Article Adaptive Behavior 2019, Vol. 27(5) 307–330 Ó The Author(s) 2019 How children develop their ability to Article reuse guidelines: combine words: a network-based sagepub.com/journals-permissions DOI: 10.1177/1059712319847993 approach journals.sagepub.com/home/adb Lluı́s Barceló-Coblijn1 , Aritz Irurtzun2, Cristina Real Puigdollers3, Emilio López-Navarro1 and Antoni Gomila1 Abstract A new way to study and represent early syntactic development is introduced that offers a promising avenue to improve on standard cumulative approaches to language learning. The analysis is inspired by complex network theory and explores an important issue in psycholinguistics: how children combine words syntactically. To this end, the article exploits the longitudinal data of multiple individual studies stored and publicly available in the CHILDES database. This analysis proves useful in two regards: in coincidence with previous approaches to syntactic development, it reveals a sim- ilar linear–nonlinear pattern of syntactic development (a combination of linear periods interrupted by abrupt transitions) in typical children regardless the language they are acquiring. It also provides a straightforward objective measure of syn- tactic impairment in atypically developing children. A striking difference between typical and atypical children lies in the connectivity of functional words, which suddenly become hubs but only in typical development. Our technique offers a formal measure of impairment with respect to typical syntactic development, which goes beyond the standard notion of retardation in development, for the Down syndrome condition. Keywords Language acquisition, analysis of corpora, complex networks, syntax, Down syndrome Handling Editor: Alberto Antonioni, University College London, UK 1. Introduction the field of language development and clinical linguis- tics. Section 2 very briefly introduces complex net- In this study, we explore the possibility of establishing works, and section 3 presents previous applications of different language development endophenotypes based complex network analyses to language and speech. on a network analysis of spontaneous speech corpora Section 4 presents the scope and goals of our study, and (Barceló-Coblijn, Benı́tez-Burraco, & Irurtzun, 2015). To section 5 the methods we propose for the analysis of do this, we establish a comparison between typically spontaneous speech. Section 6 is devoted to the results developed and atypically developed children. We com- and section 7 to their discussion. Finally, in section 8 pare typical language samples from four Romance lan- we discuss the consequences, merits, and limits of our guages (Italian, French, Catalan, Spanish), three study, and comment on the future research venues that Germanic languages (English, German, Dutch), and a it opens. language isolate (Basque), uncovering a common schema in development despite the typological differences 1 University of the Balearic Islands, Palma, Spain between those languages. Next, we compare the patterns 2 Centre National de la Recherche Scientifique (CNRS), Centre IKER of these typical populations with the patterns issued from (UMR 5478), Bayonne, France 3 the analysis of the speech of atypical populations (Dutch Universitat Autònoma de Barcelona, Barcelona, Spain and English-speaking people with Down syndrome (DS)) Corresponding author: observing clearly divergent developmental paths. Lluı́s Barceló-Coblijn, University of the Balearic Islands, cra. de The structure of the article is the following one: in Valldemossa, km 7.5, 07122 Palma, Spain. the remaining of section 1, we frame our study within Email: lluis.barcelo@uib.cat 308 Adaptive Behavior 27(5) 1.1. Differing views on language development 2014; Mundy, Kasari, Sigman, & Ruskin, 1995; The development of new tools and approaches for lan- Roberts, Price, & Malikin, 2007). In many works, individuals affected by a pathologi- guage analysis has been an important target of many cal condition are compared to controls not in terms of branches in cognitive science. One of the pitfalls that their biological age, but in terms of their alleged Mental applied linguistics faces concerns the definition of the Age (MA). Nevertheless, MA is a measure that can be target of study itself, since it is common to find radi- calculated in several different ways and is not exempt cally opposed views regarding some aspects of human of problems and discussion as Brock (2007) pointed cognition. Some issues are traditionally discussed out. That is, atypically developing (AD) individuals are dichotomically: does the ability for language develop often compared to typically developing (TD) individu- linearly or non-linearly? How do biological parameters als regarding the scores obtained in the particular MA and the environment surrounding the speaker contrib- test selected by the scholars. Those problems and differ- ute in language development within the individual (cf. ent ways to test the MA of an individual have high- the ‘‘nature vs. nurture’’ debate)? Especially, conten- lighted the necessity for looking for other tools whose tious was Chomsky’s claim on syntactic development measures can complement the information that a that domain-general learning is unable to account for researcher is already able to obtain from more tradi- language acquisition (cf. i.a.Chomsky, 1975), and that tional systems (such as tests, tasks, recordings, scans, some innate knowledge of grammar has to be hypothe- etc.). If one is interested in knowing the syntactic ability