chairman and doorman is an extreme example of this process.
3. Semantic Opacity
For compounds, the term semantic opacity can be used to describe a state in which
the meanings or functions of morphological constituents in the compound differ from
their use as independent free standing morphemes such that it is very difficult to
obtain the meaning of the compound from the meanings and arrangement of its
constituents. Under such as view, a compound such as deadline would be considered
semantically opaque because if a language user knew the meanings of dead and line,
but did not know the meaning of deadline, that language user would be unlikely to
correctly guess the meaning of the compound without considerable contextual
support. In contrast, a compound such as bedroom seems to require much less
contextual support for its correct understanding at first presentation.
The issues discussed above bear directly on questions of whether we can conceive of
compound words such as deadline and bedroom as having invariant morphological
properties across people. They also suggest that we think about semantic opacity as
actually being constituted in the psycholinguistic activities of people rather than in
the properties of stimuli. In our view, these are important considerations to bear in
mind, despite the fact that it is both convenient and intuitive to continue to speak
about lexical stimuli as though they were in possession of invariant properties.
Perhaps a brief step into another cognitive domain might provide an illustrative
example. We often speak of pets as having qualities such as being pleasing or being
annoying. But in such cases, aren’t we using these terms as a shorthand to refer to the
states that particular pets trigger in us, perhaps because of the way they look or the
way they behave? We expect that there will be common patterns of triggering, so that
many people will agree that a certain pet is pleasing or annoying. In all cases, though,
“pleasure” and “annoyance” are occurring within people, not within the pets, despite
the fact that it is both convenient and intuitive to continue to speak though the pets
were in possession of those properties.
Now, moving back to matters of semantic transparency in compounds, we note that
it is likely the case that all compounds carry with them some degree of semantic
opacity. As we have stated above, the activities of word comprehension and
production create the conditions for morphological transcendence to occur. In this
process, constituents become entrenched as positionally bound morphemes. So, for
example, the morphemes dead- and -line in deadline acquire a semantic function that
differs from that of the independent words dead and line.
To the best of our knowledge, this phenomenon was first noted by Aristotle in his
Poetics. The phenomenon of semantic opacity was problematic for Aristotle because
it conflicted with his compositional approach to complex language structures. The
solution that Aristotle proposed was that words lose some of their meaning in
6
compounds. The hypothesis of morphological transcendence is consistent with the
Aristotelean observation and situates the cause of the effect in the activities of lexical
comprehension and production. Moreover, the approach allows for a consideration
(Libben, forthcoming) of the manner in which the family composition of a compound
plays a role in the morphological transcendence of its constituents. So, for example,
the entrenchment of -line as a compound head (as opposed to free standing word) is
supported by the presence of compounds such as shoreline, snowline, breadline,
skyline, and even lifeline. The presence of these other members of the morphological
family of -line increases its distinctness from the free standing word. In principal,
this family could create the conditions under which it is easier to guess the meaning
of deadline. However, the fact that the internal semantics of deadline differs from the
others (and interestingly also from lifeline) actually adds to its opacity.
As the discussion above indicates, semantic opacity seems to be a core aspect of
compounding. It is unlikely that we can simply label compounds as either transparent
or opaque. And, it is very likely that we will see considerable variation in the manner
in which participants have internalized such compound words. This, however, does
not make the phenomenon experimentally intractable. Indeed, there has been a great
deal of research that has been focused on semantic properties of compound
processing and on matters of semantic transparency and opacity in particular. These
have included the seminal work of Taft and Forster (1976) and Sandra (1990), and
Zwitserlood (1994). The phenomenon has also been explored more recently in
Libben (2006), Monahan, Fiorentino, & Poeppel (2008), Frisson, S., NiswanderKlement, E., & Pollatsek, A. (2008), and Fiorentino & Fund-Reznicek (2009). Work by
Gagné & Spalding (2009) and Ji, Gagné, & Spalding (2011) have explored both the
effects of semantic transparency and the contributions of patterns of semantic
relations to the manner in which compound words are represented and processed.
In our investigation of semantic transparency, we sought to develop and employ a set
of new techniques for the study of compound processing that would allow us to
examine psychocentric aspects of lexical processing. As we explain below, the P3
technique that we developed (Libben, Weber & Miwa, 2012) cobines techniques of
testing lexical recognition (in this case, progressive demasking) and existing
techniques of testing language production (in this case, the typing of words). The P3
technique combines these to allow us to compare responses to lexical presentation to
patterns of lexical production. While the former can be seen as stimulus driven, the
latter must be seen as an expression of the internal mental states of participants.
