Individual behavioural patterns and neural underpinnings of verb processing in aphasia
Akinina, Yulia
DOI:
10.33612/diss.136488344
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below.
Document Version
Publisher's PDF, also known as Version of record
Publication date: 2020
Link to publication in University of Groningen/UMCG research database
Citation for published version (APA):
Akinina, Y. (2020). Individual behavioural patterns and neural underpinnings of verb processing in aphasia. https://doi.org/10.33612/diss.136488344
Copyright
Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.
CHAPTER 2
Russian Normative Data for 375
Action Pictures and Verbs
The research presented in Chapter 2 was published in the paper:
Akinina, Y., Malyutina, S., Ivanova, M., Iskra, E., Mannova, E., & Dragoy, O. (2015). Russian normative data for 375 action pictures and verbs. Behavior research methods, 47(3), 691–707. https://doi.org/10.3758/s13428-014-0492-9
The text of the paper has been updated to include more recent (up to 2018) literature on existing action pictures and verbs databases and edited for the purpose of terminological/formatting consistency.
ABSTRACT
The present article introduces a Russian-language database of 375 action pictures and associated verbs with normative data. The pictures were normed for name agreement, conceptual familiarity, and subjective visual complexity, and measures of age of acquisition, imageability, and image agreement were collected for the verbs. Values of objective visual complexity, as well as information about verb frequency, length, argument structure, instrumentality and name relation, are also provided. Correlations between these parameters are presented, along with a comparative analysis of the Russian name agreement norms and those collected in other languages. The full set of pictorial stimuli and the obtained norms may be freely downloaded from http://en.stim-database.ru for use in research and for clinical purposes.
2.1. INTRODUCTION
Since the appearance of the seminal normative studies of Lachman (1973) and Snodgrass and Vanderwart (1980), numerous normative databases of noun-object stimuli have been created with different psycholinguistic parameters. For nouns, databases have been developed for a variety of languages (for English: Barry, Morrison, & Ellis, 1997; Berman, Friedman, Hamberger, & Snodgrass, 1989; Himmanen, Gentles, & Sailor, 2003; Salmon, McMullen, & Filliter, 2010; Snodgrass & Vanderwart, 1980; for French: Alario & Ferrand, 1999; Bonin, Peereman, Malardier, Méot, & Chalard, 2003; for European Spanish: Cuetos, Ellis, & Alvarez, 1999; Sanfeliu & Fernandez, 1996; for Argentinian Spanish: Manoiloff, Arstein, Canavoso, Fernández, & Segui, 2010; for Italian: Dell’Aqua, Lotto, & Job, 2000; Nisi, Longoni, & Snodgrass, 2000; for German: Schröder , Gemballa, Ruppin, & Wartenburger, 2012; for Modern Greek: Dimitropoulou, Duñabeitia, Blitsas, & Carreiras, 2009; for Dutch: Martein, 1995; for Icelandic: Pind, Jónsdóttir, Tróggvadóttir, & Jónsson, 2000; for Russian: Tsaparina, Bonin, & Méot, 2011; for Japanese: Nishimoto, Miyawaki, Ueda, Une, & Takahashi, 2005; Nishimoto, Ueda, Miyawaki, Une, & Takahashi, 2012; for Persian: Ghasisin, Yadegari, Rahgozar, Nazari, & Rastegarianzade, 2015; for Gulf Arabic: Khwaileh, Mustafawi, Herbert, & Howard, 2018; for Tunisian Arabic: Boukadi, Zouaidi, Wilson, 2016; for Turkish: Raman, Raman, & Mertan, 2014; for Chinese: Weekes, Shu, Hao, Liu, & Tan, 2007; for English, German, Spanish, Italian, Bulgarian, Hungarian, and Mandarin Chinese: Brodeur, Dionne-Dostie, Montreuil, & Lepage, 2010; for British English, Spanish, French, Dutch, Italian and German: Duñabeitia et al., 2018), and for different age groups (for children: Berman, Friedman, Hamberger, & Snodgrass, 1989; Cycowicz, Friedman, Rothstein, & Snodgrass, 1997; for elderly people: Cuetos, Samartino, & Ellis, 2012; Ghasisin et al., 2015). Databases using the same visual stimuli, often taken (wholly or partially) from the set of black-and-white line drawings published by Snodgrass and Vanderwart (1980; e.g., Alario & Ferrand, 1999; Bates et al., 2003a; Manoiloff et al., 2010; Nisi et al., 2000; Pind et al., 2000; Sanfeliu & Fernandez, 1996; Weekes et a., 2007) or its colorized version (Rossion & Pourtois, 2004; e.g., Dimitropoulouet al., 2009; Raman et al., 2014; Tsaparina et al., 2011; Weekes et al., 2007), allow for direct cross-linguistic comparisons. Cross-linguistic studies are also carried out on the basis of databases including pictures that were originally developed or selected from different sources (Alario & Ferrand, 1999; Bates et al., 2003a; Kremin et al., 2003; Székely et al., 2004; Nishimoto et al., 2005). The databases of standardized norms provide values of different parameters (such as name and image agreement, picture visual complexity, conceptual familiarity, word age of acquisition, imageability, etc.) that can be used to balance materials in psycholinguistic experiments, can serve as independent variables in various experimental designs, or may be of interest by themselves, since they demonstrate the statistical interrelations among each other, shedding light on the mental lexicon’s structure and its functioning. Also, knowledge about the relevant psycholinguistic parameters of items is
crucial when language assessment and therapeutic tools are being developed for clinical purposes.
Following extensive normative data collection for object pictures and associated nouns in different languages and populations, an interest has recently emerged in the standardization of action pictures and verbs. This has largely been due to the growing recognition of the critical role of verbs in language disorders. Verb production is particularly vulnerable following brain damage that results in the breakdown of language, a condition known as aphasia (Bastiaanse, 1991). While being a common symptom of so-called agrammatic aphasia (Goodglass, 1993), verb deficits have also been reported for other aphasia types (Jonkers & Bastiaanse, 2007; Luzzatti, Aggujaro, & Crepaldi, 2006; Mätzig, Druks, Masterson, & Vigliocco, 2009, for a review). Careful matching for different psycholinguistic properties, such as word frequency, age of acquisition, and picture complexity, has not been able to completely eliminate a processing disadvantage for action naming as compared to object naming, in both healthy and brain-lesioned populations (Mätzig et al., 2009; Székely et al., 2005), which supports the idea about specific brain resources or greater resource recruitment allocated to verb processing. Lesions studies correlating dysfunction of particular brain regions with specific linguistic deficits have identified the frontal and, to a lesser degree, the parietal lobes of the left hemisphere as the main locus of verb production (see Cappa & Perani, 2003, for a review). However, growing evidence is suggesting that this view is too simplistic. Some individuals with verb impairments have lesions outside of the left frontal regions, and some with vast frontal lesions have verb production that is relatively spared (see Crepaldi, Berlingeri, Paulesu, & Luzatti, 2011, for a review). Also, some neurodegenerative diseases – for example, different types of frontotemporal dementia – also show impaired action naming relative to object naming (Cotelli et al., 2006). All the same, some patient groups may not show this pattern (e.g., with semantic dementia, see Cotelli et al., 2006). Furthermore, the verb-noun discrepancy effect may be eliminated in some patients when certain aspects of stimuli are controlled. For example, D’Honincthun and Pillon (2008) discovered that action naming and verb comprehension deficits were no longer found in one patient with a frontal variant of frontotemporal dementia when naming and comprehension were assessed using videotaped actions or verbal stimuli, rather than depicted actions (i.e., photographs). The pervasiveness of verb deficits in patients with different neurological syndromes and brain lesion topography necessitates the establishment of a reliable research apparatus, as well as accurate clinical tools for verb assessment, which can take into consideration normative information about verbs and associated pictorial stimuli.
