This article was downloaded by: [Archives & Bibliothèques de l'ULB] On: 28 December 2014, At: 11:30 Publisher: Routledge Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
Cognitive Neuropsychology Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/pcgn20
The representation of compound headedness in the mental lexicon: A picture naming study in aphasia a
b
b
Marco Marelli , Giusy Zonca , Antonella Contardi & Claudio Luzzatti a
c
Center for Mind/Brain Sciences, University of Trento, Rovereto, Italy
b
S. Maugeri Foundation IRCCS, Montescano Medical Center, Montescano, Italy c
Department of Psychology, University of Milano-Bicocca, Milano, Italy Published online: 07 Dec 2013.
Click for updates To cite this article: Marco Marelli, Giusy Zonca, Antonella Contardi & Claudio Luzzatti (2014) The representation of compound headedness in the mental lexicon: A picture naming study in aphasia, Cognitive Neuropsychology, 31:1-2, 26-39, DOI: 10.1080/02643294.2013.860024 To link to this article: http://dx.doi.org/10.1080/02643294.2013.860024
PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions
Cognitive Neuropsychology, 2014 Vol. 31, Nos. 1 –2, 26 –39, http://dx.doi.org/10.1080/02643294.2013.860024
The representation of compound headedness in the mental lexicon: A picture naming study in aphasia Marco Marelli1, Giusy Zonca2, Antonella Contardi2, and Claudio Luzzatti3 1
Center for Mind/Brain Sciences, University of Trento, Rovereto, Italy S. Maugeri Foundation IRCCS, Montescano Medical Center, Montescano, Italy 3 Department of Psychology, University of Milano-Bicocca, Milano, Italy
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
2
Most compound words are constituted of a head constituent (e.g., light in moonlight) and a modifier constituent (e.g., moon in moonlight); the information transmitted by these head-modifier roles is fundamental for defining the grammatical and semantic properties of the compound and for identifying a correct combination of the constituents at the conceptual level. The objective of this study is to assess how lexical processing in aphasia is influenced by the head-modifier structure of nominal compounds. A picture-naming task of 35 compounds with head-initial (pescespada, swordfish, literally fishsword) and head-final (autostrada, highway, literally carroad) forms was administered to 45 Italian aphasic patients, and their accuracy in retrieving constituents was analysed with a mixed-effects logistic regression. The interaction between headedness and constituent position was significant: The modifier emerged as being more difficult to retrieve than the head, but only for head-final compounds. The results are consistent with previous data from priming experiments on healthy subjects and provide convincing evidence that compound headedness is represented at central processing levels. Keywords: Compound words; Picture naming; Compound headedness; Aphasia.
Research on the representation of compound words has a long tradition in psycholinguistic literature: Since the seminal paper by Taft and Forster (1976), many studies, most employing chronometric methods (i.e., measuring behavioural response times or fixation durations in reading), have dealt with the question as to whether morpheme representations are involved in the processing of complex words. The issue was also
addressed qualitatively and quantitatively within the classical cognitive neuropsychological frame, by investigating the errors made by patients with acquired language disorders (for a review see Goldman, 2007; Semenza & Mondini, 2006). These analyses have largely confirmed the results obtained by healthy participants and have been profitably used to test the predictions of psycholinguistic models.
Correspondence should be addressed to Marco Marelli, CIMeC– Palazzo Fedrigotti, Corso Bettini 31, Room c111, 38068 Rovereto (TN), Italy (E-mail: [email protected]) Preliminary results of this study were presented at the 48th Annual Meeting of the Academy of Aphasia (Athens, 2010) and the 29th European Workshop on Cognitive Neuropsychology (Bressanone, 2011). The study was supported by a grant awarded to C. L. by the University of Milano-Bicocca (Fondo d’Ateneo per la Ricerca (FAR) 2009).
26
# 2013 Taylor & Francis
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
COMPOUND HEADEDNESS IN PICTURE NAMING
The first phenomenon that has been verified is that compounds are not (or not only) stored in the mental lexicon as unique and directly accessed representations. Results suggesting whole-word access have been reported (Janssen, Bi, & Caramazza, 2008), but most data clearly indicate that constituent representations are accessed when processing a compound word, as extensively demonstrated in psycholinguistics by studies adopting priming paradigms (e.g., Jarema, Busson, Nikolova, Tsapkini, & Libben, 1999; Libben, Gibson, Yoon, & Sandra, 2003; Zwitserlood, 1994) and/or reporting constituent frequency effects (e.g., Bien, Levelt, & Baayen, 2005; Dunabeitia, Perea, & Carreiras, 2007; Hyo¨na¨ & Pollatsek, 1998). This same hypothesis is supported by neuropsychological data, indicating that a similar process is likely to occur in compound production. Indeed, constituent access is strongly supported by the phenomenon described by Semenza, Luzzatti, and Carabelli (1997) and Mondini, Luzzatti, Zonca, Pistarini, and Semenza (2004): When asked to name verb – noun compounds (e.g., aspirapolvere, vacuumcleaner, literally inhaledust), patients whose performance was characterized by a grammatical class effect (with a better performance on nouns than verbs), made significantly more errors on the verb (aspira) constituent than on the noun constituent (polvere); since compounds were globally nouns, the hypothesis of a direct access is not consistent with these results. However, compound processing cannot rely only on constituents, and properties associated with the whole compound are also likely to be retrieved. Not only is this hypothesis supported by theoretical reasons (compound properties cannot be totally deduced from constituent information; Badecker, 2001; Marelli, Aggujaro, Molteni, & Luzzatti, 2012), it also has empirical backing: In the cases reported by Semenza et al. (1997; see also Semenza et al., 2011), errors were produced that maintained the original morphological structure of the stimuli, while Badecker (2001) described a patient who, when naming pictures representing compound words, was aware that his responses were incomplete if a constituent
was missing. Moreover, Marelli et al. (2012) have shown that favouring the access to structure properties helped G.R., a patient suffering from deep dyslexia with morphological impairment, in compound retrieval (see also Mondini, Jarema, Luzzatti, Burani, & Semenza, 2002). These data suggest that morphological structure is stored separately from the lexical representations of the constituents and is used during processing to form the compound word. Marelli et al. (2012) proposed that this structure is represented as an independent node in the lemma, which is the amodal and abstract processing level where grammatical and lexical properties of words would be stored (Levelt, Roelofs, & Meyer, 1999). This abstract compound node would be linked to separate and modality-specific representations of constituents at more peripheral (i.e., orthographic and phonological) levels. However, there is still very little in the way of data available as to how this information is used during processing. Are constituents hierarchically organized? In other words, does one constituent have a privileged role over the other? From a theoretical point of view, in most compounds it is possible to identify a head and a modifier, the head being the constituent specifying lexical (grammatical gender—in languages such as Italian where nouns have a masculine or feminine gender, irrespective of the natural gender of the underlying concept; grammatical class) and semantic (category, principal traits) properties of the whole compound. In German languages the roles are positionally coded, with the head always in the rightmost position (Williams, 1981). This is such a regular phenomenon that Williams (1981) even proposed a right-hand head rule, indicating that the principal morpheme of a complex word is always to be found in the final position. However, several languages (such as Romance languages) also have head-initial compounds (e.g., pescespada, swordfish, literally fishsword), or have only head-initial compounds, as in Hebrew (e.g., tapuach adamah, potato, literally apple earth). Even if the latter form is less frequent than the head-final form (this holds true also in Italian and other Romance languages, see Cognitive Neuropsychology, 2014, 31 (1 –2)
27
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
MARELLI ET AL.
Guevara & Scalise, 2008; Marelli & Luzzatti, 2012), the fact that it exists at all suggests that head-initial compounding is also a possible way to create new words. In this scenario, it is important for psycholinguistic and neurolinguistic studies to investigate whether head-modifier roles are represented in the mental lexicon, and whether head-initial and head-final structures are differentially processed. Given their particular properties, Romance languages constitute an ideal benchmark when addressing this issue. Recent results have indeed indicated that head-modifier roles have an important effect on compound processing. Patients affected by neglect dyslexia produce fewer neglect errors when reading head-initial than when reading head-final compounds (Marelli, Aggujaro, Molteni, & Luzzatti, 2013; Semenza et al., 2011). Moreover, errors made by bilingual English – French speaking subjects with aphasia in a series of tasks involve the modifier more often than the head (Jarema, Perlak, & Semenza, 2009). These data indicate that the head is relatively more salient and thus easier to process than the modifier. However, these studies did not fully compare the two possible compound structures: In the former, the analysis was limited to the performance on the first constituent due to the peculiarities of the neglect disorder, and the latter case focused on crosslinguistic comparison. As a result, the question as to whether head-initial and head-final compounds are differentially processed remains open. In earlier studies (Marelli, Crepaldi, & Luzzatti, 2009; Marelli & Luzzatti, 2012), we investigated the representation of the head-modifier structure in Italian by evaluating the processing of written noun– noun nominal compounds in healthy subjects. The results indicated that head-initial and head-final structures do indeed have different representations in the mental lexicon. In a constituent priming experiment (Marelli, Crepaldi, & Luzzatti, 2009), the participants recognized a compound more quickly when it was preceded by its head than when the prime was the modifier constituent. However, the effect was significant only in head-final compounds,
28
Cognitive Neuropsychology, 2014, 31 (1 –2)
whereas in head-initial compounds the magnitude of the priming effects did not differ between head and modifier, suggesting that only head-final compounds are built on a hierarchical structure, whereas head-initial compounds are more similar to lexicalized syntactic units. This hypothesis was further expanded upon in an investigation of frequency effects when reading Italian compounds (Marelli & Luzzatti, 2012). Constituent frequency effects facilitate compound recognition when embedded in a head-final structure (higher frequency leads to shorter response times), but are inhibitory when the compound is head-initial (higher frequency leads to longer response times). This indicates that when Italian speakers process a compound noun in their language, they take the head-final structure as the default compound representation. Since constituent frequency was found to interact with compound transparency, this effect was interpreted in semantic terms: When trying to build the combined concept on the basis of constituent meanings, the system assumes a head-final representation. When this assumption is not satisfied, the headmodifier roles must be reassigned in a time-consuming processing. This model assumes that compound processing is ultimately semantically driven, implying a conceptual combination procedure (Gagne´ & Spalding, 2007): Lexical information acquired during early processing phases interacts with stored distributional knowledge in order to derive the combined concept associated with a compound word. Since this distributional information is thought to play a role in semantic processing (Gagne´ & Spalding, 2009; Lynott & Ramscar, 2001; Marelli & Luzzatti 2012), it would be represented at central processing stages, which are involved in many different tasks, including those tapping into output processing levels. In other words, the hierarchical effect of the headfinal structure (as described above) would not only be found in silent-reading experiments, but also in tasks requiring the production of compound nouns. If this hypothesis is correct, converging evidence should emerge from the analysis of error patterns in patients with aphasia. The
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
COMPOUND HEADEDNESS IN PICTURE NAMING
present study tests this prediction in a picturenaming experiment carried out with a large sample of Italian aphasic patients. The study was organized in two main phases: The first phase consisted of a picture-naming task to identify patients with greater problems in naming morphologically complex words than in naming simple words; in the second phase, the performance of these patients was further analysed to assess the extent to which their naming disorder affected the first versus the second constituent of compound nouns and to investigate whether this error pattern was influenced by the head-modifier structure. Three possible outcomes were envisaged: 1. The head-modifier structure could impact the naming performance in a similar manner to that observed in written word processing in healthy individuals (Marelli et al., 2009): The most prominent role in head-final compounds would be played by heads rather than modifiers; conversely, when producing head-initial compounds, head and modifier constituents would be equally easy to retrieve. 2. The results could indicate that the head is easier to retrieve irrespective of the headmodifier structure involved, disconfirming that head-initial and head-final compounds undergo different processing. 3. No difference could be found between head and modifier performances, suggesting that the head-modifier structure is not represented in the mental lexicon. Outcomes 2 and 3 would indicate that data by Marelli et al. (2009) may depend on the task employed in their reading experiments and that the hierarchical effect in head-final compounds could be specific for written word processing. On the contrary, Outcome 1 would suggest that the hypothesis by Marelli and Luzzatti (2012) is correct, indicating that the head-modifier structure is stored at central levels of processing, and that this very structure has different representations on the basis of the head position.