As can be seen below, in this study we explored a number of additional opportunities
that the P3 technique affords. These include the ability to split up the progressive
demasking technique across two participants. These two participants then form an
experimental dyad in which one participant sees a word and says it aloud (generating
a lexical recognition response time). The second member of the dyad takes that
‘reading aloud’ as input and then types what he or she heard. This generates another
set of latency data that correspond to letter typing times across the word. For
compound words, such typing durations allow us to measure whether morpheme
7
boundaries are associated with typing pauses.
4. The core stimuli
In order to assess the effectiveness of the P3 technique in addressing the types of
issues we have discussed above with respect to compound processing and the
phenomenon of semantic transparency, we chose to examine a core set of
semantically transparent and opaque English compounds that have already been
studied using the lexical decision task. by Libben, Gibson, Yoon, & Sandra (2003). The
set of 40 English nominal compounds differ in the locus of their semantic
transparency/opacity. Fully transparent compounds contain two transparent
constituents. We will refer to these Transparent-Transparent compounds as TT
compounds. These are words such as sailboat, for which the meaning of the
compound is fully compatible with the meaning of its constituents. At the other
extreme, words such as humbug are semantically Opaque-Opaque (OO). In other
words the meaning of the compound cannot be derived from the meaning of either of
its morphological constituents in any obvious way. Between these extremes are
Opaque-Transparent (OT) compounds, such as nickname. These compounds have a
modifier that does not contribute to the meaning of the compound in any obvious way
and a head that does; the opposite applies to Transparent-Opaque compounds (TO),
such as jailbird.
All compounds employed in this study are bisyllabic, so that the constituent boundary
always occurs after the first syllable. The set of core compound stimuli are provided
in Table 1.
Table 1. The core compound stimulus set. Compounds are classified as TransparentTransparent (TT), Opaque-Transparent (OT), Transparent- Opaque (TO), and
Opaque-Opaque (OO).
TT
bedroom
coalmine
daylight
doorbell
farmyard
fencepost
paintbrush
rosebud
sailboat
schoolboy
OT
chopstick,
crowbar,
dashboard,
godchild,
jackknife,
nickname,
pothole,
shortcake,
strawberry,
sunfish
TO
cardshark,
doughnut,
heatwave,
jailbird,
oddball,
shoehorn,
slowpoke,
sourpuss,
spoilsport,
staircase
OO
deadline,
dingbat,
fleabag,
hallmark,
hogwash,
humbug,
ragtime,
rugrat,
stalemate,
windfall
Using the lexical decision paradigm, Libben, Gibson, Yoon, & Sandra (2003) presented
native speakers of English with these compounds under a number of conditions.
Across all the experiments that they report, TT compounds were easier to process
8
than OT compounds, which in turn were easier to process than OO compounds. They
concluded that the effects of semantic opacity must make reference to the opacity of
individual morphemes and, in particular, to whether they are morphological heads.
This core set of stimuli offered us the opportunity to benchmark the P3 technique.
The core characteristics of the technique are sketched below.
5.
The P3 technique
In this study, we employed the P3 technique (Libben, Weber & Miwa, 2012). This
experimental paradigm has the advantage that it simultaneously yields perception
and production data on morphological processing. Moreover, it provides a means of
introducing interpersonal communication to psycholinguistic experimentation.
The P3 technique has three components, namely the visual presentation of the
stimulus, an oral response and a written response. Initially, participants see words
presented on a computer screen that gradually emerge from a pattern mask.
Participants respond to the visual stimulus by saying the word out loud as soon as
they have recognized it. Subsequently, the participants type the word they have just
said out loud. It is possible to record data on a variety of dependent measures,
including naming latencies, typing onset latencies, inter-key intervals and typing
accuracy. Thus, the P3 technique combines a modification of the Progressive
Demasking technique developed by Grainger and Segui (1990) with writing
production task of the type used by Sahel, Nottbusch, Grimm & Weingarten (2008)
and Will, Nottbusch & Weingarten (2006). A key advantage of the Progressive
Demasking technique is that provides a measure of online visual word recognition
without requiring the use of nonwords (in contrast to the lexical decision task).