A number of parameters have been identified that may contribute to lexical processing, and to verb production in particular (e.g., in an action-naming task), and must therefore be controlled for in aphasia assessment batteries and carefully manipulated in experimental aphasia studies. The parameters characterize a concrete stimulus – a word or a corresponding picture. Some of these parameters are obtained through experiments
in which participants are asked to perform a task involving the stimulus – for instance, to name a picture in order to obtain name agreement scores. Some of the parameters are subjective and are obtained in questionnaires in which participants are asked to rate the stimulus on a particular scale. For instance, image agreement, imageability, visual complexity, familiarity, and age of acquisition are rated parameters, and the numbers of points on scales vary among studies. Some parameters can be taken from specialized databases (e.g., frequency), and some can be directly observed – for example, word length. Finally, specific word-class parameters are determined by experts in linguistics (for verbs, parameters include argument structure, instrumentality, and name relation to a noun). All of the parameters mentioned above will be defined and discussed below.
Name agreement is a measure of the uniformity of names used by different participants to refer to a depicted object or action. Two measures for name agreement are commonly used. The percentage of name agreement (% name agreement) is the number of participants (per hundred) who gave the most frequent response in a picture-naming task. Another measure, the H statistic, is computed by the following formula:
where k is the number of different names given to each picture, and pi is the proportion
of respondents who gave each name (Snodgrass & Vanderwart, 1980). The H statistic has been widely used in naming studies since Snodgrass and Vanderwart’s, because this measure provides information about the distribution of responses across participants. Whereas % name agreement reflects the homogeneity of naming responses, the H statistic pertains to the heterogeneity of naming behavior of participants. Name agreement has been associated with robust effects in naming. Vitkovitch and Tyrrell (1995) found that objects with higher name agreement are recognized in an object decision task as quickly as those with multiple correct names, but more quickly than objects that are often assigned erroneous names. In a naming task, objects with high name agreement were named more quickly than either of two sets with low name agreement scores. This suggests that name agreement affects the lexical access that takes place after structural recognition. Naming latencies in object and action naming also increase with decreased name agreement scores (Alario et al., 2004; Barry et al., 1997; Bonin, Boyer, Méot, Fayol, & Droit, 2004; Bonin et al., 2003; Cuetos & Alija, 2003; Ellis & Morrison, 1998; Nishimoto et al., 2012; Weekes et al., 2007). In addition, in people with aphasia (PWA), name agreement is a strong determinant of action-naming accuracy (Kemmerer & Tranel, 2000).
Image agreement is the degree to which a visually depicted object or action corresponds to the mental image generated in response to the presentation of a word. High image agreement scores tend to contribute to shorter object- and action-naming times in healthy populations (Barry et al., 1997; Bonin et al., 2004; Bonin et al., 2003; Nishimoto et al., 2012)
and to better action-naming performance in individuals with brain damage (Kemmerer & Tranel, 2000). It is generally assumed that higher image agreement scores pertain to the presence of canonical mental images (Barry et al., 1997). Conversely, at the level of object recognition, lower image agreement (i.e., an image being less similar to one’s mental representation) leads to slower recognition (e.g., Alario et al., 2004; Barry et al., 1997).
Imageability refers to the degree of effort with which a mental image of a corresponding object or action can be generated in response to a verbal stimulus. It is thought to be related to the richness of semantic representation of a word (Breedin et al., 1994; Plaut & Shallice, 1993; Rofes et al., 2018), which can be defined as the number of semantic features that are consistently accessed (Plaut & Shallice, 1993). The imageability effect, which manifests in reduced naming times for more-imageable words, has been replicated in different studies using noun stimuli (Alario et al., 2004; Ellis & Morrison, 1998; Nickels & Howard, 1995; but see Nishimoto et al., 2012). However, for verbs, the effect is inconsistent, and hence the role of imageability is unclear. Shao, Roelofs, and Meyer (2014) found that imageability was a significant predictor of action naming latencies; Cuetos and Alija (2003) reported a similar effect that approached significance; and Bonin, Boyer, Méot, Fayol, and Droit (2004) found no reliable effect. Several studies have made a direct comparison of the imageability parameter in verbs and nouns. Chiarello, Shears, and Lund (1999) reported stronger correspondence between imageability ratings and reaction times for nouns than for verbs, which implies an easier mental image generation for nouns or a lesser importance of this parameter for verb processing. In a study by Masterson and Druks (1998), verbs obtained significantly lower imageability ratings than did nouns. At the same time, some studies (e.g., Bird, Howard, & Franklin, 2003) showed that when imageability ratings were controlled for, the verb-noun dissociation disappeared. This allowed researchers to draw the conclusion that at least in some cases verb-noun discrepancies could be reduced to imageability effects, although this view was criticized (for a review, see Druks, 2002). Connel and Lynott (2012) showed that while rating imageability, the participants tended to rely on the visual modality (for nouns), which, in the authors’ opinion, was prompted by the formulation of the task “generate a mental image”. Modalities other than the visual modality might contribute more to the mental representations of verbs (e.g., they may be associated with motor imagery). The described inconsistent effects and considerations regarding imageability warrant further investigation of its role in verb processing and the inclusion of this parameter in verb norming studies. Overall, this variable should still be controlled in clinical research.
Visual complexity may refer either to subjective ratings of the amount of detail in a picture or to objective characteristics of the digitized image – for instance, the size of the file in different formats (Székely & Bates, 2000). Visual complexity is thought to affect recognition times, and therefore may then affect naming speed. Several studies have shown subjective visual complexity to be a predictor of object-naming latencies in healthy individuals (Alario et al., 2004; Ellis & Morrison, 1998), and accuracy in people with aphasia (PWA; e.g., Cuetos,
Aguado, Izura, & Ellis, 2002), although in many studies no reliable effect of this parameter on object-naming latencies has been found (e.g., Barry et al., 1997; Bonin et al., 2003; Cuetos, Ellis, & Alvarez, 1999; Snodgrass, & Yuditsky, 1996; Weekes et al., 2007). One of the problems with this measure is related to the fact that subjective visual complexity ratings rely on behavioral performance, and may be confounded by subjective familiarity, word frequency, and age of acquisition. To resolve this issue, Székely and Bates (2000) analyzed several objective measures of digitized visual stimuli, and found that PDF, TIFF and JPG formats provided valid values of objective visual complexity that strongly correlated with subjective ratings and – unlike the subjective ratings – affected naming accuracy, although with no effect on naming times in their study.
Familiarity is a conceptual parameter pertaining to the estimated degree to which the depicted object or action is familiar to participants – that is, how often they think about or deal with the object or the action. During the conceptual familiarity task, participants can be asked to make their ratings on the basis of the presented word (e.g., Cuetos & Alija, 2003; Ellis & Morrison, 1998), as well as of the picture (e.g., Bonin et al., 2004; Fiez & Tranel, 1997; Snodgrass & Vanderwart 1980); in the latter case, the instruction is to estimate the concept, and not the depicted version of it. A familiarity effect has been reported in some studies: the concepts for more familiar objects are processed faster (Cuetos et al., 1999; Snodgrass, & Yuditsky, 1996), although in other studies a small or no effect has been found (see Alario et al., 2004, for a comparative analysis of eight naming studies in six languages). Familiarity has been shown to affect object naming performance in PWA (Cuetos et al., 2002). In Kemmerer and Tranel (2000), familiarity was shown to be a reliable determinant of verb production accuracy in an action-naming task in a large group of participants with brain damage. Taking into account the discrepancies in the measurements and the inconsistency of the results, the role of familiarity in lexical retrieval during picture naming remains unclear (Schröder et al., 2012). The term familiarity can also refer to subjective frequency – that is, the degree to which the printed word is familiar to the participant (e.g. Bird, Franklin & Howard, 2001; Shao et al., 2014).