PRELIMINARY ANALYSIS Method Participants A total of 122 patients with focal left-hemisphere (LH) brain lesions were examined. All participants were suffering from mild-to-moderate aphasic impairments, as evaluated by the Italian version of the Aachen Aphasia Test (Luzzatti, Willmes, & De Bleser, 1996), or from residual aphasia—that is, the patients’ psychometric assessment did not indicate an aphasic disorder but signs of aphasia were present either in their spontaneous speech or in their performance on clinical tasks. They were recruited at the “Fondazione Salvatore Maugeri” Rehabilitation Centre, Montescano. The participants had previously been diagnosed as suffering from Broca’s aphasia (12), anomic aphasia (24), conduction aphasia (8), Wernicke’s aphasia (5), residual aphasia (64), or other aphasic forms (10). Materials A confrontation naming battery was employed, consisting of 145 stimuli, of which 110 items were presented as pictures, while 35 items were presented as definitions. The battery was an adapted version of the task employed by Mondini et al. (2004) in an earlier study. Fifty-seven stimuli were objects denoted by a simple noun, while 88 stimuli were objects denoted by a complex noun. Complex targets were either derived nouns (36 stimuli; e.g., campanile, bell tower, derived from campana, bell) or compound nouns (52 stimuli; e.g., pescespada, swordfish, literally fishsword). Spoken frequency measures were collected from the Lessico di frequenza dell’Italiano Parlato (LIP) corpus for each stimulus (De Mauro, Mancini, Vedovelli, & Voghera, 1993). A rating study on 12 unimpaired participants was also run to assess age of acquisition and imageability of the battery stimuli. Participants were asked to rate each word on a 7-point scale. Procedure Stimuli were presented in a unique list, comprising both simple and complex nouns in a randomized Cognitive Neuropsychology, 2014, 31 (1 –2)
29
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
MARELLI ET AL.
order. Definitions were presented first, after which the experimenter showed the pictures. The patients were asked to produce the noun corresponding to each definition/picture. Responses were considered correct only if given within a 2-s interval from stimulus presentation. Minor phonological alterations and articulatory distortions were ignored.
EXPERIMENTAL ANALYSIS
Data analysis The patients’ performances were individually assessed in order to reveal dissociations between complex and simple word naming. A logistic regression analysis was carried out on the data generated by each participant to predict the likelihood of a correct response on the basis of the stimulus category (complex vs. simple noun). Stimulus length (in syllables), imageability, age of acquisition, and spoken frequency were also introduced as covariates to rule out their influence on the patients’ performance. A backward procedure was adopted to simplify the statistical models: Starting from a full-factorial model, predictors were removed one by one when their absence did not significantly affect the overall goodness of fit. If after this procedure the effect of morphological complexity was significant (p , .10), the participant was included in the subsequent phase (Experimental Analysis, see below).
Materials Two sets of compound nouns were extracted from the naming task analysed in the previous phase. The first set was composed of 16 compounds in which the first constituent was the morphological head of the compound noun (head-initial compounds; e.g., pescespada, swordfish, literally fishsword); the second set included 14 compounds in which the second constituent was the morphological head of the compound (head-final compounds; e.g., autostrada, highway, literally carroad). All the stimuli are reported in the Appendix, together with their English translations. The sets were matched for spoken frequency of both the constituents, imageability, age of acquisition, frequency, stimulus length of the compound noun, naming agreement (in a sample of 38 control subjects), and proportion of stimulus types (pictures vs. definitions). Frequency measures were extracted from the LIP database (De Mauro et al., 1993). We also considered semantic transparency, which is widely accepted as affecting compound retrieval and headedness-associated effects (e.g., Libben et al., 2003; Marelli & Luzzatti, 2012). Semantic transparency was acquired by means of a rating study on 12 unimpaired participants, who were asked to rate the degree to which each compound meaning could be derived from the meaning of its constituents on a 4-point scale. The matching for the described psycholinguistic variables is summarized in Table 1.
Results Of the 122 patients assessed, 45 showed disproportionate impairment in producing morphologically complex words in comparison to simple words. This dissociation cannot be explained in terms of stimulus length, imageability, frequency, or age of acquisition, which have independent effects (if any) on the patients’ performance. The patients who participated in the study had previously been diagnosed as suffering from anomic aphasia (10), residual aphasia (28), Broca’s aphasia (3), conduction aphasia (2), and transcortical motor aphasia (2).
30
Cognitive Neuropsychology, 2014, 31 (1 –2)
Method Participants In this phase, the performance of the 45 patients who showed disproportionate impairment of morphologically complex words was subjected to further analysis.
Procedure The performance on the compound stimuli was reanalysed, focusing on accuracy in retrieving the individual constituents—that is, for each stimulus,
COMPOUND HEADEDNESS IN PICTURE NAMING
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
Table 1. Variables matching in the compound sets Variable
Range
Head-initial compounds
Head-final compounds
t-value
p
Compound frequency 1st constituent frequency 2nd constituent frequency Stimulus length (syllables) Imageability Semantic transparency Age of acquisition Naming agreement
0– 3.21 0– 5.01 0– 5.96 3– 8 1– 7 1– 4 1– 7 0– 1
0.57 3.12 2.52 5.06 4.94 2.84 3.89 0.69
0.96 2.25 2.79 4.29 5.18 2.82 3.73 0.71
1.27 1.44 0.43 1.95 0.69 0.08 0.59 0.18
.2174 .1627 .6711 .0631 .4911 .9335 .5632 .8596
Note: Word frequency values, extracted from the LIP corpus (490,000 word forms; De Mauro et al., 1993), were log-transformed.
we considered whether the constituents were correctly reported, irrespective of the accuracy in retrieving the whole compound. As a consequence, two distinct dependent measures were associated to each stimulus—that is, accuracy in reporting the first constituent, and accuracy in reporting the second constituent. Minor phonological alterations and articulatory distortions were ignored as before. Data analysis The accuracy in retrieving either constituent1 was analysed employing a logistic regression model. The peculiar properties of the dependent variable (being measured twice for each stimulus) and the intention of running a group study would inevitably lead to issues related to the non-independency of observations in a traditional generalized linear model framework; hence a mixed-effects analysis (Jaeger, 2008) was employed. If a nested structure including the random effects of participants, compounds, and constituents is specified, it is possible to prevent potential problems in parameter estimation related to the fact that observations are not independent. Stimulus type (definition vs. figure) was also included as an additional random effect. This statistical technique also guarantees
that significant effects are estimated net of the differences that characterize the individual patients: The results would thus characterize the whole group and not just a minor sample of outlying participants. The main effect of interest was the interaction between the position of the constituent considered (first vs. second Constituent Position) and Headedness of the compound (head-initial vs. head-final compounds). We also tested how this interaction was modulated by aphasia severity (as measured by the Token Test score; Huber, Poeck, Weniger, & Willmes, 1983; Luzzatti et al., 1996). The psycholinguistic variables described above were also introduced as covariates. All predictors were mean-centred, in order to ensure better estimates of the parameters in the subsequent analyses (Kraemer & Blasey, 2004). A backward procedure was adopted to simplify the statistical model: Starting from a full-factorial model, predictors were removed one by one when their absence did not significantly affect the overall model fit (on the basis of a goodness-of-fit chisquare test). Parameters were not removed when the goodness-of-fit test was significant, or if they were part of higher order interactions. Once the model was fitted, atypical outliers were identified
1 It may be argued that constituent accuracy could also be tested as a predictor in a regression analysis on compound naming accuracy, to assess to what extent the probability of retrieving the constituents affects the probability of retrieving the whole word. This effect could be modulated by constituent position and/or compound headedness, thus providing information on the representation of head-modifier roles in the mental lexicon. This analysis was run in the present study, but the interactions of interest between constituent accuracies, constituent position, and headedness were not significant—that is, constituent accuracy was helpful for compound retrieval irrespective of position and head-modifier roles.
Cognitive Neuropsychology, 2014, 31 (1 –2)
31
MARELLI ET AL.
and removed (employing 2.5 standard deviations of the residual errors as criterion). The model was then refitted to ensure that results were not driven by these outliers (Baayen, 2008).
patient’s disorder, the larger the effect of compound headedness was: Constituents of head-initial compounds are easier to retrieve, but this effect emerges only in participants who are relatively less impaired (see Figure 1). The main effects of age of acquisition, compound frequency, and semantic transparency were also significant. Constituents are easier to retrieve when they belong to compounds acquired earlier, more transparent, and with higher word frequency.