Since we employed the P3 technique for the first time, we tested participants on two
core and two control conditions. The core condition referred to as the individual
version yields perception and production data from one and the same participant.
The other core condition involves interpersonal dictation. In this version, the oral
response of one participant serves as input for another participant. The visual control
condition on the other hand does not require an oral response. The participants
indicate their recognition of the visual stimulus by pressing a key. Then they type the
word. In the computer audio condition the written response is based on auditory
input alone. However, the auditory input is produced by a computer voice and not by
another participant.
The combination of these experimental setups provides an opportunity to evaluate
the robustness of effects of semantic transparency on morphological processing.
6.
The experiments
9
6. 1 Methods
6.1.1. Participants
A total of 93 native English speakers between the age of 17 and 31 years who had not
learned a second language before age ten participated in this study. Of those, 19
participants took part in the computer audio and the individual version respectively;
19 pairs of participants took part in the interactive dictation version and 17 in the
visual version. They were all students from a variety of departments at the University
of Calgary and received either course credit or $15 for their participation.
6.1.2. Materials
The stimuli employed in this study were 40 English nominal compounds described in
Section 4 above. As shown in Table 1, the core set consisted of 10 TT compounds, 10
OT compounds, 10 TO compounds, and 10 OO compounds.
6.1.3. Procedure
As indicated above, this experiment was carried out in four different versions. In all
four versions of the experiment, the critical trials were preceded by four practice
trials, after which the participants had the opportunity to ask any remaining
questions. All core stimuli were presented in one block in randomized order. This
block began with three additional warm-up trials that were excluded from the
analysis. The main experiment took five to ten minutes per participant. It was carried
out on a MacBook Pro, and controlled by a script implemented in Psyscope B57.
Before the main experiment, participants filled out a questionnaire on their language
background and some demographic information. After the main experiment, they
were asked an online-survey in which they rated the degree of transparency of each
of the core stimuli. The online survey was implemented in Obsurvey.
Figure 1. A P3 study using progressive demasking and typing.
10
Interactive Dictation. In this version of the experiment, two participants interacted
with each other. The participants sat in two sound booths with windows facing each
other. Sound booth #1 was equipped with a computer screen and a microphone,
which was connected to a button box that timed the oral response. The other sound
booth contained a wireless keyboard and loudspeakers. A program called LineIn
enabled soft playthrough from the microphone in sound booth #1 to the speakers in
sound booth #2.
At the beginning of each trial, the participant in sound booth #1 saw a progressively
demasked visual stimulus appearing on the computer screen. Specifically, a masked
prime that was either the head or the modifier of the compound was presented for 16
ms, followed by a mask of 284 ms. The prime was presented twice, followed by the
complete compound stimulus. The length of stimulus presentation increased in steps
of 16 ms, the length of the mask decreased in steps of the same length. In each trial,
the compound was presented up to 16 times with a final duration of 288 ms.
As soon as the participant recognized the word, they responded by saying it out loud,
which stopped the presentation of the visual stimulus. The participant in sound booth
#2 typed the word dictated by the other participant and pressed the return-key to
initiate the next trial. The next stimulus was presented on the screen in sound booth
#1 1000 ms after the participant #2 had pressed the return-key.
Individual. The components of the individual version of the experiment were identical
to those of the interactive dictation study. However, a single participant performed
all tasks. Specifically, the participant performed the progressive demasking task with
naming and subsequently typed the stimulus. Participants in this version of the
experiment were equipped with a headset as opposed to a microphone to ensure that
their hands were free for typing.
Computer audio control. This version did not have a progressive demasking
component. Participants received auditory input produced by a computer voice and
typed the stimuli as soon as they recognized them without saying them out loud. The
auditory stimuli were produced by one of the voices available in the text-to-speech
application on Macintosh computers. We used the voice called "Alex", which, despite
being recognizable as a computer voice, sounds natural and is well comprehensible.
Occasional mispronunciations were corrected by altering the spelling of the stimulus.
Visual control. The visual control lacked the oral component. Participants responded
to the visual input by pressing a key as soon as they recognized the stimulus.
Subsequently, they typed the word.
6.2 Results
We discarded the data from two participants who took part in the visual version of
the experiment because they did not type anything. Moreover, we removed the
stimulus "staircase" from the analysis because it was significantly more frequent than
all other stimuli. The analysis of progressive demasking latencies is based on all
based on responses that were typed correctly.