Age of acquisition (AoA) has proven to be a very important parameter influencing lexical processing. Subjective (or rated) age-of-acquisition measures are collected in normative studies by asking adult participants to estimate the approximate age at which they believe they learned the word. Objective age-of-acquisition measures are based on actually testing the ability of children of different ages to perform different lexical tasks. This is a more rare measure that is usually highly correlated with subjective age-of-acquisition ratings (Álvarez & Cuetos, 2007; Grigoriev & Oshhepkov, 2013; Johnston & Barry, 2006). A large number of studies reported age-of-acquisition effects in different languages, populations, and experimental tasks (for a comprehensive review, see Johnston & Barry, 2006). The general pattern is that words that are learned earlier in life are processed faster and more accurately. Although some criticism has questioned the independence and validity of this measure (Bonin, Méot, Mermillod, Ferrand, & Barry, 2009; Zevin & Seidenberg,
2004), collecting age-of-acquisition ratings is still a gold standard in both noun and verb normative studies.
Frequency usually refers to the number of occurrences of a word in a large corpus of written texts, although the word frequencies in spoken language may also be used, when available (e.g., in the CELEX database; Baayen, Piepenbrock, & Van Rijn, 1993). It can be defined as frequency of a lexeme (a particular word form) or a lemma (all forms of a word). The impact of frequency on object-naming latencies is robust and has been shown in numerous studies for various languages (Alario et al., 2004; Barry et al., 1997; Cuetos et al., 1999; Ellis & Morrison, 1998): the more frequent the noun, the less time is needed to name the corresponding picture. However, the opposite pattern can be observed in an action naming task: Székely and colleagues (2005) found that high frequency was a predictor of shorter naming times for objects, but for action names the higher-frequency items took longer to produce. Székely and colleagues explain this paradoxical result by the fact that participants use high-frequency “light verbs” (i.e. less specific, general purpose verbs) for pictures that are difficult for them. Other authors report no significant effect of frequency on action naming (Cuetos & Alija, 2003; Bonin et al., 2004; Schwitter et al., 2004; Shao, Roelofs, & Meyer, 2014). Written frequency measures are usually strongly correlated with age-of-acquisition ratings, and it is controversial whether these two variables should be considered independently. Although some authors have shown that the age-of-acquisition effect on picture-naming speed interacts with frequency (Barry et al., 1997), others have reported clear independent and determinant roles of both frequency and age of acquisition (e.g., Alario et al., 2004). In clinical populations, both the frequency effect (Cuetos et al, 2002; Nickels & Howard, 1995) and a reversed frequency effect (i.e., greater difficulties with the processing of more frequent words) have been observed in different tasks (Hoffman, Jeffries, & Lambon Ralph, 2011; Marshall, Pring, Chiat, & Robson, 2001). Some studies, however, fail to replicate frequency effects in aphasic retrieval of verbs when other parameters are controlled for (Bastiaanse, Wieling, & Wolthuis, 2016).
Length is a phonological factor that can be calculated as the number of syllables or the number of phonemes in a word. The common assumption is that during word production different segments of the word are sequentially encoded in the word frame, which leads to the prediction that longer words take more time to be encoded than shorter words (e.g., Alario et al., 2004). However, the experimental evidence for length effects is inconsistent. Some authors have reported no length effect at all, whereas other findings have shown an effect of length. For instance, Cuetos et al. (1999) found an independent syllabic length effect on object-naming latencies in Spanish: reaction times were longer for longer words (a positive length effect). However, Alario et al. (2004) reported a marginal negative syllable-length effect for French: longer nouns were produced more quickly than shorter nouns. The authors claimed that certain specific conditions should be met for the effect to be seen. For instance, Meyer, Roelofs, and Levelt (2003) revealed a syllable-length effect on production speed in a blocked presentation design, but not in a mixed design. Thus,
another explanation is that the appearance of the length effect may depend on speakers’ response strategies, which may be revealed in some experimental designs but not in others (Damian, Bowers, Stadthagen-Gonzalez, & Spalek, 2010; Meyer et al., 2003). In aphasia, a noun-length effect was also observed in patients with different diagnoses, including nonfluent, fluent, and apraxic indivuduals (Nickels & Howard, 1995).
Argument structure, instrumentality, and name relation are parameters that are specific to verbs and are known to influence verb processing. Argument structure refers to the number of obligatory and optional participant roles required for a given verb (e.g., transitive verbs such as to beat require two arguments [“Peter is beating John”], whereas intransitive verbs such as to run require only one argument:[“Peter is running”]) and can be further defined in terms of qualitative categories. Studies that have focused on agrammatic aphasia in different languages have reported significant differences between verb groups, the major tendency being toward increasing verb production difficulties with the increasing number of arguments (Kim & Tompson, 2000, for English; Kiss, 2000, for Hungarian; Luzzatti et al., 2002, for Italian; Dragoy & Bastiaanse, 2010, for Russian; De Bleser & Kauschke, 2003, for German; see also Druks, 2002, for a review). This was later refined to argument complexity: the more complex the argument structure of a verb is, the harder it is to retrieve the verb (Bastiaanse & Van Zonneveld, 2005; Thompson, 2003). Instrumentality is a conceptual factor that refers to an action’s conceptual representation containing an obligatory instrument/tool that is not a body part (e.g., to cut, to draw, in contrast to noninstrumental verbs such as to tear, to push). Name relation to a noun is a lexical-phonological parameter referring to the phonological similarity between an instrumental verb and its associated tool noun: for example, to saw is a name-related verb (to saw and a saw are homophonous). Jonkers and Bastiaanse (2007) found positive effects of instrumentality on an action-naming task in a group of anomic aphasic speakers, and Malyutina, Iskra, Sevan, and Dragoy (2014) extended the findings for people with fluent and nonfluent aphasia and for elderly healthy individuals; instrumental verbs were better preserved than noninstrumental verbs. The results for name relation have been less consistent. Jonkers and Bastiaanse (2007) found a positive effect of name relation in a group of individuals with anomic aphasia, while Malyutina and colleagues (2014) demonstrated that verbs with a name relation to their associated noun were retrieved more poorly in non-fluent aphasic speakers than those which were non-name-related.
