What NN compounding in child language tells us about categorization Maria Rosenberg Ingmarie Mellenius Dept. of Language Studies Dept. of Language Studies Umeå university Umeå university SE-901 87 Umeå SE-901 87 Umeå SWEDEN SWEDEN maria.rosenberg@umu.se ingmarie.mellenius@umu.se perspectives on objects (cf. Waxman and Mar- 1 Introduction kow, 1995). “Do their [children’s] categories reflect only what their language offers, or do The present study examines novel NN com- they – must they– make use of other representa- pounds, produced on line, in Swedish child lan- tions too?” (Clark, 2004:472). guage, with focus on categorization. Given that Berman (2009) emphasizes that there is a sub- NN compounds denote objects, we concentrate stantial difference in adults’ vs. children’s lexi- on the categories those objects belong to. In that cons of established compounds, and that children way, our study aims to provide evidence of ob- have to grasp inter alia the idea of subcategoriza- ject categorization in preschool children. Two tion. Clark and Berman claim that “knowledge of questions are put forward: the pertinent lexical items, and not the construc- (i) Does perception play a crucial role for the tions they appear in, is more important for [chil- children’s coinages? dren’s] compounding” (1987:560). (ii) In what way do structural and processing In conceptual development, category struc- views on categorization apply to the data? tures change with age (Keil and Kelly, 1987). Swedish children produce compounds already Object categorization allows generalization over at age two, reflecting the fact that compounding properties of objects and of novel category is a productive word formation device. In short, members (Mandler, 2000). Swedish compounds are right-headed, written as Bornstein and Arterberry (2010) mention two one word, pronounced with a two-peak- complementing views of categorization: pro- intonation, and can exhibit liaison forms. cessing and structural. On the processing view, 2 Theoretical background categories are flexible and category membership of objects can vary in different situations (cf. e.g. Clark (2004) argues that language acquisition Jones and Smith, 1993). On the structure view, builds upon already established conceptual in- categories are hierarchically organized taxono- formation, which enables the child to categorize mies (cf. e.g. Murphy, 2002). Instead of Rosch’s objects, relations and events. Children rely main- (1978) superordinate-basic-subordinate, levels of ly on shape as they embark on the mapping of category inclusiveness can be ordered in a neu- words for objects onto their conceptual catego- tral way, such as L1 (animal), L2 (cat, dogs), L3 ries of objects, but also pay attention to texture, (collies, shepherds), L4 (scotch collies, border size, sound, motion and function. Even into collies) (Bornstein and Arterberry, 2010:3). adulthood, children continue the mapping of un- Whether categorization proceeds from con- known linguistic items onto conceptual represen- crete to abstract or the other way around is still tations. Young children occasionally form emer- under debate. Differentiation theory (e.g. Gibson, gent categories, based on non-conventional dis- 1969) stipulates that the ability to make finer tinctions (e.g. ball for round things). Clark differentiations emerges after broad conceptions (2004) notes that the pairing of word to object are acquired. Likewise, Bornstein and Arterberry enables the child to perceive similarities between (2010) indicate that more inclusive levels of cat- cognitive categories, and allows for alternate egorization appear before less inclusive ones, Copyright © by the paper’s authors. Copying permitted for private and academic purposes. In Vito Pirrelli, Claudia Marzi, Marcello Ferro (eds.): Word Structure and Word Usage. Proceedings of the NetWordS Final Conference, Pisa, March 30-April 1, 2015, published at http://ceur-ws.org 71 and that high perceptual contrasts have prece- 4.1 N1 and N2 sorting dence over low. Fisher (2011) suggests that at The sorting of N1 and N2 shows that several age 3-5, perceptual information is anchored more nouns reoccur in the children’s compounds. 