6.2.1 Progressive demasking latencies
11
As there were no significant differences between the different experimental conditions,
we merged the results from the individual, the interactive and the visual version for the
analysis of progressive demasking latencies.
remaining responses,
whereas the analyses of inter-key-intervals and total typing
In Figure 2, the progressive demasking latencies of compound words with varying
times is only
based
on responses
that were
typed correctly.
degrees
of transparency
are shown.
Fully transparent
compounds (TT) were responded to fastest, followed by OT and TO compounds. Fully opaque compounds had the
6.2.1. Progressive
demasking
latencies
slowest progressive
demasking
latencies.
These data patterns were investigated statistically, using a linear mixed effects re(Baayen et al. 2008)
with random
interceptsthe
for participants
stimAs there gression
were model
no significant
differences
between
differentandexperimental
Response
latencies
were log-transformed
to reduce skew
potential distortion
conditions,uli.we
merged
the results
from the individual,
theand
interactive
and the visual
version forfrom
theoutliers.
analysis of progressive demasking latencies.
The first column in the fixed-effects Table 2 lists levels of independent variables.
t-values in the fourth column that are greater than +/– 2.0 can be considered to be
In Figure The
2 below,
the progressive demasking latencies of compound words with
significant at the p < .05 level. In the present analysis, negative values indicate facilitavarying degrees
of transparency are shown. Fully transparent compounds (TT) were
tion because they correspond to lower response latencies.
responded toInfastest,
followed
and TOwith
compounds.
Fullywere
opaque
compounds
this case,
both typesby
of OT
compounds
transparent heads
recognized
were had the
slowestfaster
progressive
demasking
latencies.
significantly
than OO compounds,
which
are on the intercept. Moreover, the
frequency of the stimuli significantly reduced response latencies.
2150
2100
2050
2000
1950
1900
1850
1800
1750
1700
1650
1600
TT
OT
TO
OO
Figure 2. Progressive
demasking
latencies
Figure 2. Progressive
demasking
latencies.
These data patterns were investigated statistically, using a linear mixed effects
regression model (Baayen et al. 2008) with random intercepts for participants and
stimuli. Response latencies were log-transformed to reduce skew and potential
distortion from outliers.
The first column in the fixed effects Table 2 below lists levels of independent
variables. The t-values in the fourth column that are greater than +/- 2.0 can be
considered to be significant at the p < .05 level. In the present analysis, negative values
indicate facilitation because they correspond to lower response latencies.
In this case, both types of compounds with transparent heads are recognized
significantly faster than OO compounds, which are on the intercept. Moreover, the
frequency of the stimuli significantly reduced response latencies.
Gary Libben and Silke Weber
12
Table 2. Effects of compound frequency and semantic transparency on progressive
demasking
Gary Libbenlatencies
and Silke Weber
Table 2. Effects of compound frequency and semantic transparency on progressive
Random effects:
demasking
latencies.
Table 2. Effects
of compound frequency and semantic transparency on progressive
demasking latencies
Groups
Name
Random effects:
participant
Groups
stimulus
Residual
participant
stimulus
Variance
Std.Dev.
(Intercept)
Name
(Intercept)
0.0512115
(Intercept)
0.0430524
Variance0.0092626 Std.Dev.
0.0430524 0.226300 0.207491
0.207491
0.096242
(Intercept)
0.0092626
0.096242
Number
of obs: 2103, groups: participant,
55; stimulus,
39
Residual
0.0512115
0.226300
Number
of obs: 2103, groups: participant, 55; stimulus, 39
Fixed
effects:
Fixed effects:
Estimate
Estimate
(Intercept)
(Intercept)
frequency_cp
frequency_cp
transparencyOT
transparencyOT
transparencyTO
transparencyTO
transparencyTT
transparencyTT
7.71
7.71
–0.03
–0.03
–0.12–0.12
–0.07–0.07
–0.15–0.15
Std. Error
Std. Error
0.047
0.01
0.046
0.047
0.049
0.047
0.01
0.046
0.047
0.049
t value
t value
165.24
–3.00
–2.65
–1.60
–3.12
165.24
–3.00
–2.65
–1.60
–3.12
200
150300
100250
Letter typing times
Letter typing times
6.2.2 Inter-key-intervals
at the morphological
boundary
6.2.2.
