1. INTRODUCTION
1.1 Word properties and psycholinguistic experiments
When different word types, e.g. nouns and verbs, or concrete and abstract words, are compared in psycholinguistic experiments, it is crucial to be able to systematically vary the word properties of interest, while keeping other possibly confounding variables constant. A frequently used approach for evaluating semantic as well as other word properties is to ask people to make ratings of words’ properties on a Likert-type scale. For English, several research groups have gathered word ratings for different properties, among them imageability, concreteness, familiarity, subjective frequency, meaningfulness and age of acquisition (AoA) (Paivio, Yuille & Madigan Reference Paivio, Yuille and Madigan1968, Gilhoolie & Logie Reference Gilhoolie and Logie1980, Coltheart Reference Coltheart1981, Morrison, Chappell & Ellis Reference Morrison, Chappell and Ellis1997, Altarriba, Bauer & Benvenuto Reference Altarriba, Bauer and Benvenuto1999, Balota, Pilotti & Cortese Reference Balota, Pilotti and Cortese2001, Bird, Franklin & Howard Reference Bird, Franklin and Howard2001, Stadthagen-Gonzalez & Davis Reference Stadthagen-Gonzalez and Davis2006, Cortese & Khanna Reference Cortese and Khanna2008, Warriner, Kuperman & Brysbaert Reference Warriner, Kuperman and Brysbaert2013.Footnote 1 Word ratings are also available for a number of other European languages, e.g. Norwegian (Lind et al. Reference Lind, Simonsen, Hansen, Holm and Mevik2015), Portugese (Marques et al. Reference Marques, Fonseca, Morais and Pinto2007) Dutch (Ghyselinck, De Moor & Brysbaert Reference Ghyselinck, De Moor and Brysbaert2000) and French (Flieller & Tournois Reference Flieller and Tournois1994). However, no similar database with word ratings is currently available for Swedish. Since collection of word ratings is a time-consuming process and a large database of rated words should ideally be available to choose stimuli from, a convenient alternative would be to be able to translate words with ratings from already existing large databases in other languages. However, this method presupposes that word ratings are similar enough across languages for transfer of ratings to be applicable. Even if words are accurately translated, their semantic content is likely to differ more or less subtly between languages (see Simonsen et al. Reference Simonsen, Lind, Hansen, Holm and Mevik2013), and it can thus be argued that English word ratings might not accurately reflect the properties of Swedish words.
The present study was carried out in order to obtain a sample of Swedish word ratings for three of the above-mentioned variables (imageability, familiarity and age of acquisition), and to see whether these ratings correlated with corresponding English word ratings. If so, it would be reasonable to assume that directly transferring English word ratings of these properties to Swedish would in general be a valid method.
1.2 Imageability
Concrete words are generally processed with greater speed and accuracy than abstract words (Paivio Reference Paivio, Horne and Roll2010). Concreteness is related to the amount of sensory information associated with a word and is usually assessed by having subjects rate the words’ imageability or concreteness on a 1–7 scale, where 1 = least imageable/concrete and 7 = most imageable/concrete. Whereas concreteness values are based on how directly a word refers to a physical object, imageability ratings are obtained on the basis of judgments of how easily a word evokes a sensory experience or ‘mental image’ (Paivio et al. Reference Paivio, Yuille and Madigan1968; Gillhoolie & Logie Reference Gilhoolie and Logie1980; Paivio Reference Paivio1986, Reference Paivio, Horne and Roll2010). Rated concreteness is highly correlated with rated imageability, and in many studies the two terms are used interchangeably (e.g. Sabsevitz et al. Reference Sabsevitz, Medler, Seidenberg and Binder2005, Fliessbach et al. Reference Fliessbach, Weis, Klaver, Elger and Weber2006, Moroschan & Westbury Reference Moroschan and Westbury2009).