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
Results Constituent retrieval The final model included the interaction between headedness and constituent position and the interaction between headedness and aphasia severity, as well as age of acquisition, semantic transparency, and compound frequency as significant predictors. The parameters of the model are reported in Table 2 together with a summary of the simplification procedure. We were mainly interested in the interaction between constituent position and compound headedness, which emerged as significant. Patients showed greater impairment in retrieving the modifier than in retrieving the head constituent, but this effect emerged only when naming head-final compounds (accuracy: 44% vs. 50%) and not when naming head-initial compounds (accuracy: 55% vs. 55%). An interaction between headedness and aphasia severity was also found. The less severe the
Analysis by error types The patients’ responses were further classified according to error type. Errors could affect either the individual constituents or the whole compound. Three categories of errors were considered for the individual constituents: constituent omissions, lexical substitutions, and phonemic errors. Constituent omissions (e.g., “pesce”, fish, for pescespada, swordfish) also included latencies and word repairs for a specific constituent (e.g., “spada, ah no, pescespada”, sword, uh no, swordfish, for pescespada). Lexical substitutions included all cases in which a constituent was substituted with an existing related word (e.g., “pescecane”, shark, literally
Table 2. Results of the mixed-effects analysis on constituent retrieval x2
p
Removal order
Estimate
z-value
p
— — — — 10.6451 5.9668 4.9395 4.7972 4.9772 0.9759 0.5734 0.3307 0.1451 0.1167 0.0287
— — — — .0011 .0149 .0263 .0285 .0257 .3232 .4489 .5653 .7033 .7326 .8654
Not removed Not removed Not removed Not removed Not removed Not removed Not removed Not removed Not removed 6 5 4 3 2 1
0.035 0.146 20.646 20.069 0.014 0.699 0.485 20.486 0.211
0.15 1.56 3.81 6.41 3.31 3.09 2.88 2.35 2.26
.8753 .1191 .0001 .0001 .0009 .0019 .0039 .0189 .0241
Parameter Intercept Constituent position (CP) Headedness Aphasia severity Headedness × Aphasia Severity Semantic transparency Compound frequency Age of acquisition CP × Headedness Stimulus length 1st constituent frequency 2nd constituent frequency Aphasia Severity × CP Aphasia Severity × CP × Headedness Imageability
Note: The central part of the table summarizes the model-simplification procedure, including the goodness-of-fit tests and their results; constituent position, headedness, and aphasia severity were not removed because part of higher order interactions. The rightmost part of the table reports the effects of the included variables, after having removed outliers.
32
Cognitive Neuropsychology, 2014, 31 (1 –2)
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
COMPOUND HEADEDNESS IN PICTURE NAMING
Figure 1. Interaction between the effects of compound headedness and aphasia severity (as measured by the performance on the Token Test).
fishdog, for pescespada), thus producing a different compound noun. Phonemic errors included omissions, transpositions, substitutions, staccato utterances (e.g., “au-to-stra-da” for autostrada, highway). Errors affecting the whole compound were categorized as whole-word omissions, whole-word substitutions, circumlocutions, and neologistic compounds. Whole-word omissions also included latencies (more than 2 s) and repairs. Whole-word substitutions included all cases in which the patient produced an existing simple word (e.g., “un quadro”, a painting, for capolavoro, masterpiece). Circumlocutions included those cases in which the patient described the meaning of the target word by means of a sentence (“serve per portare sopra la montagna i passeggeri”, it is used to take passengers over the mountain, for funivia, cablecar). Compound neologisms were responses in the form of a novel compound, whose meaning is usually related to the target (“tagliaprezzemolo”, parsley cutter, literally cuts-parsley, for mezzaluna, dicing knife).
Table 3 reports the number of errors for each of the above categories, divided on the basis of compound headedness and constituent position (when applicable). The table shows that errors affected the whole compound more frequently than the individual constituents (521 vs. 269). The most frequent error for head-initial compounds was omission, whereas head-final compounds were most frequently substituted with simple words, x2(3) ¼ 8.21, p ¼ .0418; compound neologisms were very rare. With regard to the individual constituents, modifier constituents were affected more frequently by lexical substitutions, whereas omissions were most often observed in the case of head constituents, x2(2) ¼ 25.66, p ¼ .0001. However, this error pattern partially changes when considering compound headedness: The effect is confirmed in head-initial compounds, x2(2) ¼ 21.63, p ¼ .0001, but not in head-final ones, as in this latter case omission errors were relatively rare and equally distributed between the two constituents, x2(2) ¼ 4.49, p ¼ .1062.2 Phonemic errors were generally uncommon.
DISCUSSION In this study, we analysed the naming performance on compound nouns in a group of Italian patients affected by aphasia. The purpose of the experiment was to investigate whether the head-modifier structure of compounds influences word retrieval, and it aimed specifically at replicating the effect described by Marelli et al. (2009) with a constituent priming task in a sample of neurologically unimpaired participants. Forty-five patients, selected for their disproportionate impairment in naming complex nouns with respect to simple nouns, participated in the study. The naming impairment observed in the aphasic participants had a greater effect on the retrieval of modifiers than of head constituents. In fact, modifier constituents were more difficult
2
With regard to the effects of the covariates on the probability of specific error types, the only association found was between aphasia severity and constituent omissions: Constituents were omitted more frequently by patients suffering from more severe impairments (p ¼ .0246). Cognitive Neuropsychology, 2014, 31 (1 –2)
33
MARELLI ET AL.
Table 3. Analysis of error types in compound naming Head-initial compounds Error type
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
Constituent omissions Lexical substitutions Phonemic errors
Head-final compounds
1st constituent
2nd constituent
1st constituent
2nd constituent
42 20 12
19 49 8
7 55 14
10 27 6
Whole-word omissions Whole-word substitutions Circumlocutions Neologistic compounds
120 83 37 8
101 122 40 10
Note: Number of errors (out of a total of 1350 observations), organized by headedness and constituent position. The upper section of the table summarizes constituent-specific errors, whereas the lower section reports errors affecting the whole compound.
to retrieve than their head counterparts. These results are in line with other studies investigating the headedness effect in compound processing (Jarema et al., 1999; Marelli et al., 2009) and indicate that the head-modifier structure is represented in the mental lexicon. In other words, constituents are represented on the basis of the role (modifier vs. head) they play in the compound structure: Head constituents, which usually specify the most important semantic and lexical traits of the whole compound, are arguably most salient from a cognitive point of view and are easier to retrieve than modifier constituents, which are in turn impacted more heavily by the patients’ naming impairment. This interpretation is supported by the error type analysis. In fact, modifiers were most often substituted with other words, suggesting that the head constituent is the most important during the conceptual combination procedure: Retrieval of the head triggers a lexical search for the other element, which is probably doomed to failure due to the patients’ naming impairment, resulting in a majority of substitutions compared to omissions of the modifiers. However, the larger impairment for modifiers was observed principally in head-final compounds. The difference in the performance vis a` vis
compound headedness is in line with the priming effects observed on normal subjects, as reported by Marelli et al. (2009). In fact, a comparison of the reaction time effects in word recognition by healthy participants with the naming accuracy by aphasic patients (Figure 2) shows that the two patterns of results are quite similar. Both priming facilitation and patients’ accuracy can be assumed to reflect the ease of access to constituent representations. Applying this line of reasoning, the results of both studies indicate that in head-final compounds the head is easier to access than the modifier, whereas in head-initial compounds both constituents are similarly easy to access. In line with the theory proposed by Williams (1981), the head-final structure seems to be preferred in the mental processing of compound words. The right-hand head rule would appear to be not only a theoretical principle, but also a general psycholinguistic property of the mental lexicon: The head-final structure is assumed to be the default organization of morphologically complex words. However, in accordance with Marelli et al. (2009) and Marelli and Luzzatti (2012), we propose that this assumption is not due to a prescriptive rule, but rather follows the distributional properties of the Italian morphology of compounds3; in fact, the majority of Italian
3 The prescriptive and the distributional interpretation of the right-hand head rule (Williams, 1981) are not necessarily at odds. Di Sciullo and Williams (1987) proposed that only head-final compounds would be real morphological constructs, whereas headinitial compounds are syntactic strings imported into the lexicon. Head-final compounds may thus be more numerous because they are generated through a standard morphological operation.
34
Cognitive Neuropsychology, 2014, 31 (1 –2)
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
COMPOUND HEADEDNESS IN PICTURE NAMING
Figure 2. Comparison between the results of the present study (left panel, percentage of correct constituent retrieval) and priming effects in lexical decision by healthy subjects (right panel, adapted from Marelli et al., 2009; permission granted from John Benjamins’ Publishing Company).
compounds adopts a head-final structure (Marelli & Luzzatti, 2012). In other words, it is to be expected that the representations of head-final compounds would be more structured as these compounds constitute the usual way in which two nominal constituents are combined in Italian (Marelli et al., 2009). This would be in line with constituents being more difficult to process in head-final compounds, as suggested by the interaction between headedness and aphasia severity: The complex and structured representation associated with head-final compounds makes constituent retrieval more difficult even for less impaired patients. The assumption of right-headedness also fits the pattern of omission errors in the error analysis: The head is often omitted in head-initial compounds, in line with the hypothesis that by default it is expected to be in the final position, whereas constituent omissions affect heads and modifiers equally (and are markedly less observed) in head-final compounds. The results of the present experiment are not a mere replica of the findings described in Marelli et al. (2009). In the first place, the present study employed a different set of compound stimuli to that used in the psycholinguistic experiment. This
difference strongly supports the reliability of the observed effect and rules out the possibility that the phenomenon may be due to an unfortunate selection of experimental stimuli. In the second place, the same effect was observed in two different tasks: The psycholinguistic study required a lexical decision to be made, thus tapping into the input stages of lexical processing; in the present experiment, patients were asked to name objects on the basis of pictures/definitions, thus having to rely on output procedures. This converging evidence provides an indication as to which level of representation generates the observed effect: In fact, the same level of processing has necessarily to be involved in both tasks. The most likely explanation is that information regarding a compound structure is stored at a central and amodal processing level; the conceptual/semantic level appears to be the best candidate to account for the headedness effect. In line with Marelli and Luzzatti (2012), the semantic system itself would be involved since the conceptual combination between constituents (Gagne´ & Spalding, 2009) takes place at this level, and the procedure has to be driven by previous knowledge about how constituents are combined (i.e., which constituent is the head and which the Cognitive Neuropsychology, 2014, 31 (1 –2)
35
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
MARELLI ET AL.