Inter-key-intervals
at the morphological
boundary
6.2.2
Inter-key-intervals
at the
morphological
boundary
The second
dependent variable
we analyzed
was the inter-key-intervals
at the boundarysecond
between
thedependent
first and the
second
constituent
of the
compounds.
Figure
3 illustrates at the boundThe
dependent
variable
we analyzed
was
thewere
inter-key-intervals
The
second
variable
we analyzed
the inter-key-intervals
at the
inter-key-intervals
at
four
different
positions
in
the
compound.
The
plus
1
position
ary
betweenbetween
the first and
constituent
of the compounds.
3 illustrates
boundary
the the
firstsecond
and the
second constituent
of the Figure
compounds.
Figure 3
refers to the constituent boundary, i.e. the interval between the last letter of the first
illustrates
inter-key-intervals
at
four
different
positions
in
the
compound.
The
plus 1
inter-key-intervals
at four different positions in the compound. The plus 1 position
constituent and the first letter of the second constituent. Accordingly, minus 1 refers to
position
refers
to the
thelast
constituent
boundary,
i.e. the
interval
the
refers
to the
constituent
boundary,
i.e. the
interval
between
thebetween
lastCruletter
oflast
the letter
first of
the interval
between
and the pre-final
letter
of the first
constituent,
etc.
the
first
and
first
ofconstituent.
the second
constituent.
Accordingly,
cially,
theconstituent
inter-key
at the
the of
morpheme
boundary
is longer Accordingly,
than
the inter-keyconstituent
and theinterval
first
letter
theletter
second
minus
1 refersminus
to
1intervals
refers to
the
interval
between
the
last
and
the
pre-final
letter
of
the
first
constituent,
at
all
surrounding
positions.
the interval between the last and the pre-final letter of the first constituent, etc. Cruetc. Crucially, the inter-key interval at the morpheme boundary is longer than the
cially, the inter-key300interval at the morpheme boundary is longer than the inter-keyinter-key-intervals at all surrounding positions.
intervals at all surrounding
positions.
250
50
200
0
150 Minus 2
Minus 1
Plus 1
Plus 2
Letter location (Plus 1 is the..
100
Figure 3. Inter-key intervals
50 at the constituent boundary (Plus 1) and surrounding letter
locations
0
Minus 1
Plus 1
Plus 2
Minus 2
Letter
location (Plus
1 is the.. (Plus 1) and surrounding
Figure 3. Inter-key intervals at the
constituent
boundary
letter locations.
Figure 3. Inter-key intervals at the constituent boundary (Plus 1) and surrounding letter
locations
Figure 4 further distinguishes inter-key-intervals by transparency type. It illustrates
that inter-key-intervals are reliably slowest at the morpheme boundary. However,
13
Semantic transparency and compounding the effect size appears to be dependent on degree and location of transparency. The
inter-key-intervals are longest for fully transparent compounds and shortest for fully
350
Semantic transparency
and compounding opaque compounds.
The other two testing conditions
fall in between.
300
250
350
200
300
150
250
100
200
50
150
0
100
TT
50
Minus two
Minus one
Plus one
Plus two
Minus two
Minus one
Plus one
Plus two
OT
TO
OO
Figure 4. Inter-key-intervals by letter location and transparency type
0
TT
OT
TO
OO
Figure 44.
further
inter-key-intervals
by and
transparency
type.
Figure
4.distinguishes
Inter-key-intervals
by letter
letter location
transparency
typeIt illustrates
Figure
Inter-key-intervals
by
location
and transparency
type.
that inter-key-intervals are reliably slowest at the morpheme boundary. However, the
effect
size appearslinear
to be dependenteffects
on degree
and locationanalysis
of transparency.
The interThe
following
regression
compares
inter-key-intervals
Figure 4 furthermixed
distinguishes inter-key-intervals
by transparency
type. It
illustrates
key-intervals
are
longest
for
fully
transparent
compounds
and
shortest
for
fully
opaque
at the constituent
boundary.
compounds
withHowever,
transparent
heads
that inter-key-intervals
areAgain,
reliably responses
slowest at theto
morpheme
boundary.
the
compounds. The other two testing conditions fall in between.
differ significantly
fromtothose
to fully
compounds.
Note, however,
effect size appears
be dependent
onopaque
degree and
location of transparency.