126 strongly than conceptual information; cognitive N1 of the 383 compounds were either identical flexibility develops with age. or belonging to the same morphological family, Yet, according to Smith (1984), preschool such as morotsvatten ‘carrot-s-water and moröt- children show the ability of both concrete cate- termacka ‘carrots-sandwich’. With respect to N2, gorization, due to perceptual characteristics, and this number was as high as 143. abstract categorization, leaning on conceptual The largest morphological family found in our relationships. Nguyen and Murphy (2003) posit data contains vatten ‘water’. 12 compounds are three categorization forms: taxonomic (see attested (7 compounds from one child, whereof 4 above), script and thematic. Script-based catego- have vatten as N1, and 3 as N2). The two other ries include objects (e.g. egg, cereal) with the children used vatten in 4 and 1 instances respec- same functional role in a routine event (e.g. eat- tively, such as vattenkaffe ‘water-coffee’ or the ing breakfast). Thematic categories involve ob- aforementioned morotsvatten ‘carrot-s-water’. jects that usually appear together (e.g. bowl- Other nouns that reoccurred nearly ten times cereal). Nguyen and Murphy (2003) show that among the innovations of all three children were children, aged 3 to7, use taxonomic and script bil ‘car’ kläder ‘clothes’, mamma ‘mommy’ and categorization in a flexible way. väg ‘road’ (cf. 4.6). 3 Data and method It is worth noting that although the same nouns were used in several compounds, they did The data consists of 383 spontaneously produced not always uphold the same relation to the other NN compounds from three monolingual Swedish constituent: pizzabil ‘pizza-car’ was used for a children, aged 1 to 6, collected longitudinally car with a pizza print on it (viz. perceptually), and including contextual information. The chil- whereas dimbil ‘fog-car’ referred to an imagina- dren often give an explanation of the intended tive car spraying fog (viz. abstractly). meaning, e.g. hundstall ‘dog-stable’, ‘where Overall, the overlap between the same nouns dogs live, outside’. Hence, they seem to under- being used in several compounds and as first or stand the semantics of their novel compound. We second constituent, can be taken as support for use a strict selection criterion: only non- Clark’s and Berman’s (1987) claim (cf. 2) that established compounds in contemporary Swedish children use lexical items that they are familiar are considered. with in their compounding. As a first step to analyze our data, we sort the compounds in two ways: (i) based on N1; (ii) 4.2 Level of inclusiveness based on N2. This is a way of locating items be- As for the level of inclusiveness, the compounds longing to a same morphological family (cf. in our data are situated on L1 (björkgrej ‘birch- Schreuder and Baayen, 1997). As a second step, thing’), L2 (brödrosta ‘toaster’), L3 (äppelsvans the data is analyzed according to: (iii) level of ‘apple-tail’) or L4 (hjärtklackskorna ‘heart-heel- inclusiveness; (iv) script; (v) thematicity; (vi) shoes), with L3 as the predominant level. If we perception (real-world referent or not, high con- look only at N1 or N2 in isolation, they can also trasts vs. low). As a third step, other characteris- correspond to items located at L1 (djur ‘animal’, tics appearing from the children’s compounds are L2 (björn ‘bear’), or L3 (äppeljuice ‘apple-juice’) analyzed. in three-psart compounds. Moreover, there are some compounds in our 4 Analysis data containing a taxonomic relation between the constituents: two examples are ugglafågel ‘owl- In the analysis we provide evidence of categori- bird’ and skinndjur ‘skin-animals’. zation concerning larger groups of compounds. Below follows some preliminary findings. Note 4.3 Script-based categories that the compounds can be analyzed according to different parameters and, thus, some of them go Entire sets of the compounds can be analyzed as into several labels, depending on the parameter having the same role with respect to a script, in taken under account. which the compounds fulfill the same part. All three children categorize clothes according to season or weather, as indicated by N1: sommar- 72 vantar ‘summer-gloves’, snöstrumpor ‘snow- redundancy, is one way to arrive at overcategori- stockings’ or vinterficka ‘winter-pocket’. zation, as we see it. For instance, kogräs ‘cow- There are also compounds in our data where grass’ denotes ‘ordinary grass, that cows eat’ N1 and N2 participate in the same scripts that according to one child. Additionally, an ordinary concern different types of edibles: grötmjölk car is referred to as motorbil ‘motor-car’, or ‘porridge-milk’ (eating breakfast”) or pizzaham- handfinger ‘hand-finger’ is used instead of just burgare ‘pizza-hamburger’ (eating dinner”) or finger for the body part. In these three examples, saftglass ‘syrup-ice cream’ (eating dessert). N2 alone would have been the target like word to use, but the children limit its use further. 