1.3 Familiarity
It is well-known that word processing is affected by how common words are, and word frequencies taken from spoken or written language corpora are often used as an indication of how often a word may have been encountered. Another way to quantify people's experience with words is to ask them to rate how familiar words are on a 1–7 scale where 1 = least familiar and 7 = most familiar (Gilhoolie & Logie Reference Gilhoolie and Logie1980). Familiarity ratings can be used as a complement to word frequencies, or be used in cases where word frequencies are not available, but familiarity judgments may also measure something else than just how often the words are encountered, possibly involving semantic properties. For example, Westbury (Reference Westbury2013) found that a set of affective predictors accounted for 100% of the variance in English familiarity ratings. In some studies, familiarity has even been found to be a better predictor of word processing performance than word frequency (Stadthagen-Gonzalez & Davis Reference Stadthagen-Gonzalez and Davis2006).
1.4 Age of acquisition (AoA)
Early acquired words can be assumed to be processed differently from words acquired later in life, and experience of earlier learned words is likely to be greater than experience of more recently learned words. The AoA variable is quantified on a 1–7 scale, where 1 = lowest age interval (0–2 years of age) and 7 = highest interval (13 years and older) (Gilhoolie & Logie Reference Gilhoolie and Logie1980). Making subjective ratings of when a particular word was acquired may seem difficult and imprecise, but AoA estimates have been found to correspond reliably to objective measures of word acquisition age (Stadthagen-Gonzalez & Davis Reference Stadthagen-Gonzalez and Davis2006).
2. METHOD
2.1 Participants and materials
Nineteen native Swedish speakers (13 female) in the age range of 19–65 years (M = 38, SD = 15) performed word ratings of imageability, familiarity and age of acquisition for 99 Swedish words. All ratings were performed anonymously. The words were all nouns denoting concrete objects and entities as well as emotions and abstract states (Appendix 1). Written word frequencies were obtained from the Stockholm Umeå Corpus (SUC; Ejerhed et al. Reference Ejerhed, Källgren, Wennstedt and Åström1992). SUC frequencies ranged from 1 to 130 occurrences per million (M = 2.48, SD = 0.914). Word length ranged from one to four syllables (M = 24.3, SD = 28.021).
2.2 Procedure
The word rating test was carried out as a web-based rating form (Figure 1), published on an internet page with the MIDAS software (further described in Appendix 2). All words were rated with regard to the variables föreställning (imageability), vanlighet (familiarity), inlärningsålder (age of acquisition) and känsloladdning (emotional arousal).Footnote 2 Word ratings were made on scales ranging from 1 to 7 (1 = the lowest imageability/familiarity/AoA/arousal, 7 = the highest imageability/familiarity/AoA/arousal), following translated versions of the instructions used in the Gilhoolie–Logie norms (Gilhoolie & Logie Reference Gilhoolie and Logie1980), a set of rating norms which the imageability, familiarity and age of acquisition scores in the Medical Research Council (MRC) Psycholinguistic Database scores are partially based on. The words were presented in random order, with one practice word prior to the real ratings. Each word had to be rated for all variables before the next word could be accessed. It was not possible to go back and change any answers. The instructions (Appendix 3, for English instructions see Gilhoolie & Logie Reference Gilhoolie and Logie1980) could be viewed at any time during the test by clicking the icon Instruktioner (‘Instructions’) at the upper right corner of the web page.
Figure 1 Screenshot of the web-based rating form.
2.3 Data analysis
Scores on the 1–7-point scale were transformed by multiplying them by 100, in order to get values on the same scale as those in the MRC database (100–700). In order to see how similar the Swedish imageability, familiarity and age of acquisition ratings were to the corresponding English word ratings, all words were translated and English word ratings were obtained from the MRC database (Coltheart Reference Coltheart1981). The MRC database was chosen since it contains a large number of English words and is easily searchable via a web-based interface.Footnote 3 The MRC database value for each English word was compared with the mean rating of each word from the present study using Spearman correlation. Written frequencies (Kucera–Francis) were obtained from the MRC database in order to see whether or not the Swedish and English word frequencies correlated. All statistical testing was performed in SPSS.