The inter-that interThe following linear mixed effects regression analysis compares inter-key-interkey-intervals
longestfor
for fully
transparentcompounds,
compounds andwhereas
shortest forPDM
fully opaque
key-intervals
wereare
longer
transparent
latencies were
vals at the constituent boundary. Again, responses to compounds with transparent
compounds. The other two testing conditions fall in between.
shorter.
heads differ significantly from those to fully opaque compounds (OO compounds are
The following linear mixed effects regression analysis compares inter-key-interon the intercept). Note, however, that inter-key-intervals were longer for transparent
vals
at theof
constituent
boundary. Again, and
responses
to compounds
with transparent
Table
3. Effects
compound
semantic
transparency
on inter-keycompounds,
whereas
PDM
latencies frequency
were shorter.
heads differ significantly from those to fully opaque compounds (OO compounds are
intervals.
on the intercept). Note, however, that inter-key-intervals were longer for transparent
Table 3. Effects of compound frequency and semantic transparency on inter-key-intervals
compounds, whereas PDM latencies were shorter.
Random Effects:
Groups
Table 3. Effects of compound frequency and semantic transparency on inter-key-intervals
Name
Random Effects: (Intercept)
participant
stimulus Groups
(Intercept) Name
Residual
0.149634
participant
(Intercept)
Number ofstimulus
obs: 2341, groups: participant, 73;(Intercept)
stimulus, 39
Residual
0.149634
Fixed effects:
Variance
Std.Dev.
0.075219
0.023330 Variance
0.38683
0.075219
0.023330
0.38683
0.27426
0.15274 Std.Dev.
Number of obs: 2341, groups:
participant, 73; stimulus,
39
Estimate
Std. Error
0.27426
0.15274
t value
(Intercept)Fixed effects:
5.55911
0.06761
82.22
frequency_cp
–0.04890 Estimate
0.01652 Std. Error
–2.96 t value
transparencyOT
0.17164
0.07262
2.36
(Intercept)
5.55911
0.06761
82.22
transparencyTO
0.11149
0.07414
1.50
frequency_cp
–0.04890
0.01652
–2.96
transparencyTT
0.20521
0.07742
2.65
transparencyOT
0.17164
0.07262
2.36
transparencyTO
0.11149
0.07414
1.50
Figure 5 below indicates that all versions of the experiment yielded
transparencyTT
0.20521
0.07742
2.65
the same basic
pattern. Specifically, fully transparent compounds resulted in the longest inter-keyintervals at the constituent boundary and fully opaque compounds involved the
shortest inter-key-intervals. Interestingly, the interactive version yielded the same
pattern as that of the individual versions. This represents a confirmation that the
14
progressive demasking technique that is effectively ‘split up’ in the P3 paradigm,
nevertheless targets comparable phenomena.
Gary Libben and Silke Weber
350
300
250
TT
OT
TO
00
200
150
100
50
0
Interactive
Individual
Computer
Visual
voice
(just return key)
5. Inter-key-intervalsat
at the
the constituent
boundary
by experiment
version
Figure 5.Figure
Inter-key-intervals
constituent
boundary
by experiment
version and
and
transparency
type
transparency type.
Figure 5 above indicates that all versions of the experiment yielded the same basic
7. Discussion
pattern. Specifically, fully transparent compounds resulted in the longest inter-key-
intervals at the constituent boundary and fully opaque compounds involved the shortInterestingly,
the interactive
version yieldedof
the
pattern as on lexical
Our goalest
ininter-key-intervals.
this paper has been
to explore
the consequences
a same
perspective
that
of
the
individual
versions.
This
represents
a
confirmation
that
the
progressive
processing that emphasizes the extent to which language structures arede-essentially
maskingintechnique
is effectively
‘split of
up’morphological
in the P3 paradigm,structure
nevertheless
psychological
nature.that
Our
discussion
intargets
compounding
comparable phenomena.
focused on the role that language processing plays in shaping the morphological
structure of compounds and, in particular, in giving rise to positionally bound
morphemic constituents that may vary in the extent to which they differ from the
7. Discussion
meanings of their free standing word counterparts.
Our goal in this paper has been to explore the consequences of a perspective on lexical
We presented
anthat
experimental
technique
is consistent
with a
processing
emphasizes thepsycholinguistic
extent to which language
structuresthat
are essentially
psypsychocentric
perspective
because
it
combines
aspects
of
lexical
recognition,
lexical
chological in nature. Our discussion of morphological structure in compounding
naming, focused
and writing.