Databases with normative data about verbs and action pictures have increased in number in these recent years. General information about existing verb and action databases, their material and parameters, is presented in Table 2.1. The aim of Fiez and Tranel (1997) was to develop a set of naming and recognition tests for the evaluation of lexical and conceptual processing of actions. For that purpose, norms for 280 English verbs and colored photographs of the corresponding actions were collected. The focus of the Masterson and Druks (1998) was on making a list of English nouns and verbs and creating visual stimuli for them (164 objects and 102 actions) matched on a number of parameters,
which would allow for the comparison of participants’ performance in experimental, assessment, and treatment tasks. Chiarello, Shears, and Lund (1999) examined dissociations between English verbs (N = 427), nouns (N = 555), and words of balanced verb-noun usage (N = 215), collecting, among several corpus-based statistical measures, imageability ratings for the words. The purpose of Bird, Franklin, and Howard (2001) was to collect imageability and age-of-acquisition ratings for a large set of words (N = 2,645), with 892 verbs and 213 function words among them, and in particular, to explore the relationship between the word’s class and the rated values. The interrelation of age of acquisition and several other parameters was also calculated. Cuetos and Alija (2003) offered a normative database for 100 Spanish verbs, using the visual material from Druks and Masterson (2000). Székely et al. (2004) performed the largest cross-linguistic project, which included databases of normative values for stimulus sets in seven languages (American English, German, Mexican Spanish, Italian, Bulgarian, Hungarian, and the Taiwan variant of Mandarin Chinese), the action and verb processing part being represented by 275 items. A comprehensive normative database for French action photographs can be taken from Fiez and Tranel (1997), and the corresponding verbs were presented in Bonin, Boyer, Méot, Fayol, and Droit (2004). Schwitter et al. (2004) also provided psycholinguistic norms for French, this time for drawings and not photographs, some of which (71 items) were taken from Druks and Masterson (2000), and the rest (41 items) were originally developed. A comparison of two data sets for one language, based on photographs and black-and-white drawings, showed that when photographs were used, fewer items obtained % name agreement scores higher than or equal to 80%, which allowed Schwitter et al. (2004) to recommend the use of drawings for research and clinical practice. Cameirão and Vicente (2010) described the collection of age-of-acquisition norms for 1,749 Portuguese words of different word classes, including 373 verbs, and they presented a database with these norms and several other psycholinguistic parameters. The work of Shao, Roelofs, and Meyer (2014) introduced a normative database of 124 Dutch verbs and action pictures: 100 items from Druks and Masterson (2000), and 24 items from Konopka and Meyer (2012). Alonso, Díez, & Fernandez (2016) collected subjective age of acquisition for 4640 infinitive and reflexive verb forms in Spanish. Imbir (2016) focused on affective norms – parameters measuring emotional aspects of word processing – for 4905 Polish words, including 1126 verbs. Values of several affective parameters were collected along with concreteness, imageability, and subjective age of acquisition; values of frequency and length were also provided. Bayram, Aydin, Ergenc, & Akbostanci (2017) presented a stimuli database in Turkish, where verbs (N = 160) and nouns (N = 160) were specifically controlled for action/ non-action semantics. Birchenough, Davies, & Connelly (2017) collected normative values of subjective age of acquisition for 3259 German words (N of verbs = 473). Soares, Costa, Machado, Comesaña, & Oliveira (2017) assembled a dataset for European Portuguese (N = 3800, N of verbs and verb forms = 826) which contained normative values of imageability, concreteness, and subjective frequency, along with several measures of length, objective
frequency and orthographic neighborhood estimates. Bonin, Méot, & Bugaiska (2018), for French, investigated the relationship between concreteness and previously collected normative values in 1659 words (N of verbs = 157). Finally, Khwaileh, Mustafawi, Herbert, & Howard (2018) presented an extensive noun (N = 319) and verb (N = 141) stimuli database for Gulf Arabic. This list is by no means comprehensive: there are many other excellent psycholinguistic databases that include verbs among other words. Here, for reasons of space, we only describe studies where verbs are explicitly indicated within the database or the study contains a part-of-speech analysis of any kind.
Until recently, no normative database with psycholinguistic parameters has existed for the Russian language, despite high demand. Russian is a major East Slavic language belonging to the Indo-European language family, and is the official language of the Russian Federation. Russian is also spoken in former USSR countries (Armenia, Azerbaijan, Belarus, Estonia, Georgia, Kazakhstan, Kyrgyzstan, Latvia, Lithuania, Moldova, Tajikistan, Turkmenistan, Ukraine, and Uzbekistan) and by emigrant communities in different parts of the world (Bulgaria, Canada, China, Croatia, Czech Republic, Finland, Germany, Greece, Israel, Mongolia, Mozambique, Norway, Paraguay, Poland, Romania, Serbia, Slovakia, Sweden, United States, and Uruguay). Lewis, Simons, and Fennig (2013) estimated the total number of Russian speakers in the world to be 161,727,650, with additional 110 million people speaking it as a second language. Russian has a rich derivational and inflectional morphology, uses the Cyrillic alphabet for writing, and features many other interesting characteristics that make it a valuable target for advanced (psycho)linguistic research.
With the motivation to provide normative data for Russian, and thus to make it possible to control materials in Russian experimental research, several studies have been recently published. A study by Grigoriev, Baljasnikova, and Oshhepkov (2009) presented imageability data for 470 Russian nouns. In a subsequent study (Grigoriev, Oshhepkov, Baljasnikova, & Orlova, 2010), data on name agreement (% name agreement and the H statistic), familiarity, and picture-name agreement were collected for 286 object pictures taken from Snodgrass and Vanderwart (1980), 90 of which were modified, and for eight additional pictures taken from Morrison, Chappell, and Ellis (1997). Finally, in a recently published article, Grigoriev and Oshhepkov (2013) reported objective age-of-acquisition measures for the 286 pictures from Grigoriev et al. (2010). In addition, another normative study for Russian was performed by a different research group (Tsaparina et al., 2011), which continued the collection of norms for object pictures from Snodgrass and Vanderwart (1980) and provided data on 260 Russian nouns and colorized images created by Rossion and Pourtois (2004), based on the drawing from Snodgrass and Vanderwart (1980). The parameters included in the study were name agreement (% name agreement and the H statistic), image agreement, conceptual familiarity, imageability, age of acquisition, objective word frequency, and objective visual complexity. Grigoriev and Oshhepkov (2013) did not perform a direct comparison between their normative data and those reported by Tsaparina et al. (2011), except for the high correlation between objective age of acquisition obtained in their study and the rated age of acquisition from Tsaparina et al. (2011).