4.4 Thematic categories A quite odd categorization made by all three Thematic categories, items with close semantic children, independently, is to add the goal of a association based on, e.g., contiguity, are numer- direction to the direction: kalasväg ‘party-road’ ous within the compounds. An example is or mormorväg ‘granny-road’. Recall that väg häxafiskspö ‘witch-fish-wand’, where the child ‘road’ was one of the nouns that reoccurred fre- aims at a wand used by a witch, but confuses quently among the novel compounds (cf. 4.1). trollspö ‘magic-wand’ with fiskespö ‘fishing- Hence, the three children seem to find it im- pole’, and than adds the user of the item in ques- portant to name particular roads. tion (actually a case of “overcategorization”, cf. Furthermore, nearly 20 of the children’s com- 4.6). pounds contain one of the words mamma Several themes are found. One is “sweets”, ‘mommy’, pappa ‘daddy’ or bebis ‘baby’ as N1 giving rise to numerous compounds, semantical- or N2, such as mammfluga ‘mommy-fly’, ly associated or not, such as silvergodis ‘silver- fågelpappa ‘bird-daddy’ or bebismyra ‘baby- candy’ and godisstrumpor ‘candy-stockings’. ant’. All three children coined such compounds, Most of the thematic categorization found in which we interpret as a kind of emergent catego- the children’s innovation is abstract and ground- rization, as well as of overcategorization. There ed in conceptual information. Furthermore, the were two types of relations involved in these thematic relations are mostly of an inherent na- compounds: animals or insects subcategorized ture, such as manifested by djungelträd ‘jungle- according to human kinship terms as in the pre- tree’, rather than temporal, such as fotbollsplanet ceding examples; mommy or daddy subcatego- ‘football-planet’. rized according to some habit, such as ciga- rettpappa ‘cigarette-daddy’. 4.5 Perception 4.7 Ad hoc categorization Compounds categorized according to Shape are attested, such as R-paprika ‘a piece of paprika Barsalou (1983) uses the label ad hoc categories that looks like a R’, or mössaboll ‘hat crumpled for categories constructed on the spot to achieve into the shape of a ball’. Shape may concern ei- certain goals, such as “things to sell at a garage ther the head or the non-head of the compound. sale”. These categories are much less established Texture is involved in many of the children’s in memory than common categories. We inter- compounds, such as: pälsmatta ‘fur-carpet’. pret ad hoc categories to encompass compounds Prints are also a frequent way to distinguish such as Downing’s (1977) “apple-juice seat”, among clothes they want to wear, or vehicles that and also the examples from Clark, Gelman and they see, such as the above-mentioned “pizza- Lane (1985) claimed to involve a temporal rela- car”. tion, in contrast to compounds with inherent rela- Yet, note that many of the children’s coinages, tions. According to Clark, Gelman and Lane which involve perception, can do so in an imagi- (1985), children would more often use novel nary way, or in other words, as mental imagery. compounds to express inherent relations among A compound, such as champagnetröja ‘cham- objects. The opposite stand is taken by Mellenius pagne-sweater’, was uttered to denote a non- (1997), supported by Berman (2009), who claims existent sweater that the child just dreamt up children’s novel compounds are “highly ‘con- when playing. text-dependent’ and hence more likely to express temporary rather than intrinsic relations” 4.6 Overcategorization (2009:311). We will use the term “overcategorization” to la- Some innovations in our data can be analyzed bel some striking features among the children’s as ad hoc instances that the children coin sponta- compounds. Underextension, often involving neously without a real naming demand. They are 73 typically difficult to understand, or does not Analyses. Developmental Psychology, 46(2):350- make sense, outside the context of the utterance. 365. An example is one child in the data that invents a Eve V. Clark. 