modifier). This interpretation is in line with results suggesting that the nature of headedness effects is primarily semantic (Marelli & Luzzatti, 2012) and with the assumption that headedness depends on the semantic association between a compound and its constituents (e.g., El Yagoubi et al., 2008). This hypothesis is also supported by the observed effect of semantic transparency: When they are part of more transparent compounds, constituents are correctly retrieved with a higher frequency, thus indicating that constituent retrieval is easier when the compound meaning can be obtained through a conceptual combination procedure. However, data have been reported that are in contrast with this “semantic hypothesis”, so underlining the importance of lexical factors in determining the head-modifier structure of a compound (e.g., Frisson, Niswander-Klement, & Pollatsek, 2008; Marelli et al., 2013). An alternative interpretation locates the source of the headedness effect at lemma level (Levelt et al., 1999), a module that stores grammatical and lexical information about words. This level is assumed to be central as well as amodal and to contain the same information (grammatical gender, grammatical class) as that shared by a compound and its head constituent. Therefore compound headedness may be represented in a lemma architecture that comprises both compound and constituent representations, as is the case of the multiple-lemma model proposed by Marelli et al. (2012). In this approach, the access to the constituent lemma nodes is subordinated to the activation of the compound node. Headedness can be straightforwardly formalized as the difference in strength of the connections between the compound and its constituents: The link between a compound and its head is assumed to be stronger than that between a compound and its modifier. As the present result suggests, in Italian the weight difference would be greater in head-final (hierarchically organized) than in head-initial compounds. A multiple-lemma architecture is also suggested by the significant effects of the age of acquisition and compound frequency variables, which are associated with the whole compound but emerged as significant predictors of the
36
Cognitive Neuropsychology, 2014, 31 (1 –2)
accuracy in producing the individual constituents. In other words, access to compound properties would be a prerequisite for the constituent retrieval, which clearly fits well the lemma network described by Marelli et al. (2012). Indeed, it is possible that the information on headedness is coded at both the lemma and the semantic levels as both lexical (i.e., pertaining to the lemma) and semantic aspects contribute to the headedness definition. The compound structure, specifying head and modifier positions, is arguably stored in the compound lemma, which is by principle independent of the meanings of the constituents. However, the actual head-modifier roles have to be represented in the semantic system, since their associated properties are used (together with relational information, see Gagne´ & Spalding, 2009) in the conceptual combination of constituent meanings. The results obtained from the present study do not provide a definitive answer regarding the representation of compound headedness, and further investigations are needed to clarify the issue. However, the study does provide crucial evidence to the headedness effect being originated at central processing levels, disconfirming the hypothesis that previously observed phenomena may be a by-product of the employed paradigm and/or be specific to written word processing. The approach adopted in the present study was fundamental for observing these results. In fact, headednessassociated effects have usually small effect size and are qualified by interactions with other variables (e.g., Marelli & Luzzatti, 2012), and group studies are often needed to obtain statistically significant effects. For these reasons, the error analysis in neuropsychological patients is important: This approach magnifies effects that are very limited in size (Semenza & Mondini, 2006), thus facilitating assessment of the relatively small influence of headedness and enabling testing of the predictions of psycholinguistic models. Manuscript received 27 March 2013 Revised manuscript received 26 July 2013 Revised manuscript accepted 24 October 2013 First published online 6 December 2013
COMPOUND HEADEDNESS IN PICTURE NAMING
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
REFERENCES Baayen, R. H. (2008). Analyzing linguistic data. A practical introduction to statistics using R. Cambridge: University Press. Badecker, W. (2001). Lexical composition and the production of compounds: Evidence from errors in naming. Language and Cognitive Processes, 16, 337– 366. Bien, H., Levelt, W., & Baayen, R. (2005). Frequency effects in compound production. PNAS, 102, 17876 –17881. De Mauro, T., Mancini, F., Vedovelli, M., & Voghera, M. (1993). Lessico di frequenza dell’italiano parlato. Milano: Etaslibri. Di Sciullo, M., & Williams, E. (1987). On the definition of word. Cambridge: MIT Press. Dunabeitia, J. A., Perea, M., & Carreiras, M. (2007). The role of the frequency of constituents in compound words: Evidence from Basque and Spanish. Psychonomic Bulletin & Review, 14(6), 1171 –1176. El Yagoubi, R., Chiarelli, V., Mondini, S., Perrone, G., Danieli, M., & Semenza, C. (2008). Neural correlates of Italian nominal compounds and potential impact of headedness effect: An ERP study. Cognitive Neuropsychology, 25, 559 –581. 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. Gagne´, C. L., & Spalding, T. L. (2007). Conceptual combination: Implications for the mental lexicon. In G. Libben & G. Jarema (Eds.), The representation and processing of compound words (pp. 145 –169). Oxford: University Press. Gagne´, 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. Goldman, R. (2007). Compounding in Aphasia: A crosslinguistic review (BA Thesis). Swarthmore College, Department of Linguistics, Swartmore, PA. Retrieved from http://hdl.handle.net/10066/10186 Guevara, E., & Scalise, S. (2008, February). Heads in compounding: Variation and patterns. 13th international morphology meeting – Book of Abstracts, Vienna, Austria, p. 43. Huber, W., Poeck, K., Weniger, D., & Willmes, K. (1983). Der Aachener Aphasie Test (AAT). Go¨ttingen: Hogrefe.
Hyo¨na¨, J., & Pollatsek, A. (1998). Reading Finnish compound words: Eye fixations are affected by component morphemes. Journal of Experimental Psychology: Human Perception and Performance, 24, 1612– 1627. Jaeger, T. F. (2008). Categorical data analysis: Away from ANOVAs (transformation or not) and towards logit mixed models. Journal of Memory and Language, 59, 434–446. Janssen, N., Bi, Y., & Caramazza, A. (2008). A tale of two frequencies: Determining the speed of lexical access for Mandarin Chinese and English compounds. Language and Cognitive Processes, 23, 1191– 1223. Jarema, G., Busson, C., Nikolova, R., Tsapkini, K., & Libben, G. (1999). Processing compounds: A cross-linguistic study. Brain and Language, 68, 362–369. Jarema, G., Perlak, D., & Semenza, C. (2009). The processing of compounds in bilingual aphasia. Aphasiology, 24, 126– 140. Kraemer, H. C., & Blasey, C. M. (2004). Centring in regression analyses: A strategy to prevent errors in statistical inference. International Journal of Methods in Psychiatric Research, 13, 141–151. Levelt, W. J., Roelofs, A., & Meyer, A. S. (1999). A theory of lexical access in speech production. The Behavioral and Brain Sciences, 22, 1–38; discussion 38 –75. Libben, G., Gibson, M., Yoon, Y. B., & Sandra, D. (2003). Compound fracture: The role of semantic transparency and morphological headedness. Brain and Language, 84, 50–64. Luzzatti, C., Willmes, K., & De Bleser, R. (1996). Aachener Aphasie Test: versione italiana (2nd ed.). Firenze: Organizzazioni Speciali. Lynott, D., & Ramscar, M. J. A. (2001, September). Can we model conceptual combination using distributional measures? Twelfth Irish Conference on Artificial Intelligence and Cognitive Science, Maynooth, Ireland, pp. 1– 10. Marelli, M., Aggujaro, S., Molteni, F., & Luzzatti, C. (2012). The multiple-lemma representation of Italian compound nouns: A single case study of deep dyslexia. Neuropsychologia, 50, 852–861. Marelli, M., Aggujaro, S., Molteni, F., & Luzzatti, C. (2013). Understanding the mental lexicon through neglect dyslexia: A study on compound noun reading. Neurocase, 19, 128–144. Marelli, M., Crepaldi, D., & Luzzatti, C. (2009). Head position and the mental representation of nominal Cognitive Neuropsychology, 2014, 31 (1 –2)
37
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
MARELLI ET AL.
compounds: A constituent priming study in Italian. The Mental Lexicon, 4, 430– 454. Marelli, M., & Luzzatti, C. (2012). Frequency effects in the processing of Italian nominal compounds: Modulation of headedness and semantic transparency. Journal of Memory and Language, 66, 644 –664. Mondini, S., Jarema, G., Luzzatti, C., Burani, C., & Semenza, C. (2002). Why is “Red Cross” different from “Yellow Cross”? A neuropsychological study of noun –adjective agreement within Italian compounds. Brain and Language, 81, 621 –634. Mondini, S., Luzzatti, C., Zonca, G., Pistarini, C., & Semenza, C. (2004). The mental representation of Verb-Noun compounds in Italian: Evidence from a multiple single-case study in aphasia. Brain and Language, 90, 470– 477. Semenza, C., Arcara, G., Facchini, S., Meneghello, F., Ferraro, M., Passarini, L., . . . Mondini, S. (2011). Reading compounds in neglect dyslexia: The headedness effect. Neuropychologia, 49, 3116–3120. Semenza, C., De Pellegrin, S., Battel, I., Garzon, M., Meneghello, F., & Chiarelli, V. (2011). Compounds
38
Cognitive Neuropsychology, 2014, 31 (1 –2)
in different aphasia categories: A study on picture naming. Journal of Clinical and Experimental Neuropsychology, 33, 1099–1107. Semenza, C., Luzzatti, C., & Carabelli, S. (1997). Morphological representation of compound nouns: A study on Italian aphasic patients. Journal of Neurolinguistics, 10, 33–43. Semenza, C., & Mondini, S. (2006). Neuropsychology of compound words. In G. Libben & G. Jarema (Eds.), The representation and processing of compound words (pp. 71 –95). Oxford: Oxford University Press. 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. Williams, E. (1981). On the notions “lexically related”, and “head of a word”. Linguistic Inquiry, 12, 245– 274. Zwitserlood, P. (1994). The role of semantic transparency in the processing and representation of Dutch compounds. Language and Cognitive Processes, 9, 341–368.
COMPOUND HEADEDNESS IN PICTURE NAMING
APPENDIX
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
Compound stimuli employed in the experimental phase of the experiment Headedness
Stimulus
Translations
Literal translation
Head-initial Head-initial Head-initial Head-initial Head-initial Head-initial Head-initial Head-initial Head-initial Head-initial Head-initial Head-initial Head-initial Head-initial Head-initial Head-initial Head-final Head-final Head-final Head-final Head-final Head-final Head-final Head-final Head-final Head-final Head-final Head-final Head-final Head-final
acquasanta arcobaleno busta paga cabina telefonica calzamaglia capolinea capotavola carro armato carro attrezzi carta carbone cassaforte macchina fotografica montepremi pesce spada pianoterra vagone ristorante aeroporto aliscafo autostrada capolavoro cronometro fotocopia funivia mezzaluna microscopio motoscafo pronto soccorso seggiovia supermercato terremoto
holy water rainbow payslip telephone booth tights end of the line head of the table tank tow truck carbon paper strongbox camera jackpot swordfish ground floor dining car airport hydroplane highway masterpiece stopwatch photocopy cable car dicing knife microscope motorboat first aid chairlift supermarket earthquake
water-holy arch-lightning envelope salary stateroom-telephonic sock-knit head-line head-table wagon-armed wagon-tools paper-coal box-strong machine-photographic mountain-prizes fish-sword floor-ground coach-restaurant air-port wings-hull car-street head-work time-meter photo-copy cables-way half– moon small-look movement-hull ready-rescue seat-way super-market lands-movement
Note: Constituents in italics are usually part of neoclassical compounds and are not free morphemes.