3. RESULTS
All Swedish and English words with ratings are listed in Appendix 1. As can be seen in Table 1 and Figures 2–4, all three word properties were significantly correlated between the two languages (all ps < .001). Very strong correlations were present for imageability (rs = .865) and age of acquisition (rs = .816) and a moderate correlation was seen for familiarity, (rs = .393). The range of familiarity values was more restricted than the range of the other variables (see Table 1), i.e. the words were generally considered to be rather familiar, possibly contributing to the lower correlation for this variable. Statistical testing with Pearson correlation showed that written word frequencies from the SUC (Ejerhed et al. Reference Ejerhed, Källgren, Wennstedt and Åström1992) and the MRC database (Coltheart Reference Coltheart1981) correlated (r = .473, p < .001)
Table 1. Correlations and descriptive statistics of Swedish and English word ratings.
Ima = imageability, AoA = age of acquisition, Fam = familiarity
Figure 2 Scatterplot showing the correlation between English imageability ratings from the MRC database (y axis) and Swedish imageability ratings (x axis).
Figure 3 Scatterplot showing the correlation between English age of acquisition ratings from the MRC database (y axis) and Swedish age of acquisition ratings (x axis).
Figure 4 Scatterplot showing the correlation between English familiarity ratings from the MRC database (y axis) and Swedish familiarity ratings (x axis).
4. DISCUSSION
The present study compared subjects’ ratings of imageability, familiarity and age of acquisition in a sample of Swedish nouns with the ratings of their English translations in the MRC database. The Swedish word ratings were moderately to strongly correlated with the English ratings, indicating that MRC database values can be reliably transferred to Swedish translations of the words. This opens up the possibility of translating a large number of already available English word ratings and using the Swedish translations of the words for psycholinguistic experiments. It should, however, be noted, that since written word frequencies also correlated between the two languages, this might account for some of the shared variance in word ratings.Footnote 4
Although the present study showed that transferring word ratings is a valid option in the absence of Swedish ratings (and even as a complement to Swedish word ratings if they existed but did not comprise the same set of words), it should be stressed that there would be several advantages of a genuine Swedish database. One advantage would be that, other variables whose values cannot be transferred, such as word frequencies and data concerning form-based word properties, could also be included in such a database. It is also the case that some word meanings (especially highly culture-specific ones) might be difficult to translate and ratings of such words’ properties can be expected to differ between languages (Simonsen et al. Reference Simonsen, Lind, Hansen, Holm and Mevik2013). In the present study, some words exhibited larger variation between their Swedish and English ratings. Examples of this include the Swedish word sorg ‘sorrow’, which was rated lower in imageability compared to the English word sorrow (429 compared to 589), but higher in familiarity (589 compared to 486). Swedish ilska ‘anger’ was rated as being substantially more imageable (626) than English anger (488). The Swedish word position ‘position’ had a notably higher AoA rating (526) than its English translation position (375). One explanation for this variation might be that these words’ meanings do not overlap entirely between the two languages.
Furthermore, there are word properties other than the ones compared in the present study that may be less correlated between the two languages (e.g. meaningfulness, Coltheart Reference Coltheart1981) as well as other variables, not available in the MRC database, which it might be useful to have Swedish ratings for (e.g. abstract conceptual features (Crutch et al. Reference Crutch, Troche, Reilly and Ridgway2013). Thus, in the long run it would be ideal to create a Swedish database, preferably searchable via a web-based interface, similar to e.g. the MRC database (Coltheart Reference Coltheart1981) and Norwegian Words (Lind et al. Reference Lind, Simonsen, Hansen, Holm and Mevik2015).
Finally, Swedish and English are structurally similar languages, spoken in similar cultures. Thus, although the results offer support for transferring word ratings between these two languages, they might be less generalizable for translations across less similar languages and cultures. The field would benefit from extending the cross-linguistic comparisons to other languages. Swedish ratings could be compared to those already available in, for example, Norwegian, French, Dutch and Portugese and, ideally, also to word ratings in languages outside of the Indo-European language family.
ACKNOWLEDGEMENTS
This research was supported by the Swedish Research Council: grant 421-2004-8918. We would like to express our gratitude to two anonymous reviewers, whose constructive feedback has greatly improved the paper.