Table 2.1. Revi ew o f n orm ative stu di es an d d atabases o f verbs an d acti on pi ctur es . Fi ez & Tr an el (1 99 7) M as te rs on & D ru ks (1 99 8) Ch ia re llo , Sh ea rs & L un d (1 999 ) Bi rd , F ra nk lin & H ow ar d (2 00 1) Cu et os & A lij a (2 00 3) Sz ékel y et al . (2 00 4) Bon in , B oy er , Mé ot , F ay ol , & D ro it (2 00 4) Sc hw itt er e t al . (2 00 4) Ca m eir ão & Vi cen te (2 01 0) Sh ao , Ro elo fs, & Me ye r ( 20 14 ) La ng ua ge (s) Eng lis h Eng lis h Eng lis h Eng lis h Sp an ish En gl ish , Hu ng ar ia n, Sp an ish , I ta lia n, Bu lg ar ia n, a nd Ch in es e Fr en ch Fr en ch Po rtu gu es e D ut ch V isu al m at er ial s Co lo re d ph ot og ra ph s Bl ac k-an d-wh ite dr aw in gs -M as te rs on & D ru ks (2 00 0) Bl ac k-an d-wh ite dr aw in gs Fi ez & T ra ne l (1 99 7) M as te rs on & D ru ks 2 00 0 (7 1 i te m s), or ig ina l ad di tion al dr aw in gs ( 41 item ) -M as te rs on & D ru ks 2 00 0 ( 10 0 item s), Kon op ka & M ey er 2 01 2 ( 24 item s) N um be r o f i te m s 28 0 10 2 427 89 2 10 0 275 14 2 11 2 37 3 12 4 NA ✓ ✓ ✓ ✓ ✓ ✓ ✓ IA ✓ ✓ ✓ ✓ ✓ Fa m ✓ ✓ ✓ ✓ ✓ ✓ SVC ✓ ✓ ✓ ✓ ✓ ✓ ✓ Fr eq ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ AoA ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Ima g ✓ ✓ ✓ ✓ ✓ ✓ Le ng th ✓ ✓ ✓ ✓ ✓ ✓ N am in g La ten ci es ✓ ✓ ✓ ✓ O th er pa ra m et er s Se m an tic cat eg or ie s fo r ve rb s a nd nou ns D ist ribu tion al ty pi cal ity Conc re te ne ss O bj ec tive A oA , ob jec tive v isu al com pl ex ity D ur at ion o f de pi ct ed a ct ion s Im ag e va ria bli lit y Conc re tn es s ne ig hb or ho od de ns ity , n um be r of or tho gr ap hi c neig hb or s, nu m be r o f ph on ol og ic al neig hb or s
Ta bl e 2 .1 (c on tin ue d) Al on so , D íez , & Fe rn an de z (2 01 6) Im bi r (2 01 6) Ba yr am , A yd in , Er ge nc , & Ak bo sta nc i ( 20 17) Bi rc he no ug h, Da vi es , & Co nn ell y ( 20 17) So ar es , C os ta , M ac had o, Co m es añ a, & O liv eir a (2 01 7) Bo ni n, M éo t, & Bu ga isk a, (2 01 8) Kh wa ileh, Mu sta fa wi , H er be rt , & H ow ar d, (2 01 8) La ng ua ge (s) Sp an ish Po lis h Tu rk ish Ge rm an Po rt ug ue se Fr en ch Gu lf A ra bic Vi su al m at er ial s Bl ac k- and-wh ite dr aw in gs Bl ac k- and-wh ite dr aw in gs N um be r of it em s 46 40 11 26 16 0 47 3 82 6 15 7 14 1 N A ✓ ✓ IA ✓ Fa m ✓ SVC ✓ ✓ Fr eq ✓ ✓ ✓ ✓ Ao A ✓ ✓ ✓ ✓ ✓ Ima g ✓ ✓ ✓ ✓ Len gt h ✓ ✓ ✓ ✓ N am in g L at en cie s ✓ O th er p ar am et er s aff ec tive pa ra m et er s (v al enc e, a rou sa l, dom in an ce , or ig in , sig ni fic an ce) , conc re te ne ss ac tion /n on -ac tio n s em an tic s, m or ph em e c ou nt conc re te ne ss , sy lla bi c s tr uc tu re , or tho gr ap hi c ne ig hb or ho od siz e, or tho gr ap hi c Le ve ns hte in dis ta nce conc re te ne ss , co nt ext av ail ab ili ty , aff ec tive pa ra m et er s (v al enc e, a rou sa l) wor d f am ili ar ity No te : “ ✓ ” s ta nd s f or p ar am et er s n or m ed in t he s tu dy o r r ep or te d t he re o n t he b as is o f o th er s ou rc es . N A = n am e ag re em en t ( di ff er en t m ea su re s) ; I A = i m ag e a gr ee m en t; F am = c on ce pt ua l f am ili ar it y; S CV = s ub je ct iv e v is ua l co m pl ex it y; F re q = f re qu en cy ( di ff er en t m ea su re s) ; A oA = a ge o f a cq ui si ti on ; I m ag = i m ag ea bi lit y; L en gt h = wo rd len gt h ( di ff er en t m ea su re s) .
Despite the emerging interest in normative data collection for nouns and object pictures in Russian, to our knowledge there no available databases include Russian verbs and the corresponding action pictures, in addition to reporting reliable normative data collected from a considerable number of language speakers and taking into account other linguistic characteristics that are critical for verb processing. A study of Pashneva (2013) presents name agreement norms and data on imageability, concept familiarity, and age of acquisition for 275 action pictures retrieved form the database of Bates et al. (2003a). As the first stage, a picture study was performed in which the dominant names for action pictures were obtained. In the second stage, the normative data on imageability, concept familiarity, and age of acquisition were collected for the dominant names obtained in the first stage. Although this was the first attempt to create a verb and action stimuli database for Russian, the reliability of the results is disputable, due to the small number of participants in the first stage of the study (N = 34) and because the overall obtained name agreement scores were rather low (mean = 55.18, SD = 23.37 for % name agreement; mean = 1.89, SD = 0.92 for H statistic).
Our aim was to create a comprehensive verb-action picture normative database that would contain stimuli and the values of relevant (psycho)linguistic parameters that affect lexical processing. Given the lack of similar resources for Russian, our database will be useful not only for research purposes, but also for the development of speech-related diagnostic and therapeutic tools for Russian-speaking populations all over the world. The parameters included in the database are name agreement (% name agreement and H statistics), objective and subjective visual complexity, image agreement, imageability, conceptual familiarity, age of acquisition, verb lemma frequency, number of arguments, length in syllables (third-person and infinitive forms), instrumentality, and name relatedness.
2.2. METHOD
2.2.1. Stimuli
The decision to create original visual stimuli was made because the number of depictable verbs that we wished to include in the study exceeded the number of items in the existing picture sets. The initial data set consisted of 414 depictable verbs and corresponding pictures of actions. For pictures we used 414 black-and-white drawings of actions specifically created by an artist for this project (the entire set is available online at http:// en.stim-database.ru ).
Figure 2.1. The example of the visual stimulus for the verb doit’: ‘to milk’
2.2.2. Procedure
Objective verb and action picture parameters were obtained from available dictionaries or on the basis of the independent judgment of at least two professional linguists. If the judgments of the two experts differed, the case was discussed with a third expert, and a decision was made. Verb argument structure descriptions were taken from the dictionary (Ozhegov & Shvedova, 1992; available at http://dic.academic.ru/contents.nsf/ogegova/), with the number of arguments being defined as the number of obligatory arguments mentioned in the definition of the verb. Then the depiction of the verb was checked: if the
argument was not present in the picture, it was not counted (e.g., the verb fotografirovat’1:
‘to take photos’ may have a direct object, but it was not present in the picture, so the item was assigned “1” as the number of arguments, which referred to the participant present in the picture). If the argument of the verb could not be depicted because it is not a visible material object (e.g., pet’: ‘to sing’), the argument structure was assigned the value of “1/2”. It should be highlighted that the argument structure presented in the database is based on the concrete action depictions; hence, it is not recommended to use it as canonical argument structure for experiments that do not include those particular visual stimuli. Verb lemma frequency was extracted from Lyashevskaya and Sharov (2009), a Russian frequency dictionary based on the Russian National Corpus of written and spoken texts of various genres (the total size being 92 million words) available at http://dict.ruslang. 1 All the examples are written in transliteration and accompanied with English translation.
ru/freq.php. The values were also transformed to avoid data skewness using the formula log(1+x), where x is the verb lemma frequency (Barry et al., 1997; Cuetos & Alija, 2003). The values of instrumentality and name relation were assigned by professional linguists. Since verb morphology is different from noun morphology in Russian, and full verb-noun homophony is not possible, the notion of name relation was broadened, so that a verb was considered name-related if it has the same root as the noun referring to the instrument/tool, but not if the verb and noun were completely homophonous, as in the case of English (e.g., a saw – to saw). Length was defined as the number of syllables in the form of the present tense third-person singular/plural, depending on the number of actors in the picture, and also as the number of syllables in the infinitive form. The objective visual complexity was defined as the file size in JPG format, in kilobytes, following the suggestion of Tsaparina et al. (2011).