We
described
the means
which the
thismorphological
technique struccan be used to
on the role
that
language processing
playsby
in shaping
compounds and,
in particular,
in giving
rise to positionally
bound morphemic
explore ture
theofinterplay
among
response
variables
and claimed
that a promising
constituents
that technique
may vary in the
extentittocan
which
differ
the meanings
of
characteristic
of this
is that
bethey
used
to from
compare
individual
lexical
their
free
standing
word
counterparts.
processing with processing that involves both a language producer and a language
receiver. We presented an experimental psycholinguistic technique that is consistent with a
psychocentric perspective because it combines aspects of lexical recognition, lexical
naming, and writing. We described the means by which this technique can be used to
The dataexplore
thatthewe
report suggest that the new P3 technique, which combines,
interplay among response variables and claimed that a promising characprogressive
naming
latency,
wordindividual
typing,lexical
yields
data that are
teristicdemasking,
of this technique
is that it can
be used and
to compare
processing
comparable,
and
perhaps
even
clearer,
than
those
obtained
in
lexical
decision
studies
with processing that involves both a language producer and a language receiver.
with the same
stimuli.
datawhich
showed
a clear
The data
that we The
reportprogressive
suggest that thedemasking
new P3 technique,
combines
pro- effect of
semanticgressive
transparency
such
that
compounds
demasking, naming
latency,
and word
typing, yields with
data thatsemantically
are comparable, opaque
and perhaps
even clearer,
those obtained
in lexical
decision
with
the same
morphological
heads
were than
recognized
more
slowly
thanstudies
those
with
semantically
stimuli.
The progressive heads.
demasking
data showed
a clear
of semanticdemasking
transpar- results
transparent
morphological
Taken
together,
theeffect
progressive
such that
compounds
with semantically
opaque
morphological
headsmorphological
were
point toency
a graded
view
of semantic
opacity that
has
an important
component. Compounds with two transparent constituents were recognized most
quickly. Slower recognition times were recorded for OT, TO and OO compounds, in
15
that order. This is exactly the order that had been found in lexical decision studies
using the same stimulus set.
The analysis of typing times yielded the same gradation associated with opacity. It is
interesting to note that the locus of this opacity effect was the typing speed at the
character immediately following the morpheme boundary. This points to the ability
of the technique to target morphological effects. It also suggests that the effects of
semantic opacity emerge during the production process. From a psychocentric
perspective, they are indeed inextricably tied to the production process. In our view,
a promising line of future research would be the investigation of how and why
transparency effects occur at the morpheme boundary. This would allow us to
investigate more fully a hypothesis that dates back to Sandra (1990), who proposed
that opaque compounds are not morphologically decomposed in the way that
semantically transparent ones are. An alternative possibility is that morphosemantic
computation naturally occurs are morphological breakpoints, and the typing task,
with its ability to generate letter-by-letter latencies, is sensitive to this.
In our view, the fact that the typing technique is sensitive to the semantic
transparency effects obtained in lexical decision and in the PDM technique brings to
the foreground an important consideration: In the typing part of the experiment,
participants are not responding to a stimulus, but rather are producing one. So, any
account of the transparency effect that appeals to a ‘surprise’ element would have
difficulty in accounting for the data pattern. Put another way, if there are semantic
transparency stimulus effects at play in the typing task, those stimulus effects are
being generated from within the participants.
Finally, we note an intriguing result of the use of the P3 technique in both individual
and interactive contexts. Our data indicated that it was the interactive version of the
experiment that yielded response patterns that were most clearly differentiated in
terms of semantic transparency. We see this as opening up the possibility that
differences in the production of compounds varying in semantic transparency by the
speaker in the dyad results in typing differences in the production of compounds by
the second member of the dyad. The mechanisms that would underlie this sort of
cueing are not yet clear.
To sum up, we suggest that a psychocentric perspective on compound processing and
on the nature of semantic transparency opens up a number of conceptual and
experimental opportunities. The P3 technique that we have presented was designed
to use those opportunities and to combine measures in a manner that does not force
us into an inappropriately simplified view of language processing, but rather allows
for a fuller investigation of the psychodynamics of lexical processing. This approach
is linked to the Derwing (1973) claim that both linguistic units and linguistic
structures have their empirical realization as part of the language process of speech
production and comprehension.