2004. How Language Acquisition triplet of compounds with glass ‘ice cream’ with Builds on Cognitive Development. TRENDS in the goal “things that could possibly constitute ice Cognitive Sciences, 8(10):472-478. cream”: träglass ‘wood-ice cream’, sockerglass Eve V. Clark and Ruth A. Berman. 1987. Types of ‘sugar-ice cream’ and glassögon ‘ice cream- Linguistic Knowledge: Interpreting and Producing eyes’; the latter denotes, according to the child, Compound Nouns. Journal of Child Language, ‘eye-glasses but made of ice cream (glass) in- 14(3):547-567. stead of glass (glas)’. Another example is Eve V. Clark, Susan A. Gelman and Nancy M. Lane. kungtröja ‘king-sweater’, coined on the spot 1985. Compound Nouns and Category Structure in when playing: ‘if you wear that sweater you will Young Children. Child Development 56(1):84-94. be the king’. However, our data points in the direction that Pamela Downing. 1977. On the Creation and Use of the children’s innovations more often express English Compound Nouns. Language 53(4):810- 842. inherent relations than temporal relations, but this issue certainly merits further investigation. Anna V. Fisher. 2011. Processing of Perceptual In- formation Is More Robust than Processing of Con- 5 Conclusion ceptual Information in Preschool-Age Children: Evidence from Costs of Switching. Cognition, The study provides evidence of on-line categori- 119(2):253-264. zation based on spontaneous production of novel Eleanor J. Gibson. 1969. Principles of Perceptual NN compounds from three Swedish children. Learning and Development. New York: Appleton- Compared to experimental situations, limited by Century-Crofts. the material used and the children’s will and en- ergy to participate, our collection of data is Stevan R. Harnad (ed). 1987. Categorical Perception: The Groundwork of Cognition. Cambridge: Cam- unique. It shows that high contrast perceptual bridge University Press. features give rise to much subcategorization, however not at the expense of conceptual subcat- Susan S. Jones and Linda B. Smith. 1993. Place and egorization, equally important in our data. Perception in Children’s Concepts. Cognitive De- Since we lack clear longitudinal facts of how velopment, 8:113-139. object categorization emerges within the chil- Frank C. Keil and Michael H. Kelly. 1987. Develop- dren, the structure view is hard to apply. We can mental changes in category structures. In S. Harnad state that L3 and L4 categories appear around age (ed.), Categorical Perception: The Groundwork of 2, but lack numbers about their overall frequency Cognition. Cambridge: Cambridge University in relation to more inclusive categories. Given Press. that the children show cognitive flexibility in Jean M. Mandler. 2000. Perceptual and Conceptual their categorization of an object in a particular Processes in Infancy. Journal of Cognition and way by producing an NN compound, the pro- Development, 1(1):3-36. cessing view conforms better to our data. To Ingmarie Mellenius. 1997. The Acquisition of Nomi- conclude, the children often categorize objects in nal Compounding in Swedish. [Travaux de a much more detailed way than adults do. l’Institut de Linguistique de Lund 31]. Lund: Lund University Press. References Gregory L. Murphy. 2002. The Big Book of Concepts. Lawrence W. Barsalou. 1983. Ad Hoc Categories. Cambridge, MA: MIT Press. Memory & Cognition, 11(3):211-227. Simone P. Nguyen and Gregory L. Murphy. 2003. An Ruth A. Berman. 2009. Children’s Acquisition of Apple is More Than Just a Fruit: Cross- Compound Constructions. In Rochelle Lieber and Classification in Children’s Concepts. Child De- Pavol Štekauer (eds.), The Oxford Handbook of velopment, 24(6):1783-1806. Compounding, 298-322. Oxford: Oxford Universi- Eleanor Rosch. 1978. Principles of Categorization. In ty Press. Eleanor Rosch and Barbara B. Lloyd (eds.), Cogni- Marc H. Bornstein and Martha E. Arterberry. 2010. tion and Categorization, 27-48. Hillsdale, NJ: Erl- The Development of Object Categorizaion in baum. Young Children: Hierarchical Inclusiveness, Age, Perceptual Attribute, and Group versus Individual 74 Robert Schreuder and Harald B. Baayen. 1997. How Complex Simplex Words Can Be. Journal of Memory and Language, 37:118-139. Linda B. Smith. 1984. Young Children’s Understand- ing of Attributes and Dimensions: A Comparison of Conceptual and Linguistics Measures. Child Development, 55(2):363-380. Sandra R. Waxman and Dana B. Markow. 1995. Words as Invitations to Form Categories: Evidence from 12- to 13-Month-Old Infants. Cognitive Psy- chology, 29:257-302. 75