The normative data for the pictures (name agreement, action familiarity, and subjective visual complexity) and for verbs (age of acquisition, imageability, and image agreement) were collected via the online survey platform http://virtualexs.ru/. The verbs were divided into six subsets, of 58 to 81 items each (mean = 74.7). Two lists were created for each subset, resulting in 12 lists overall. One list for each verb subset contained picture-based tasks – naming, familiarity, and subjective complexity judgment, whereas the second list contained word-based tasks – age of acquisition, imageability judgment, and image agreement.
In the picture-based lists, for the collection of name agreement scores, the pictures (for an example, see Figure 2.1) were displayed on a white background, and the participants were asked for a one-word answer to the question “What is the character (characters) doing in the picture?” This task resulted in the production of verbs in the third-person singular form, which has been proven to be more natural for Russian than the production of the infinitive form in an action-naming task (Akinina & Dragoy, 2012; Kozintseva et al., 2013). The participants were explicitly instructed not to give multiword expressions as answers. In the action familiarity task, the participants were asked to estimate how familiar the depicted action was in terms of how often they performed the action, observed others performing it, or thought about it, using the five-point scale from 1 = barely familiar to 5 = very familiar2. The subjective visual complexity scores were collected by asking participants to evaluate the complexity of the picture, but not of the action itself, on the basis of the numbers of lines and details present in the picture, using a five-point scale from 1 = simple picture to 5 = complex picture.
2 The 5-point scales used in the study were adapted from (Snodgrass & Vanderwart, 1980) and are widely used in noun and object norming studies, although some verb and action norming studies are based on 7-point scales (e.g. Bayram et al., 2017; Bird et al., 2001; Chiarello et al., 1999; Cuetos & Alija, 2003; Mfastrson & Druks, 1998; Schwitter et al., 2004; Shao et al., 2014; Soares et al., 2017). The normative data for Russian nouns and objects are collected in 5-point scales to allow for a comparison of the current data on verbs and actions with future data on nouns and objects.
The norms in word-based lists were obtained on the basis of verbal modality. The verb in the infinitive form (e.g., doit’: ‘to milk’) was displayed on the screen. In the age-of-acquisition task, the participants were instructed to estimate the approximate age at which, in their opinion, they had learned the verb, on a five-point scale from 1 = 0-3 years to 5 = 12 years and later, with each point representing an interval of 3 years. The instructions for the imageability task were formulated as follows: the participants were asked to indicate how easy it was for them to imagine the action depicted in the picture, using a five-point scale from 1 = easy to imagine to 5 = hard to imagine (the modality of images evoked by the verb – visual, auditory, tactile, etc. – was not specified). The imageability scale in this study is reversed compared to other verb norming studies (Bayram et al., 2017; Bird et al., 2001; Bonin et al., 2004; Chiarello, Shears & Lund, 1999; Cuetos & Alija, 2003; Imbir, 2016; Khwaileh et al., 2018; Mastrson & Druks, 1998; Shao et al., 2014; Soares et al., 2017;), where 1 is usually used to designate the least-imageable verbs, and the opposite end of the scale corresponds to the most- imageable verbs. A reversed scale was used in the present study because it might better reflect the amount of effort needed to evoke mental images of an action, with smaller numbers indicating less effort, and greater numbers standing for more effort. In the image agreement task, the participants were asked to generate a mental image of the action corresponding to the displayed verb, then to proceed to the next slide featuring the picture (see Figure 2.1) developed for the verb, and to evaluate the degree of the match between the generated mental image and the presented picture, on a five-point scale from 1 = does not match at all to 5 = matches very well.
2.2.3. Participants
The links to the surveys were distributed online. Each participant received a link to no more than one list for a subset of the verbs (either a word-based list or a picture-based list, but not both), but respondents were allowed to participate in multiple surveys, as long as each of them involved different verb subsets. For 11 of the lists, a total number of 100 participants’ responses were obtained, and for one word-based list, the responses of 102 participants were obtained; 1,202 surveys were completed, in total. In a demographic questionnaire preceding the main test, the participants reported themselves as being neurologically healthy native speakers of Russian (869 females, 332 males, one of unidentified sex; age range: 16-76, mean = 27.46, SD = 10.52; the level of education ranged from the secondary school to a doctoral degree, with 85.19% of the participants having at least some higher education).
2.3. RESULTS AND DISCUSSION
Normative data were collected for the total of 414 verbs and their pictorial counterparts. For 39 verbs, the obtained dominant (most frequent) name for the pictorial representation differed from the expected target name; in these cases, the data for pictures and for verbs were separated, regarded as incomplete, and excluded from further analysis, but they are provided in the supplementary materials of the main database. For the remaining 375 verbs, the means and standard deviations for the parameters of age of acquisition, familiarity, subjective complexity, imageability, and image agreement were calculated and entered into the database, along with the previously identified values for frequency, argument structure, instrumentality, name relation, and length (two measures). The distribution of different verbs used by participants to name each picture is also present in the database. The % name agreement values were estimated on the basis of the expected target name for each picture. The complete database is available for download in Excel format at http://en.stim-database.ru.
2.3.1. Descriptive statistics
Descriptive statistics for the normative parameters (two measures of name agreement, subjective visual complexity, image agreement, age of acquisition, imageability, objective visual complexity, frequency, and length) are presented in Table 2.2. Table 2.3 contains the distribution of items for instrumentality, name relation, and argument structure.
Table 2.2. Descriptive statistics for 375 Russian verbs and action pictures
%NA H SVC OVC IA AoA Imag Fam Freq LogFreq Length: Pers Length: Inf
Mean 76.01 1.20 2.69 264.28 3.96 1.87 1.29 3.69 45.27 1.12 3.22 2.66 Median 80.00 1.07 2.70 237.00 4.15 1.81 1.23 3.70 8.70 0.99 3 2 SD 18.18 0.76 0.43 113.68 0.71 0.47 0.23 0.71 108.66 0.64 1.23 0.98 Minimum value 28.00 0.00 1.69 93.30 1.89 1.07 1.00 1.90 0.00 0.00 1 1 Maximum value 100.00 3.66 3.83 844.00 4.96 3.35 2.57 4.97 957.10 2.98 6 6 25th procentile 63.00 0.59 2.39 185.50 3.43 1.50 1.13 3.14 3.30 0.63 2 2 75th procentile 91.50 1.70 2.98 313.50 4.55 2.20 1.39 4.26 32.05 1.52 4 3 Quartile range 28.50 1.11 0.60 128.00 1.12 0.70 0.26 1.12 28.75 0.89 2 1 Skewness -0.72 0.63 0.03 1.26 -0.77 0.58 1.70 -0.14 4.67 0.67 0.37 0.60
Note: %NA = percentage of name agreement; H = H statistic; SCV = subjective visual complexity; OVC =
objective visual complexity; IA = image agreement; AoA = age of acquisition; Imag = imageability; Fam = familiarity; Freq = lemma frequency per million; LogFreq = lemma frequency per million transformed according to the formula log(1+x); Length: Pers = length in syllables in the form of third person; Length: Inf = length in the infinitive form.