Cognitive Neuropsychology, 2014, 31 (1 –2)
39
Cognitive Neuropsychology Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/pcgn20
The representation of compound headedness in the mental lexicon: A picture naming study in aphasia a
b
b
Marco Marelli , Giusy Zonca , Antonella Contardi & Claudio Luzzatti a
c
Center for Mind/Brain Sciences, University of Trento, Rovereto, Italy
b
S. Maugeri Foundation IRCCS, Montescano Medical Center, Montescano, Italy c
Department of Psychology, University of Milano-Bicocca, Milano, Italy Published online: 07 Dec 2013.
Click for updates To cite this article: Marco Marelli, Giusy Zonca, Antonella Contardi & Claudio Luzzatti (2014) The representation of compound headedness in the mental lexicon: A picture naming study in aphasia, Cognitive Neuropsychology, 31:1-2, 26-39, DOI: 10.1080/02643294.2013.860024 To link to this article: http://dx.doi.org/10.1080/02643294.2013.860024
PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions
Cognitive Neuropsychology, 2014 Vol. 31, Nos. 1 –2, 26 –39, http://dx.doi.org/10.1080/02643294.2013.860024
The representation of compound headedness in the mental lexicon: A picture naming study in aphasia Marco Marelli1, Giusy Zonca2, Antonella Contardi2, and Claudio Luzzatti3 1
Center for Mind/Brain Sciences, University of Trento, Rovereto, Italy S. Maugeri Foundation IRCCS, Montescano Medical Center, Montescano, Italy 3 Department of Psychology, University of Milano-Bicocca, Milano, Italy
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
2
Most compound words are constituted of a head constituent (e.g., light in moonlight) and a modifier constituent (e.g., moon in moonlight); the information transmitted by these head-modifier roles is fundamental for defining the grammatical and semantic properties of the compound and for identifying a correct combination of the constituents at the conceptual level. The objective of this study is to assess how lexical processing in aphasia is influenced by the head-modifier structure of nominal compounds. A picture-naming task of 35 compounds with head-initial (pescespada, swordfish, literally fishsword) and head-final (autostrada, highway, literally carroad) forms was administered to 45 Italian aphasic patients, and their accuracy in retrieving constituents was analysed with a mixed-effects logistic regression. The interaction between headedness and constituent position was significant: The modifier emerged as being more difficult to retrieve than the head, but only for head-final compounds. The results are consistent with previous data from priming experiments on healthy subjects and provide convincing evidence that compound headedness is represented at central processing levels. Keywords: Compound words; Picture naming; Compound headedness; Aphasia.
Research on the representation of compound words has a long tradition in psycholinguistic literature: Since the seminal paper by Taft and Forster (1976), many studies, most employing chronometric methods (i.e., measuring behavioural response times or fixation durations in reading), have dealt with the question as to whether morpheme representations are involved in the processing of complex words. The issue was also
addressed qualitatively and quantitatively within the classical cognitive neuropsychological frame, by investigating the errors made by patients with acquired language disorders (for a review see Goldman, 2007; Semenza & Mondini, 2006). These analyses have largely confirmed the results obtained by healthy participants and have been profitably used to test the predictions of psycholinguistic models.
Correspondence should be addressed to Marco Marelli, CIMeC– Palazzo Fedrigotti, Corso Bettini 31, Room c111, 38068 Rovereto (TN), Italy (E-mail: [email protected]) Preliminary results of this study were presented at the 48th Annual Meeting of the Academy of Aphasia (Athens, 2010) and the 29th European Workshop on Cognitive Neuropsychology (Bressanone, 2011). The study was supported by a grant awarded to C. L. by the University of Milano-Bicocca (Fondo d’Ateneo per la Ricerca (FAR) 2009).
26
# 2013 Taylor & Francis
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
COMPOUND HEADEDNESS IN PICTURE NAMING
The first phenomenon that has been verified is that compounds are not (or not only) stored in the mental lexicon as unique and directly accessed representations. Results suggesting whole-word access have been reported (Janssen, Bi, & Caramazza, 2008), but most data clearly indicate that constituent representations are accessed when processing a compound word, as extensively demonstrated in psycholinguistics by studies adopting priming paradigms (e.g., Jarema, Busson, Nikolova, Tsapkini, & Libben, 1999; Libben, Gibson, Yoon, & Sandra, 2003; Zwitserlood, 1994) and/or reporting constituent frequency effects (e.g., Bien, Levelt, & Baayen, 2005; Dunabeitia, Perea, & Carreiras, 2007; Hyo¨na¨ & Pollatsek, 1998). This same hypothesis is supported by neuropsychological data, indicating that a similar process is likely to occur in compound production. Indeed, constituent access is strongly supported by the phenomenon described by Semenza, Luzzatti, and Carabelli (1997) and Mondini, Luzzatti, Zonca, Pistarini, and Semenza (2004): When asked to name verb – noun compounds (e.g., aspirapolvere, vacuumcleaner, literally inhaledust), patients whose performance was characterized by a grammatical class effect (with a better performance on nouns than verbs), made significantly more errors on the verb (aspira) constituent than on the noun constituent (polvere); since compounds were globally nouns, the hypothesis of a direct access is not consistent with these results. However, compound processing cannot rely only on constituents, and properties associated with the whole compound are also likely to be retrieved. Not only is this hypothesis supported by theoretical reasons (compound properties cannot be totally deduced from constituent information; Badecker, 2001; Marelli, Aggujaro, Molteni, & Luzzatti, 2012), it also has empirical backing: In the cases reported by Semenza et al. (1997; see also Semenza et al., 2011), errors were produced that maintained the original morphological structure of the stimuli, while Badecker (2001) described a patient who, when naming pictures representing compound words, was aware that his responses were incomplete if a constituent
was missing. Moreover, Marelli et al. (2012) have shown that favouring the access to structure properties helped G.R., a patient suffering from deep dyslexia with morphological impairment, in compound retrieval (see also Mondini, Jarema, Luzzatti, Burani, & Semenza, 2002). These data suggest that morphological structure is stored separately from the lexical representations of the constituents and is used during processing to form the compound word. Marelli et al. (2012) proposed that this structure is represented as an independent node in the lemma, which is the amodal and abstract processing level where grammatical and lexical properties of words would be stored (Levelt, Roelofs, & Meyer, 1999). This abstract compound node would be linked to separate and modality-specific representations of constituents at more peripheral (i.e., orthographic and phonological) levels. However, there is still very little in the way of data available as to how this information is used during processing. Are constituents hierarchically organized? In other words, does one constituent have a privileged role over the other? From a theoretical point of view, in most compounds it is possible to identify a head and a modifier, the head being the constituent specifying lexical (grammatical gender—in languages such as Italian where nouns have a masculine or feminine gender, irrespective of the natural gender of the underlying concept; grammatical class) and semantic (category, principal traits) properties of the whole compound. In German languages the roles are positionally coded, with the head always in the rightmost position (Williams, 1981). This is such a regular phenomenon that Williams (1981) even proposed a right-hand head rule, indicating that the principal morpheme of a complex word is always to be found in the final position. However, several languages (such as Romance languages) also have head-initial compounds (e.g., pescespada, swordfish, literally fishsword), or have only head-initial compounds, as in Hebrew (e.g., tapuach adamah, potato, literally apple earth). Even if the latter form is less frequent than the head-final form (this holds true also in Italian and other Romance languages, see Cognitive Neuropsychology, 2014, 31 (1 –2)
27
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
MARELLI ET AL.
Guevara & Scalise, 2008; Marelli & Luzzatti, 2012), the fact that it exists at all suggests that head-initial compounding is also a possible way to create new words. In this scenario, it is important for psycholinguistic and neurolinguistic studies to investigate whether head-modifier roles are represented in the mental lexicon, and whether head-initial and head-final structures are differentially processed. Given their particular properties, Romance languages constitute an ideal benchmark when addressing this issue. Recent results have indeed indicated that head-modifier roles have an important effect on compound processing. Patients affected by neglect dyslexia produce fewer neglect errors when reading head-initial than when reading head-final compounds (Marelli, Aggujaro, Molteni, & Luzzatti, 2013; Semenza et al., 2011). Moreover, errors made by bilingual English – French speaking subjects with aphasia in a series of tasks involve the modifier more often than the head (Jarema, Perlak, & Semenza, 2009). These data indicate that the head is relatively more salient and thus easier to process than the modifier. However, these studies did not fully compare the two possible compound structures: In the former, the analysis was limited to the performance on the first constituent due to the peculiarities of the neglect disorder, and the latter case focused on crosslinguistic comparison. As a result, the question as to whether head-initial and head-final compounds are differentially processed remains open. In earlier studies (Marelli, Crepaldi, & Luzzatti, 2009; Marelli & Luzzatti, 2012), we investigated the representation of the head-modifier structure in Italian by evaluating the processing of written noun– noun nominal compounds in healthy subjects. The results indicated that head-initial and head-final structures do indeed have different representations in the mental lexicon. In a constituent priming experiment (Marelli, Crepaldi, & Luzzatti, 2009), the participants recognized a compound more quickly when it was preceded by its head than when the prime was the modifier constituent. However, the effect was significant only in head-final compounds,
28
Cognitive Neuropsychology, 2014, 31 (1 –2)
whereas in head-initial compounds the magnitude of the priming effects did not differ between head and modifier, suggesting that only head-final compounds are built on a hierarchical structure, whereas head-initial compounds are more similar to lexicalized syntactic units. This hypothesis was further expanded upon in an investigation of frequency effects when reading Italian compounds (Marelli & Luzzatti, 2012). Constituent frequency effects facilitate compound recognition when embedded in a head-final structure (higher frequency leads to shorter response times), but are inhibitory when the compound is head-initial (higher frequency leads to longer response times). This indicates that when Italian speakers process a compound noun in their language, they take the head-final structure as the default compound representation. Since constituent frequency was found to interact with compound transparency, this effect was interpreted in semantic terms: When trying to build the combined concept on the basis of constituent meanings, the system assumes a head-final representation. When this assumption is not satisfied, the headmodifier roles must be reassigned in a time-consuming processing. This model assumes that compound processing is ultimately semantically driven, implying a conceptual combination procedure (Gagne´ & Spalding, 2007): Lexical information acquired during early processing phases interacts with stored distributional knowledge in order to derive the combined concept associated with a compound word. Since this distributional information is thought to play a role in semantic processing (Gagne´ & Spalding, 2009; Lynott & Ramscar, 2001; Marelli & Luzzatti 2012), it would be represented at central processing stages, which are involved in many different tasks, including those tapping into output processing levels. In other words, the hierarchical effect of the headfinal structure (as described above) would not only be found in silent-reading experiments, but also in tasks requiring the production of compound nouns. If this hypothesis is correct, converging evidence should emerge from the analysis of error patterns in patients with aphasia. The
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
COMPOUND HEADEDNESS IN PICTURE NAMING
present study tests this prediction in a picturenaming experiment carried out with a large sample of Italian aphasic patients. The study was organized in two main phases: The first phase consisted of a picture-naming task to identify patients with greater problems in naming morphologically complex words than in naming simple words; in the second phase, the performance of these patients was further analysed to assess the extent to which their naming disorder affected the first versus the second constituent of compound nouns and to investigate whether this error pattern was influenced by the head-modifier structure. Three possible outcomes were envisaged: 1. The head-modifier structure could impact the naming performance in a similar manner to that observed in written word processing in healthy individuals (Marelli et al., 2009): The most prominent role in head-final compounds would be played by heads rather than modifiers; conversely, when producing head-initial compounds, head and modifier constituents would be equally easy to retrieve. 2. The results could indicate that the head is easier to retrieve irrespective of the headmodifier structure involved, disconfirming that head-initial and head-final compounds undergo different processing. 3. No difference could be found between head and modifier performances, suggesting that the head-modifier structure is not represented in the mental lexicon. Outcomes 2 and 3 would indicate that data by Marelli et al. (2009) may depend on the task employed in their reading experiments and that the hierarchical effect in head-final compounds could be specific for written word processing. On the contrary, Outcome 1 would suggest that the hypothesis by Marelli and Luzzatti (2012) is correct, indicating that the head-modifier structure is stored at central levels of processing, and that this very structure has different representations on the basis of the head position.