16
References
Aristotle. “Poetics” (E.M. Edgehill, Translation), retrieved from
http://classics.mit.edu//Aristotle/poetics.html December 5, 2009.
Baayen, R. H., Davidson, D.J., & Bates, D. (2008). Mixed-effects modeling with crossed
random effects for subjects and items. Journal of Memory and Language, 59, 390412.
Derwing, B. L. (1973). Transformational grammar as a theory of language
acquisition: A study in the empirical, conceptual and methodological foundations
of contemporary linguistics. London & New York: Cambridge University Press.
Fiorentino, R. and Fund-Reznicek, E. (2009). Masked morphological priming of
compound constituents. The Mental Lexicon, 4(2), 159-193.
Frisson, S., Niswander-Klement, E., & Pollatsek, A. (2008). The role of semantic
transparency in the processing of English compound words. British Journal of
Psychology, 99, 87-107
Gagné, C. L., & Spalding, T. L. (2009). Constituent integration during the processing
of compound words: Does it involve the use of relational structures? Journal of
Memory and Language, 60, 20-35.
Grainger, J., & Segui, J. (1990). Neighborhood frequency effects in visual word
recognition: A comparison of lexical decision and masked identification
latencies. Perception & Psychophysics, 47, 191-198.
Jarema, G., Busson, C., Nikolova, R., Tsapkini, K. & Libben, G. (1999). Processing
compounds: A cross-linguistic study. Brain and Language, 68, 362-369.
Ji, H., Gagné, C. L., & Spalding, T. L. (2011) Benefits and costs of lexical
decomposition and semantic integration during the processing of transparent
and opaque English compounds. Journal of Memory and Language, 65, 406-430
Libben, G. (forthcoming). Compound fracture and morphological superstates.
Libben, G. (1998). Semantic transparency in the processing of compounds:
consequences for representation, processing, and impairment. Brain and
Language, 61, 30-44.
Libben, G. (2006). Why study compounds? An overview of the issues. In G. Libben &
G. Jarema, (Eds.), The representation and processing of compound words. Oxford:
Oxford University Press (pp. 1-21).
Libben, G. (2010). Compounds, semantic transparency, and morphological
transcendence. In S. Olson (ed.), New impulses in word-formation (Linguistische
Berichte Sonderheft 17), Buske, Hamburg, 212-232.
Libben, G., Gibson, M., Yoon, Y-B. & Sandra D. (2003). Compound fracture: The role
of semantic transparency and morphological headedness. Brain and Language,
84, 26-43.
17
Libben, G., Weber, S., & Miwa, K. (2012). P3: A technique for the study of perception,
production, and participant properties. The Mental Lexicon, 7:2, 237-248.
Menn, L. (1996). The real-time accessibility of grammatical morphemes: Miceeaters, alumni relations, and The El Alhambra Motel. Keynote Address to the
International Association for Applied Linguistics, Jyväskylä, Finland/
Monahan, P., Fiorentino, R., and Poeppel, D. (2008). Masked repetition priming using
magnetoencephalography. Brain and Language, 106, 65-71.
Olofsson, A. (1990). On the origins of chopper 'helicopter' and moped, American
Speech, 65:3, 278-284.
Sahel, S., Nottbusch, G, Grimm, A. & Weingarten, R. (2008). Written production of
German compounds: effects of lexical frequency and semantic transparency.
Written Language & Literacy 11(2), 211–227.
Sandra, D. (1990). On the representation and processing of compound words:
automatic access to constituent morphemes does not occur. Quarterly Journal of
Experimental Psychology, 42, 529-567.
Singh, R. (2006): “Whole Word Morphology”. In: Keith Brown, (Editor-in-Chief)
Encyclopedia of Language & Linguistics, Second Edition, Volume 13. Oxford:
Elsevier, 578–579.
Taft M. & Forster, K.I. (1976). Lexical storage and retrieval of polymorphemic and
polysyllabic words. Journal of Verbal Learning and Verbal Behavior, 15, 607-620.
Will, U., Nottbusch, G., & Weingarten, R. (2006). Linguistic units in word typing:
Effects of word presentation modes and typing delay. Written Language &
Literacy, 9 (1), 153-176.
View publication stats
Download