APPENDIX 1
Words with ratings
Ima = imageability, Fam = familiarity, AoA = age of acquisition, Emo = emotional arousal; Swe = Swedish, Eng = English
APPENDIX 2
Method for data collection and database construction
The present study was carried out using the software MIDAS (Mysql Interface and Database Abstraction System). MIDAS is a web content and database management system which can be used to gather various sources of linguistic data and make them easily accessable and searchable through a common interface where search criteria for various parameters, e.g. word frequency, word class and other variables can be specified. MIDAS was used for creating the word rating web page as well as for organizing the data. With MIDAS, all data entered into the system can be downloaded as a .csv-file and directly imported to SPSS for statistical testing.
Word frequencies were obtained from the Stockholm Umeå Corpus (SUC) (Ejerhed et al. Reference Ejerhed, Källgren, Wennstedt and Åström1992) and the Gothenburg Spoken Language Corpora (GSLC) (Alwood Reference Allwood1999), the number of syllables for each word was manually counted, and this information was entered into the database together with data from the MRC database and the word ratings obtained in the present study. In this way, a mini-database was constructed. For access to the database, please contact the first author of the present study: frida.blomberg@ling.lu.se.
APPENDIX 3
Swedish instructions
Instruktioner
Du kommer att få poängsätta ett antal ord gällande några olika egenskaper. En skala med 7 steg kommer att användas i samtliga fall. Känn dig fri att använda hela skalan, men bry dig inte om hur ofta du använder en viss siffra så länge den motsvarar din verkliga bedömning av ordet. Det är inte meningen att du ska lägga ner lång tid på varje fråga – fyll i testet ganska snabbt och baserat på din intuition, men reflektera ändå över frågorna innan du svarar. Observera också att det inte finns några “rätt” eller “fel” svar, utan syftet med testet är att du ska ge en bild av hur du uppfattar orden.
Här kommer ordegenskaperna som du kommer att bedöma:
1) Vanlighet Det varierar hur vanliga olika ord är, dvs hur ofta de förekommer i vardagen och hur välbekanta de känns. En del ord är mycket välbekanta, medan andra kan vara mindre välbekanta eller nästan helt okända. Din uppgift är att poängsätta ordens vanlighet, beroende på hur vanliga du upplever att de är. Skalan sträcker sig från 1–7, där 1 är mycket ovanligt och 7 är mycket vanligt. De ord som du upplever som mycket vanliga ska alltså ges en hög vanlighetspoäng. De ord som du upplever som mycket ovanliga ska ges en låg vanlighetspoäng.
2) Känsloladdning Det varierar hur starkt olika ord är associerade med känsloupplevelser. En del ord väcker inre känslor som kan vara starkt positiva eller negativa, andra ord kan väcka mindre tydliga känsloupplevelser, och ytterligare ord är helt neutrala och väcker ingen känsloupplevelse alls. Din uppgift är att poängsätta ordens känsloladdning, beroende på hur starka känslor de väcker. Skalan sträcker sig mellan 1–7, där 1 motsvarar ett helt neutralt ord och 7 ett starkt känsloladdat ord. De ord som väcker starka känsloassociationer ska ges en hög känsloladdningspoäng. De ord som i mycket liten utsträckning eller inte alls väcker känsloassociationer ska ges en låg känsloladdningspoäng.
3) Inlärningsålder I vilken ålder kan du uppskattningsvis ha lärt dig ordet? En skala med 7 åldersintervall kommer att användas. Intervallen är 0–2 år, 3–4 år, 5–6 år, 7–8 år, 9–10 år, 11–12 år samt 13 år och uppåt.
4) Föreställningar Det varierar hur lätt olika ord väcker inre föreställningar av t.ex. saker, händelser eller upplevelser. En del ord väcker snabbt och lätt föreställningar av synintryck, ljudintryck, känselintryck, lukter och smaker, medan andra ord kan göra det med viss ansträngning (t.ex. efter en lång fördröjning) och vissa ord väcker inte någon inre föreställning alls. Din uppgift är att poängsätta orden beroende på hur lätt de väcker inre föreställningar. Skalan sträcker sig mellan 1–7, där 1 är svårast att föreställa sig och 7 är lättast att föreställa sig. De ord som snabbt och lätt väcker inre föreställningar ska ges en hög föreställbarhetspoäng. De ord som med svårighet eller inte alls väcker inre föreställningar ska ges en låg föreställbarhetspoäng.