Table 2.3 Argument structure, instrumentality, and name relation for 375 Russian verbs
Instrumentality and Name relation Argument structure
Verb class N of verbs Percentage N of arguments N of verbs Percentage
Non-instrumental 230 61.33% 1 148 39.47%
Instrumental 145 38.67% 1/2 4 1.07%
- Name related 39 26.90% 2 217 57.87%
- Non-name-related 106 73.10% 3 6 1.60%
The median of % name agreement was 80%, a number that is often used as a threshold for selecting experimental and clinical materials (e.g., Cuetos & Alija, 2003). The mean of H = 1.20 and the positive skewness of this measure suggest relatively high name agreement scores for the data set. The statistics on subjective visual complexity indicate that the pictures were rated as having medium levels of complexity, on average (see the low inter-quartile range = 0.60 and the low skewness = 0.03), although the range of values (1.69-3.83) suggests that the whole scale was used. The scores of image agreement were high (mean = 3.96) and positively skewed, meaning that the participants tended to estimate the pictures as matching with their generated mental images, although the whole scale was also used (range = 1.89-4.96). The statistics on age of acquisition suggest that the verbs were generally rated as having been acquired early (mean = 1.87, range = 1.07-3.35). The imageability scores were low (mean = 1.29), with a relatively narrow range (1-2.57), indicating that few verbs in this data set that are hard to picture in mind. The actions in the database were also rated as mostly familiar (mean = 3.69) and the whole scale was used by the participants (range = 1.90-4.97).
2.3.2. Correlation analysis
The correlations among the obtained values of psycholinguistic parameters show the interrelations among measures. A pairwise two-tailed Pearson correlation analysis was performed on the following parameters: % name agreement, H statistics, subjective and objective visual complexity, image agreement, age of acquisition, familiarity, log-transformed frequency, and length in the form of third person (see Table 2.4). The correlations with length in the infinitive form are not reported, since the two length measures are highly correlated [r(373) = .919, p < .001].
Table 2.4. Correlation matrix for normative parameters for 375 Russian verbs and pictures of actions
NA H SVC OVC IA AoA Imag Fam LogFreq
H -.932* SVC -.218* .259* OVC -.128 .148 .534* IA .209* -.225* -.160 -.076 AoA .021 .002 .162 .160 .150 Imag -.218* .279* .287* .083 -.290* .427* Fam .076 -.132 -.317* -.178 .120 -.341* -.253* LogFreq .013 .001 -.069 -.154 -.253* -.396* .072 .418* Length -.027 -.003 .071 .265* .058 .165 -.055 .026 -.363*
Note: %NA = percentage of name agreement; H = H statistic; SVC = subjective visual complexity; OVC =
objective visual complexity; IA = image agreement; AoA = age of acquisition; Imag = imageability; Fam = familiarity; LogFreq = log-transformed frequency; Length = length in the third-person form. The α was set to p < 0.001 (p < .05 Bonferroni corrected); significant correlations are marked with asterisks.
Table 2.5. Corr elati on s r eported in pr evi ous verb n ormin g stu di es , compar ed to th e pr esen t stu dy Th e pr es ent stud y Fi ez & Tr an el (1 99 7) Ch ia re llo , Sh ea rs & L un d (1 999 ) Bi rd , Fr ank lin & H ow ar d (2 00 1) Cu et os & A lij a (2 00 3) Bon in , Bo ye r, Mé ot , Fay ol & D ro it (2 00 4) Sc hw itt er et a l. (2 00 4) C am eir ão & V ice nt e (2 01 0) Sha o, Ro elo fs & Me ye r (2 014 ) Al on so , D íez , & Fer na nde z (2 016 ) Im bi r (2 016 ) Bi rc he no ug h, D av ies , & Con ne lly ( 20 17 ) So ar es , C os ta , M ac had o, Co m es añ a, & O liv eir a ( 20 17 ) Kh wa ileh, Mu sta fa wi , H er be rt , & H ow ar d, (2 018 % N A - H --* -* % N A - S VC -n. s. n. s. n. s. n. s. % N A - I A + +* n. s. +* +* % N A - A oA n. s. -n. s. -% N A - I m ag -+* +* +* % N A - F am n. s. n. s. n. s. % N A - Fr eq n. s. + n. s. n. s. n. s. % N A - L en gt h n. s. n. s. n. s. n. s. H - S VC + n. s. n. s. n. s. n. s. H - I A --* -* n. s. H - A oA n. s. n. s. + + H - I m ag + -* -* -* H - F am n. s. -n. s. n. s. H - Fr eq n. s. -n. s. n. s. n. s. H - L en gt h n. s. n. s. n. s. SV C - I A n. s. -n. s. n. s. n. s. n. s. SV C - A oA n. s. + n. s. n. s. + + SV C - I m ag + n. s. n. s. -* n. s. SV C - F am --* n. s. -* n. s. SV C - Fr eq n. s. -n. s. -n. s. -SV C - L en gt h n. s. n. s. n. s. n. s. n. s. IA - A oA n. s. +* n. s. n. s. -IA - I m ag -n. s. n. s. +* +* IA - F am + n. s. n. s. n. s. IA - Fr eq --* -* n. s. n. s. n. s. IA - L en gt h n. s. n. s. n. s. n. s. Ao A - I m ag + -* -* -* -* -* -* -* Ao A - F am --* -* -* -* + Ao A - Fr eq --* -* -* -* -* -* -* -* -* n. s. Ao A - L en gt h n. s. + + n. s. + + + n. s. Im ag - F am -n. s. n. s. n. s. -Im ag - Fr eq n. s. n. s. + n. s. n. s. n. s. n. s. + n. s. n. s. Im ag - L en gt h n. s. -n. s. -n. s. n. s. Fa m - Fr eq + n. s. +* +* +* +* n. s. Fa m - L en gt h n. s. -n. s. n. s. n. s. Fr eq - L en gt h --* n. s. -* -* -* No te : % NA = p er ce nt ag e o f n am e ag re em en t; H = H s ta ti st ic ; S VC = s ub je ct iv e vi su al c om pl ex it y; O VC = o bj ec ti ve v is ua l c om pl ex it y; IA = i m ag e a gr ee m en t; A oA = a ge o f a cq ui si ti on Im ag = i m ag ea bi lit y; F am = c on ce pt ua l f am ili ar it y; F re q = f re qu en cy . T he p ar am et er Le ng th in t he p re se nt t ab le r ef er s t o t he l en gt h i n s yl la bl es . I n ( Bi rd , F ra nk lin & H ow ar d, 2 00 1) Fa m w as n ot s pe ci fie d f or w or d c la ss . S ch w it te r e t a l. ( 20 04 ) d id n ot e xp lic it ly s pe ci fy w he th er t he Fr eq w as c al cu la te d f or t he l em m a o r t he i nfi ni ti ve f or m . A ll c or re la ti on s i n t he tabl e h ave p < . 05 o r l es s; o nl y t he c or re la ti on s a m on g v ar ia bl es t ha t a re p re se nt i n t he c ur re nt s tu dy a re r ep or te d. “+ ” m ea ns a p os it iv e c or re la ti on , “ -” m ea ns a n eg at iv e c or re la ti on “n .s .” m ea ns t ha t t he c or re la ti on d id n ot r ea ch t he l ev el o f s ig ni fic an ce , a nd a b la nk c el l m ea ns t ha t t he v ar ia bl e o r t he c or re la ti on w as n ot a na ly ze d. S ig ni fic an t c or re la ti on s t ha t a re th e s am e a s i n t he p re se nt s tu dy (i nc lu di ng t ho se w it h t he c or re ct io n f or t he i nv er se i m ag ea bi lit y s ca le i n t he p re se nt s tu dy ) a re m ar ke d w it h a n a st er is k.
Table 2.5 presents the previously reported correlations among the normative parameters of the verbs and action pictures and their correspondence to the findings in the present study. The correlations obtained in the present study are generally consistent with those reported for other verb databases. Each of the significant correlations obtained in the present study is discussed below; the correlations that did not reach the level of significance are not mentioned, unless the observed lack of relationship between psycholinguistic parameters contradicts previous findings.