PRELIMINARY ANALYSIS Method Participants A total of 122 patients with focal left-hemisphere (LH) brain lesions were examined. All participants were suffering from mild-to-moderate aphasic impairments, as evaluated by the Italian version of the Aachen Aphasia Test (Luzzatti, Willmes, & De Bleser, 1996), or from residual aphasia—that is, the patients’ psychometric assessment did not indicate an aphasic disorder but signs of aphasia were present either in their spontaneous speech or in their performance on clinical tasks. They were recruited at the “Fondazione Salvatore Maugeri” Rehabilitation Centre, Montescano. The participants had previously been diagnosed as suffering from Broca’s aphasia (12), anomic aphasia (24), conduction aphasia (8), Wernicke’s aphasia (5), residual aphasia (64), or other aphasic forms (10). Materials A confrontation naming battery was employed, consisting of 145 stimuli, of which 110 items were presented as pictures, while 35 items were presented as definitions. The battery was an adapted version of the task employed by Mondini et al. (2004) in an earlier study. Fifty-seven stimuli were objects denoted by a simple noun, while 88 stimuli were objects denoted by a complex noun. Complex targets were either derived nouns (36 stimuli; e.g., campanile, bell tower, derived from campana, bell) or compound nouns (52 stimuli; e.g., pescespada, swordfish, literally fishsword). Spoken frequency measures were collected from the Lessico di frequenza dell’Italiano Parlato (LIP) corpus for each stimulus (De Mauro, Mancini, Vedovelli, & Voghera, 1993). A rating study on 12 unimpaired participants was also run to assess age of acquisition and imageability of the battery stimuli. Participants were asked to rate each word on a 7-point scale. Procedure Stimuli were presented in a unique list, comprising both simple and complex nouns in a randomized Cognitive Neuropsychology, 2014, 31 (1 –2)
29
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
MARELLI ET AL.
order. Definitions were presented first, after which the experimenter showed the pictures. The patients were asked to produce the noun corresponding to each definition/picture. Responses were considered correct only if given within a 2-s interval from stimulus presentation. Minor phonological alterations and articulatory distortions were ignored.
EXPERIMENTAL ANALYSIS
Data analysis The patients’ performances were individually assessed in order to reveal dissociations between complex and simple word naming. A logistic regression analysis was carried out on the data generated by each participant to predict the likelihood of a correct response on the basis of the stimulus category (complex vs. simple noun). Stimulus length (in syllables), imageability, age of acquisition, and spoken frequency were also introduced as covariates to rule out their influence on the patients’ performance. A backward procedure was adopted to simplify the statistical models: Starting from a full-factorial model, predictors were removed one by one when their absence did not significantly affect the overall goodness of fit. If after this procedure the effect of morphological complexity was significant (p , .10), the participant was included in the subsequent phase (Experimental Analysis, see below).
Materials Two sets of compound nouns were extracted from the naming task analysed in the previous phase. The first set was composed of 16 compounds in which the first constituent was the morphological head of the compound noun (head-initial compounds; e.g., pescespada, swordfish, literally fishsword); the second set included 14 compounds in which the second constituent was the morphological head of the compound (head-final compounds; e.g., autostrada, highway, literally carroad). All the stimuli are reported in the Appendix, together with their English translations. The sets were matched for spoken frequency of both the constituents, imageability, age of acquisition, frequency, stimulus length of the compound noun, naming agreement (in a sample of 38 control subjects), and proportion of stimulus types (pictures vs. definitions). Frequency measures were extracted from the LIP database (De Mauro et al., 1993). We also considered semantic transparency, which is widely accepted as affecting compound retrieval and headedness-associated effects (e.g., Libben et al., 2003; Marelli & Luzzatti, 2012). Semantic transparency was acquired by means of a rating study on 12 unimpaired participants, who were asked to rate the degree to which each compound meaning could be derived from the meaning of its constituents on a 4-point scale. The matching for the described psycholinguistic variables is summarized in Table 1.
Results Of the 122 patients assessed, 45 showed disproportionate impairment in producing morphologically complex words in comparison to simple words. This dissociation cannot be explained in terms of stimulus length, imageability, frequency, or age of acquisition, which have independent effects (if any) on the patients’ performance. The patients who participated in the study had previously been diagnosed as suffering from anomic aphasia (10), residual aphasia (28), Broca’s aphasia (3), conduction aphasia (2), and transcortical motor aphasia (2).
30
Cognitive Neuropsychology, 2014, 31 (1 –2)
Method Participants In this phase, the performance of the 45 patients who showed disproportionate impairment of morphologically complex words was subjected to further analysis.
Procedure The performance on the compound stimuli was reanalysed, focusing on accuracy in retrieving the individual constituents—that is, for each stimulus,
COMPOUND HEADEDNESS IN PICTURE NAMING
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
Table 1. Variables matching in the compound sets Variable
Range
Head-initial compounds
Head-final compounds
t-value
p
Compound frequency 1st constituent frequency 2nd constituent frequency Stimulus length (syllables) Imageability Semantic transparency Age of acquisition Naming agreement
0– 3.21 0– 5.01 0– 5.96 3– 8 1– 7 1– 4 1– 7 0– 1
0.57 3.12 2.52 5.06 4.94 2.84 3.89 0.69
0.96 2.25 2.79 4.29 5.18 2.82 3.73 0.71
1.27 1.44 0.43 1.95 0.69 0.08 0.59 0.18
.2174 .1627 .6711 .0631 .4911 .9335 .5632 .8596
Note: Word frequency values, extracted from the LIP corpus (490,000 word forms; De Mauro et al., 1993), were log-transformed.
we considered whether the constituents were correctly reported, irrespective of the accuracy in retrieving the whole compound. As a consequence, two distinct dependent measures were associated to each stimulus—that is, accuracy in reporting the first constituent, and accuracy in reporting the second constituent. Minor phonological alterations and articulatory distortions were ignored as before. Data analysis The accuracy in retrieving either constituent1 was analysed employing a logistic regression model. The peculiar properties of the dependent variable (being measured twice for each stimulus) and the intention of running a group study would inevitably lead to issues related to the non-independency of observations in a traditional generalized linear model framework; hence a mixed-effects analysis (Jaeger, 2008) was employed. If a nested structure including the random effects of participants, compounds, and constituents is specified, it is possible to prevent potential problems in parameter estimation related to the fact that observations are not independent. Stimulus type (definition vs. figure) was also included as an additional random effect. This statistical technique also guarantees
that significant effects are estimated net of the differences that characterize the individual patients: The results would thus characterize the whole group and not just a minor sample of outlying participants. The main effect of interest was the interaction between the position of the constituent considered (first vs. second Constituent Position) and Headedness of the compound (head-initial vs. head-final compounds). We also tested how this interaction was modulated by aphasia severity (as measured by the Token Test score; Huber, Poeck, Weniger, & Willmes, 1983; Luzzatti et al., 1996). The psycholinguistic variables described above were also introduced as covariates. All predictors were mean-centred, in order to ensure better estimates of the parameters in the subsequent analyses (Kraemer & Blasey, 2004). A backward procedure was adopted to simplify the statistical model: Starting from a full-factorial model, predictors were removed one by one when their absence did not significantly affect the overall model fit (on the basis of a goodness-of-fit chisquare test). Parameters were not removed when the goodness-of-fit test was significant, or if they were part of higher order interactions. Once the model was fitted, atypical outliers were identified
1 It may be argued that constituent accuracy could also be tested as a predictor in a regression analysis on compound naming accuracy, to assess to what extent the probability of retrieving the constituents affects the probability of retrieving the whole word. This effect could be modulated by constituent position and/or compound headedness, thus providing information on the representation of head-modifier roles in the mental lexicon. This analysis was run in the present study, but the interactions of interest between constituent accuracies, constituent position, and headedness were not significant—that is, constituent accuracy was helpful for compound retrieval irrespective of position and head-modifier roles.
Cognitive Neuropsychology, 2014, 31 (1 –2)
31
MARELLI ET AL.
and removed (employing 2.5 standard deviations of the residual errors as criterion). The model was then refitted to ensure that results were not driven by these outliers (Baayen, 2008).
patient’s disorder, the larger the effect of compound headedness was: Constituents of head-initial compounds are easier to retrieve, but this effect emerges only in participants who are relatively less impaired (see Figure 1). The main effects of age of acquisition, compound frequency, and semantic transparency were also significant. Constituents are easier to retrieve when they belong to compounds acquired earlier, more transparent, and with higher word frequency.
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
Results Constituent retrieval The final model included the interaction between headedness and constituent position and the interaction between headedness and aphasia severity, as well as age of acquisition, semantic transparency, and compound frequency as significant predictors. The parameters of the model are reported in Table 2 together with a summary of the simplification procedure. We were mainly interested in the interaction between constituent position and compound headedness, which emerged as significant. Patients showed greater impairment in retrieving the modifier than in retrieving the head constituent, but this effect emerged only when naming head-final compounds (accuracy: 44% vs. 50%) and not when naming head-initial compounds (accuracy: 55% vs. 55%). An interaction between headedness and aphasia severity was also found. The less severe the
Analysis by error types The patients’ responses were further classified according to error type. Errors could affect either the individual constituents or the whole compound. Three categories of errors were considered for the individual constituents: constituent omissions, lexical substitutions, and phonemic errors. Constituent omissions (e.g., “pesce”, fish, for pescespada, swordfish) also included latencies and word repairs for a specific constituent (e.g., “spada, ah no, pescespada”, sword, uh no, swordfish, for pescespada). Lexical substitutions included all cases in which a constituent was substituted with an existing related word (e.g., “pescecane”, shark, literally
Table 2. Results of the mixed-effects analysis on constituent retrieval x2
p
Removal order
Estimate
z-value
p
— — — — 10.6451 5.9668 4.9395 4.7972 4.9772 0.9759 0.5734 0.3307 0.1451 0.1167 0.0287
— — — — .0011 .0149 .0263 .0285 .0257 .3232 .4489 .5653 .7033 .7326 .8654
Not removed Not removed Not removed Not removed Not removed Not removed Not removed Not removed Not removed 6 5 4 3 2 1
0.035 0.146 20.646 20.069 0.014 0.699 0.485 20.486 0.211
0.15 1.56 3.81 6.41 3.31 3.09 2.88 2.35 2.26
.8753 .1191 .0001 .0001 .0009 .0019 .0039 .0189 .0241
Parameter Intercept Constituent position (CP) Headedness Aphasia severity Headedness × Aphasia Severity Semantic transparency Compound frequency Age of acquisition CP × Headedness Stimulus length 1st constituent frequency 2nd constituent frequency Aphasia Severity × CP Aphasia Severity × CP × Headedness Imageability
Note: The central part of the table summarizes the model-simplification procedure, including the goodness-of-fit tests and their results; constituent position, headedness, and aphasia severity were not removed because part of higher order interactions. The rightmost part of the table reports the effects of the included variables, after having removed outliers.
32
Cognitive Neuropsychology, 2014, 31 (1 –2)
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
COMPOUND HEADEDNESS IN PICTURE NAMING
Figure 1. Interaction between the effects of compound headedness and aphasia severity (as measured by the performance on the Token Test).
fishdog, for pescespada), thus producing a different compound noun. Phonemic errors included omissions, transpositions, substitutions, staccato utterances (e.g., “au-to-stra-da” for autostrada, highway). Errors affecting the whole compound were categorized as whole-word omissions, whole-word substitutions, circumlocutions, and neologistic compounds. Whole-word omissions also included latencies (more than 2 s) and repairs. Whole-word substitutions included all cases in which the patient produced an existing simple word (e.g., “un quadro”, a painting, for capolavoro, masterpiece). Circumlocutions included those cases in which the patient described the meaning of the target word by means of a sentence (“serve per portare sopra la montagna i passeggeri”, it is used to take passengers over the mountain, for funivia, cablecar). Compound neologisms were responses in the form of a novel compound, whose meaning is usually related to the target (“tagliaprezzemolo”, parsley cutter, literally cuts-parsley, for mezzaluna, dicing knife).
Table 3 reports the number of errors for each of the above categories, divided on the basis of compound headedness and constituent position (when applicable). The table shows that errors affected the whole compound more frequently than the individual constituents (521 vs. 269). The most frequent error for head-initial compounds was omission, whereas head-final compounds were most frequently substituted with simple words, x2(3) ¼ 8.21, p ¼ .0418; compound neologisms were very rare. With regard to the individual constituents, modifier constituents were affected more frequently by lexical substitutions, whereas omissions were most often observed in the case of head constituents, x2(2) ¼ 25.66, p ¼ .0001. However, this error pattern partially changes when considering compound headedness: The effect is confirmed in head-initial compounds, x2(2) ¼ 21.63, p ¼ .0001, but not in head-final ones, as in this latter case omission errors were relatively rare and equally distributed between the two constituents, x2(2) ¼ 4.49, p ¼ .1062.2 Phonemic errors were generally uncommon.
DISCUSSION In this study, we analysed the naming performance on compound nouns in a group of Italian patients affected by aphasia. The purpose of the experiment was to investigate whether the head-modifier structure of compounds influences word retrieval, and it aimed specifically at replicating the effect described by Marelli et al. (2009) with a constituent priming task in a sample of neurologically unimpaired participants. Forty-five patients, selected for their disproportionate impairment in naming complex nouns with respect to simple nouns, participated in the study. The naming impairment observed in the aphasic participants had a greater effect on the retrieval of modifiers than of head constituents. In fact, modifier constituents were more difficult
2
With regard to the effects of the covariates on the probability of specific error types, the only association found was between aphasia severity and constituent omissions: Constituents were omitted more frequently by patients suffering from more severe impairments (p ¼ .0246). Cognitive Neuropsychology, 2014, 31 (1 –2)
33
MARELLI ET AL.
Table 3. Analysis of error types in compound naming Head-initial compounds Error type
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
Constituent omissions Lexical substitutions Phonemic errors
Head-final compounds
1st constituent
2nd constituent
1st constituent
2nd constituent
42 20 12
19 49 8
7 55 14
10 27 6
Whole-word omissions Whole-word substitutions Circumlocutions Neologistic compounds
120 83 37 8
101 122 40 10
Note: Number of errors (out of a total of 1350 observations), organized by headedness and constituent position. The upper section of the table summarizes constituent-specific errors, whereas the lower section reports errors affecting the whole compound.
to retrieve than their head counterparts. These results are in line with other studies investigating the headedness effect in compound processing (Jarema et al., 1999; Marelli et al., 2009) and indicate that the head-modifier structure is represented in the mental lexicon. In other words, constituents are represented on the basis of the role (modifier vs. head) they play in the compound structure: Head constituents, which usually specify the most important semantic and lexical traits of the whole compound, are arguably most salient from a cognitive point of view and are easier to retrieve than modifier constituents, which are in turn impacted more heavily by the patients’ naming impairment. This interpretation is supported by the error type analysis. In fact, modifiers were most often substituted with other words, suggesting that the head constituent is the most important during the conceptual combination procedure: Retrieval of the head triggers a lexical search for the other element, which is probably doomed to failure due to the patients’ naming impairment, resulting in a majority of substitutions compared to omissions of the modifiers. However, the larger impairment for modifiers was observed principally in head-final compounds. The difference in the performance vis a` vis
compound headedness is in line with the priming effects observed on normal subjects, as reported by Marelli et al. (2009). In fact, a comparison of the reaction time effects in word recognition by healthy participants with the naming accuracy by aphasic patients (Figure 2) shows that the two patterns of results are quite similar. Both priming facilitation and patients’ accuracy can be assumed to reflect the ease of access to constituent representations. Applying this line of reasoning, the results of both studies indicate that in head-final compounds the head is easier to access than the modifier, whereas in head-initial compounds both constituents are similarly easy to access. In line with the theory proposed by Williams (1981), the head-final structure seems to be preferred in the mental processing of compound words. The right-hand head rule would appear to be not only a theoretical principle, but also a general psycholinguistic property of the mental lexicon: The head-final structure is assumed to be the default organization of morphologically complex words. However, in accordance with Marelli et al. (2009) and Marelli and Luzzatti (2012), we propose that this assumption is not due to a prescriptive rule, but rather follows the distributional properties of the Italian morphology of compounds3; in fact, the majority of Italian
3 The prescriptive and the distributional interpretation of the right-hand head rule (Williams, 1981) are not necessarily at odds. Di Sciullo and Williams (1987) proposed that only head-final compounds would be real morphological constructs, whereas headinitial compounds are syntactic strings imported into the lexicon. Head-final compounds may thus be more numerous because they are generated through a standard morphological operation.
34
Cognitive Neuropsychology, 2014, 31 (1 –2)
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
COMPOUND HEADEDNESS IN PICTURE NAMING
Figure 2. Comparison between the results of the present study (left panel, percentage of correct constituent retrieval) and priming effects in lexical decision by healthy subjects (right panel, adapted from Marelli et al., 2009; permission granted from John Benjamins’ Publishing Company).
compounds adopts a head-final structure (Marelli & Luzzatti, 2012). In other words, it is to be expected that the representations of head-final compounds would be more structured as these compounds constitute the usual way in which two nominal constituents are combined in Italian (Marelli et al., 2009). This would be in line with constituents being more difficult to process in head-final compounds, as suggested by the interaction between headedness and aphasia severity: The complex and structured representation associated with head-final compounds makes constituent retrieval more difficult even for less impaired patients. The assumption of right-headedness also fits the pattern of omission errors in the error analysis: The head is often omitted in head-initial compounds, in line with the hypothesis that by default it is expected to be in the final position, whereas constituent omissions affect heads and modifiers equally (and are markedly less observed) in head-final compounds. The results of the present experiment are not a mere replica of the findings described in Marelli et al. (2009). In the first place, the present study employed a different set of compound stimuli to that used in the psycholinguistic experiment. This
difference strongly supports the reliability of the observed effect and rules out the possibility that the phenomenon may be due to an unfortunate selection of experimental stimuli. In the second place, the same effect was observed in two different tasks: The psycholinguistic study required a lexical decision to be made, thus tapping into the input stages of lexical processing; in the present experiment, patients were asked to name objects on the basis of pictures/definitions, thus having to rely on output procedures. This converging evidence provides an indication as to which level of representation generates the observed effect: In fact, the same level of processing has necessarily to be involved in both tasks. The most likely explanation is that information regarding a compound structure is stored at a central and amodal processing level; the conceptual/semantic level appears to be the best candidate to account for the headedness effect. In line with Marelli and Luzzatti (2012), the semantic system itself would be involved since the conceptual combination between constituents (Gagne´ & Spalding, 2009) takes place at this level, and the procedure has to be driven by previous knowledge about how constituents are combined (i.e., which constituent is the head and which the Cognitive Neuropsychology, 2014, 31 (1 –2)
35
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
MARELLI ET AL.