As expected, a strong3 negative correlation [r(373) = -.932, p < .001] was found between
% name agreement and H, which is due to the fact that the former is a component of the latter by definition: the value of % name agreement is a part of the H formula. The name agreement measures (both % name agreement and H) correlated with imageability [r(373) = -.218, p < .001; r(373) = .279, p < .001, respectively], which is consistent with the previous findings (Bonin et al., 2004; Khwaileh et al., 2018; Shao et al., 2014): the opposite direction of the correlation is explained by the reverse normative scale used in the present study (where 1 = easy to imagine and 5 = hard to imagine, unlike in other studies, where 1 refers to the least imageable point of scale). The correlation between name agreement and imageability suggests that verbs that more easily evoke mental images tend to be named more uniformly and to obtain more responses that are the same as the target name. The name agreement measures also correlated with image agreement. For %name agreement, r(373) = .209, p < .001; these correlations were revealed by Bonin and colleagues (2004), Shao and colleagues (2014) and Khwaileh and colleagues (2018), as well. For H statistics, r(373) = -.225, p < .001; the effect was also found in several studies (Bonin et al., 2004; Fiez & Tranel, 1997). The correlation between H and image agreement indicates that items that have a good match between the mental image and the picture are given more uniform names. These tendencies may occur due to the presence of a conventional mental image related to the verb. This conventional mental image may facilitate image generation in the imageability task. In turn, if a conventional image is represented in the picture, it may lead to consistent naming responses that more frequently coincide with the images the participants pictured in mind in the image agreement task. Finally, the correlations between the name agreement measures (both % name agreement and H) and subjective visual complexity [r(373) = -.218, p < .001; r(373) = .259, p < .001, respectively] suggest that the pictures that were rated as being more complex received less uniform naming responses, and vice versa. Again, this may be due to the perceived visual simplicity of the conventional image (note that there is no correlation between the name agreement measures and objective visual complexity, meaning that those images are not objectively simpler, but they seem simpler due to their conventionality). However, in other verb databases the correlations between the name agreement measures and subjective visual
3 In the use of the effect size terms, we adhere to the conventions of Cohen (1988): “small” correlations = .1 – .3, “moderate” = .3 – .5, “large” >.5.
complexity did not reach the level of significance (Bonin et al., 2004; Cuetos & Alija, 2003; Fiez & Tranel, 1997; Khwaileh et al., 2018; Schwitter et al., 2004; Shao et al., 2014).
The correlation between subjective and objective visual complexity was strong and positive [r(373) = .534, p < .001], suggesting that objective measures of visual complexity can be used as an estimate of speakers’ perspectives on the amount of processing required for visual recognition of an action. Subjective visual complexity negatively correlated with familiarity [r(373) = -.317, p < .001]. The same correlation of subjective visual complexity with familiarity was found in the previous studies (see Bonin et al., 2004; Fiez & Tranel, 1997), which indicates that more familiar actions are usually depicted with fewer details. The fact that correlations with familiarity and imageability [r(373) = .287, p < .001] were only observed for subjective visual complexity, and not objective visual complexity, may imply that subjective ratings of visual complexity, imageability, and familiarity are biased by the same semantic properties of the concept. Apart from that, in our study the familiarity question preceded the visual complexity question, which may also have caused a priming effect for the visual complexity task (but note that the imageability task is in another experimental list). Therefore it is not clear whether subjective visual complexity is an independent parameter, or which visual complexity measure – subjective or objective – would reflect the real visual processing effort. A regression study with a behavioral dependent variable such as action naming reaction times could clarify this issue.
As expected, age of acquisition correlated significantly with imageability, familiarity, and frequency, which means that verbs that refer to actions that are more frequent, imageable, and familiar tend to be rated as having been learned earlier. This result is highly consistent and has been replicated in almost every norming study of verbs, if age of acquisition was considered (Alonso et al., 2016; Birchenough et al., 2017; Bird et al., 2001; Bonin et al., 2004; Cameirão & Vicente, 2010; Cuetos & Alija, 2003; Imbir, 2016; Khwaileh et al., 2018; Schwitter et al., 2004; Shao et al., 2014). The opposite (positive) correlation between age of acquisition and imageability in the present study is explained by the use of the reverse imageability scale. However, a very persistent correlation between age of acquisition and length (Birchenough et al., 2017; Bird et al., 2001; Cameirão & Vicente, 2010; Cuetos & Alija, 2003; Shao et al., 2014) was not maintained in the present study, although it approached significance.
For imageability, correlations with subjective visual complexity [positive, r(373) = .287, p < .001], image agreement [negative, r(373) = -.290, p < .001] and familiarity [negative, r(373) = -.253, p < .001] were observed. The positive correlation between imageability and subjective visual complexity suggests that verbs that require more effort to generate a mental image for are also related to relatively more complex pictures in our database. This reflects the expected link between the difficulty of generating an image and the objective complexity of a drawing. Similarly, the negative correlation between imageability and image agreement indicates that verbs that require more effort to generate a mental image for are naturally related to pictures that are nonpreferred matches for the images
generated by the participants. We assume that these observations pertain to the absence of a conventional verb-related image. Imageability and familiarity are negatively correlated, which indicates that the actions that are rated as more familiar tend to be related to verbs that prompt an easier generation of mental images.
Familiarity was shown to be negatively correlated with subjective visual complexity [r(373) = -.317, p < .001], age of acquisition [r(373) = -.341, p < .001] and imageability [r(373) = -.253, p < .001]. A positive correlation with log transformed frequency [r(373) = .418, p < .001] was observed. The moderate negative correlation between familiarity and subjective visual complexity indicates that the pictures with more familiar actions are rated as less complex, and vice versa. This result was also found in other verb databases (Bonin et al., 2004; Fiez & Tranel, 1997). The moderate correlation between familiarity and log-transformed frequency is in line with previous research (Bird et al., 2001; Bonin et al., 2004; Cuetos & Alija, 2003; Schwitter et al., 2004) and suggests that more frequently used verbs are related to more familiar actions. The correlation between familiarity and image agreement was found in the present study for the first time, and indicates that for more familiar actions, the image offered to the participants tended to be scored as matching with the mental image – again, this may be due to the presence of a conventional image.
The replication of the general correlation patterns between the variables confirms the reliability of the normative values obtained in the present study.
2.3.3. Analysis of name disagreement sources
One of the most common research and clinical uses of the developed experimental stimuli involves a naming task, so it is important to analyze the normative variability in picture naming and its possible sources. In the present database, 49% of items (182 out of 375) obtained name agreement scores lower than 80%. These pictures were classified as either having multiple names (i.e., alternative names which could be considered correct, although they do not coincide with the target name) or erroneous names. In this study, the erroneous answers were defined as having the same semantic-category coordinates (e.g., chikhaet: ‘sneezes’ instead of kashljaet: ‘coughs’) or responses that were not in any way connected to the target verb and were probably due to action recognition failures (e.g., topit: ‘stokes’ instead of pechjot: ‘bakes’). The ratio of pictures that had erroneous names among the answers was relatively low: 107 out of 182 items with % name agreement < 80 obtained multiple names without any erroneous names; for the other 75, erroneous answers occurred, and the percentage of erroneous responses was low (range = 1-39, mean = 6.26, SD = 7.81, median = 3, Q3 = 6).
Table 2.6 presents a comparison between our data and the available data on percentages of name agreement and the H statistics from other databases (Cuetos & Alija, 2003; Fiez & Tranel, 1997; Shao et al., 2014). The differences between our database and that of Fiez & Tranel (1997) on both measures did not reach significance, but name agreement in the present database was found to be significantly lower than in the findings of Cuetos &