modifier). This interpretation is in line with results suggesting that the nature of headedness effects is primarily semantic (Marelli & Luzzatti, 2012) and with the assumption that headedness depends on the semantic association between a compound and its constituents (e.g., El Yagoubi et al., 2008). This hypothesis is also supported by the observed effect of semantic transparency: When they are part of more transparent compounds, constituents are correctly retrieved with a higher frequency, thus indicating that constituent retrieval is easier when the compound meaning can be obtained through a conceptual combination procedure. However, data have been reported that are in contrast with this “semantic hypothesis”, so underlining the importance of lexical factors in determining the head-modifier structure of a compound (e.g., Frisson, Niswander-Klement, & Pollatsek, 2008; Marelli et al., 2013). An alternative interpretation locates the source of the headedness effect at lemma level (Levelt et al., 1999), a module that stores grammatical and lexical information about words. This level is assumed to be central as well as amodal and to contain the same information (grammatical gender, grammatical class) as that shared by a compound and its head constituent. Therefore compound headedness may be represented in a lemma architecture that comprises both compound and constituent representations, as is the case of the multiple-lemma model proposed by Marelli et al. (2012). In this approach, the access to the constituent lemma nodes is subordinated to the activation of the compound node. Headedness can be straightforwardly formalized as the difference in strength of the connections between the compound and its constituents: The link between a compound and its head is assumed to be stronger than that between a compound and its modifier. As the present result suggests, in Italian the weight difference would be greater in head-final (hierarchically organized) than in head-initial compounds. A multiple-lemma architecture is also suggested by the significant effects of the age of acquisition and compound frequency variables, which are associated with the whole compound but emerged as significant predictors of the
36
Cognitive Neuropsychology, 2014, 31 (1 –2)
accuracy in producing the individual constituents. In other words, access to compound properties would be a prerequisite for the constituent retrieval, which clearly fits well the lemma network described by Marelli et al. (2012). Indeed, it is possible that the information on headedness is coded at both the lemma and the semantic levels as both lexical (i.e., pertaining to the lemma) and semantic aspects contribute to the headedness definition. The compound structure, specifying head and modifier positions, is arguably stored in the compound lemma, which is by principle independent of the meanings of the constituents. However, the actual head-modifier roles have to be represented in the semantic system, since their associated properties are used (together with relational information, see Gagne´ & Spalding, 2009) in the conceptual combination of constituent meanings. The results obtained from the present study do not provide a definitive answer regarding the representation of compound headedness, and further investigations are needed to clarify the issue. However, the study does provide crucial evidence to the headedness effect being originated at central processing levels, disconfirming the hypothesis that previously observed phenomena may be a by-product of the employed paradigm and/or be specific to written word processing. The approach adopted in the present study was fundamental for observing these results. In fact, headednessassociated effects have usually small effect size and are qualified by interactions with other variables (e.g., Marelli & Luzzatti, 2012), and group studies are often needed to obtain statistically significant effects. For these reasons, the error analysis in neuropsychological patients is important: This approach magnifies effects that are very limited in size (Semenza & Mondini, 2006), thus facilitating assessment of the relatively small influence of headedness and enabling testing of the predictions of psycholinguistic models. Manuscript received 27 March 2013 Revised manuscript received 26 July 2013 Revised manuscript accepted 24 October 2013 First published online 6 December 2013
COMPOUND HEADEDNESS IN PICTURE NAMING
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
REFERENCES Baayen, R. H. (2008). Analyzing linguistic data. A practical introduction to statistics using R. Cambridge: University Press. Badecker, W. (2001). Lexical composition and the production of compounds: Evidence from errors in naming. Language and Cognitive Processes, 16, 337– 366. Bien, H., Levelt, W., & Baayen, R. (2005). Frequency effects in compound production. PNAS, 102, 17876 –17881. De Mauro, T., Mancini, F., Vedovelli, M., & Voghera, M. (1993). Lessico di frequenza dell’italiano parlato. Milano: Etaslibri. Di Sciullo, M., & Williams, E. (1987). On the definition of word. Cambridge: MIT Press. Dunabeitia, J. A., Perea, M., & Carreiras, M. (2007). The role of the frequency of constituents in compound words: Evidence from Basque and Spanish. Psychonomic Bulletin & Review, 14(6), 1171 –1176. El Yagoubi, R., Chiarelli, V., Mondini, S., Perrone, G., Danieli, M., & Semenza, C. (2008). Neural correlates of Italian nominal compounds and potential impact of headedness effect: An ERP study. Cognitive Neuropsychology, 25, 559 –581. 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. Gagne´, C. L., & Spalding, T. L. (2007). Conceptual combination: Implications for the mental lexicon. In G. Libben & G. Jarema (Eds.), The representation and processing of compound words (pp. 145 –169). Oxford: University Press. Gagne´, 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. Goldman, R. (2007). Compounding in Aphasia: A crosslinguistic review (BA Thesis). Swarthmore College, Department of Linguistics, Swartmore, PA. Retrieved from http://hdl.handle.net/10066/10186 Guevara, E., & Scalise, S. (2008, February). Heads in compounding: Variation and patterns. 13th international morphology meeting – Book of Abstracts, Vienna, Austria, p. 43. Huber, W., Poeck, K., Weniger, D., & Willmes, K. (1983). Der Aachener Aphasie Test (AAT). Go¨ttingen: Hogrefe.
Hyo¨na¨, J., & Pollatsek, A. (1998). Reading Finnish compound words: Eye fixations are affected by component morphemes. Journal of Experimental Psychology: Human Perception and Performance, 24, 1612– 1627. Jaeger, T. F. (2008). Categorical data analysis: Away from ANOVAs (transformation or not) and towards logit mixed models. Journal of Memory and Language, 59, 434–446. Janssen, N., Bi, Y., & Caramazza, A. (2008). A tale of two frequencies: Determining the speed of lexical access for Mandarin Chinese and English compounds. Language and Cognitive Processes, 23, 1191– 1223. Jarema, G., Busson, C., Nikolova, R., Tsapkini, K., & Libben, G. (1999). Processing compounds: A cross-linguistic study. Brain and Language, 68, 362–369. Jarema, G., Perlak, D., & Semenza, C. (2009). The processing of compounds in bilingual aphasia. Aphasiology, 24, 126– 140. Kraemer, H. C., & Blasey, C. M. (2004). Centring in regression analyses: A strategy to prevent errors in statistical inference. International Journal of Methods in Psychiatric Research, 13, 141–151. Levelt, W. J., Roelofs, A., & Meyer, A. S. (1999). A theory of lexical access in speech production. The Behavioral and Brain Sciences, 22, 1–38; discussion 38 –75. Libben, G., Gibson, M., Yoon, Y. B., & Sandra, D. (2003). Compound fracture: The role of semantic transparency and morphological headedness. Brain and Language, 84, 50–64. Luzzatti, C., Willmes, K., & De Bleser, R. (1996). Aachener Aphasie Test: versione italiana (2nd ed.). Firenze: Organizzazioni Speciali. Lynott, D., & Ramscar, M. J. A. (2001, September). Can we model conceptual combination using distributional measures? Twelfth Irish Conference on Artificial Intelligence and Cognitive Science, Maynooth, Ireland, pp. 1– 10. Marelli, M., Aggujaro, S., Molteni, F., & Luzzatti, C. (2012). The multiple-lemma representation of Italian compound nouns: A single case study of deep dyslexia. Neuropsychologia, 50, 852–861. Marelli, M., Aggujaro, S., Molteni, F., & Luzzatti, C. (2013). Understanding the mental lexicon through neglect dyslexia: A study on compound noun reading. Neurocase, 19, 128–144. Marelli, M., Crepaldi, D., & Luzzatti, C. (2009). Head position and the mental representation of nominal Cognitive Neuropsychology, 2014, 31 (1 –2)
37
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
MARELLI ET AL.
compounds: A constituent priming study in Italian. The Mental Lexicon, 4, 430– 454. Marelli, M., & Luzzatti, C. (2012). Frequency effects in the processing of Italian nominal compounds: Modulation of headedness and semantic transparency. Journal of Memory and Language, 66, 644 –664. Mondini, S., Jarema, G., Luzzatti, C., Burani, C., & Semenza, C. (2002). Why is “Red Cross” different from “Yellow Cross”? A neuropsychological study of noun –adjective agreement within Italian compounds. Brain and Language, 81, 621 –634. Mondini, S., Luzzatti, C., Zonca, G., Pistarini, C., & Semenza, C. (2004). The mental representation of Verb-Noun compounds in Italian: Evidence from a multiple single-case study in aphasia. Brain and Language, 90, 470– 477. Semenza, C., Arcara, G., Facchini, S., Meneghello, F., Ferraro, M., Passarini, L., . . . Mondini, S. (2011). Reading compounds in neglect dyslexia: The headedness effect. Neuropychologia, 49, 3116–3120. Semenza, C., De Pellegrin, S., Battel, I., Garzon, M., Meneghello, F., & Chiarelli, V. (2011). Compounds
38
Cognitive Neuropsychology, 2014, 31 (1 –2)
in different aphasia categories: A study on picture naming. Journal of Clinical and Experimental Neuropsychology, 33, 1099–1107. Semenza, C., Luzzatti, C., & Carabelli, S. (1997). Morphological representation of compound nouns: A study on Italian aphasic patients. Journal of Neurolinguistics, 10, 33–43. Semenza, C., & Mondini, S. (2006). Neuropsychology of compound words. In G. Libben & G. Jarema (Eds.), The representation and processing of compound words (pp. 71 –95). Oxford: Oxford University Press. 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. Williams, E. (1981). On the notions “lexically related”, and “head of a word”. Linguistic Inquiry, 12, 245– 274. Zwitserlood, P. (1994). The role of semantic transparency in the processing and representation of Dutch compounds. Language and Cognitive Processes, 9, 341–368.
COMPOUND HEADEDNESS IN PICTURE NAMING
APPENDIX
Downloaded by [Archives & Bibliothèques de l'ULB] at 11:30 28 December 2014
Compound stimuli employed in the experimental phase of the experiment Headedness
Stimulus
Translations
Literal translation
Head-initial Head-initial Head-initial Head-initial Head-initial Head-initial Head-initial Head-initial Head-initial Head-initial Head-initial Head-initial Head-initial Head-initial Head-initial Head-initial Head-final Head-final Head-final Head-final Head-final Head-final Head-final Head-final Head-final Head-final Head-final Head-final Head-final Head-final
acquasanta arcobaleno busta paga cabina telefonica calzamaglia capolinea capotavola carro armato carro attrezzi carta carbone cassaforte macchina fotografica montepremi pesce spada pianoterra vagone ristorante aeroporto aliscafo autostrada capolavoro cronometro fotocopia funivia mezzaluna microscopio motoscafo pronto soccorso seggiovia supermercato terremoto
holy water rainbow payslip telephone booth tights end of the line head of the table tank tow truck carbon paper strongbox camera jackpot swordfish ground floor dining car airport hydroplane highway masterpiece stopwatch photocopy cable car dicing knife microscope motorboat first aid chairlift supermarket earthquake
water-holy arch-lightning envelope salary stateroom-telephonic sock-knit head-line head-table wagon-armed wagon-tools paper-coal box-strong machine-photographic mountain-prizes fish-sword floor-ground coach-restaurant air-port wings-hull car-street head-work time-meter photo-copy cables-way half– moon small-look movement-hull ready-rescue seat-way super-market lands-movement
Note: Constituents in italics are usually part of neoclassical compounds and are not free morphemes.
Cognitive Neuropsychology, 2014, 31 (1 –2)
39