sized to account for this process. Authors like Mehler of a speaker, one will see that here too the systems used (1974, et seq.) have even hypothesized that learning a to calculate that ability are indeed many (and some- language amounts to ‘‘unlearning’’ others (see also times they yield contradictory results). Baker, 2005; Lenneberg, 1967; Piattelli-Palmarini, A basic aspect of the available measures is how they 1989, among many others). were conceived, that is, whether the creators of that The opposite view to biological-centered approaches particular measure/device conceived a priori that the (or bio-approaches) concentrates on the use of lan- language ability is something that develops linearly or guage: how speakers use language, when, and in which not (see Müller and Ball, 2013, for an overview). contexts. A common assumption in usage-based Research methods can be divided in language produc- approaches to language acquisition has been that syn- tion and comprehension methods. Among the former, tactic development is a linear, cumulative process for example, there are measures such as the Assessment (‘‘gradual’’ and ‘‘piecemeal,’’ in Tomasello’s, 2003, of Intelligibility of Dysarthric Speech (Yorkston & terms). If the biological-centered viewpoint has used Beukelman, 1981), the Mean Length of Utterance the switchboard metaphor, according to which children (MLU, Rice et al., 2010), or the Index of Productive select a parameter value from a limited set of possibili- Syntax (IPS; Scarborough, 1990), which are used to ties (ideally a binary parameter, ‘‘on/off,’’ say, head- track syntactic development. These are implicitly com- first vs head-last), usage-based approaches consider mitted to a linear view of the ontogeny of syntax. that syntactic categories and structures are mastered These measures, though, are viewed as imperfect ‘‘one after another,’’ so that it should be possible to proxies for syntactic development due to the well- index the moment a particular category and/or rule is known problems of the variability in morphological acquired. While bio-approaches consider that children structure (words may comprise just one or several mor- transfer categorial information from one word to phemes), and the critical role of functional, closed-class another, usage-based approaches are reluctant (if not words for mastering syntax (Leonard & Finneran, contrary) to this kind of learning. 2003; Rollins, Snow, & Willett, 1996). Informative as it is, a crucial weakness of the MLU measure is that sen- tences of an equal number of words may be very differ- 1.2. Language development and clinical linguistics ent from a syntactic point of view (they may imply In consonance with the current black-or-white scenario, radically different syntactic structures involving hypo- research in clinical linguistics is often approached taxis vs parataxis, different sorts of dependencies and dichotomically as well. For example, although there is ‘‘displacement’’ strategies, etc.). As a consequence, a research view that considers that linguistic skills in these linear measures turn out to be of limited useful- pathologies like DS do progress in a divergent way over ness in characterizing atypical development. Another the life course, another view held for a long time that tool often used nowadays by neurologists (among other DS development was a slowed down process (this view researchers) is the Boston Naming Test (Kaplan, is still held nowadays by some authors). Thus, individu- Goodglass, & Weintraub, 1983), although the accuracy als affected by some pathologies are taken to lag behind of this test is still not completely satisfactory since their typical peers (often under the term delay, for sometimes individuals reach a score that would suggest example, Abbedutto, Warren, & Conners, 2007; the patient has no problem at all, even when the indi- Galeote, Soto, Sebastián, Checa, & Sánchez-Palacios, vidual has been diagnosed on the contrary. Other ways Barceló-Coblijn et al. 309 employed to assess the syntactic proficiency of AD (Gottesman & Gould, 2003). Accordingly, this new individuals have been the Words and Sentences Scale of way to observe language development takes into the MacArthur–Bates Communicative Development account that the development of an organism tends to Inventory (MCDI), by asking the parents about what follow a particular developmental path. In other words, the child is able to perform (Fenson et al., 1993), or the psychological ability for language would have been making the participants to repeat a series of sentences. somehow canalized during evolution (Balari & In general, results of works trying to compare different Lorenzo, 2013; Benı́tez-Burraco & Boeckx, 2014) and clinical profiles rely on this type of tools. In the case of clinical linguistics should take into account the many asking the parents, the researcher needs to trust what levels at which the development of an organism can be they think the AD child is able to perform, and the intervened or affected (Benı́tez-Burraco & Murphy, results of sentence repetition tasks can vastly vary 2016). This third way recognizes the biological predis- between pathologies (e.g. people affected by Williams position of humans to develop language in typical con- syndrome seem to perform orally better than individu- ditions and the important role of the environment als with DS). Comprehension techniques such as the surrounding the individual. De-canalization (Gibson, picture-matching task do not show in and of them- 2009) on the other way could explain ‘‘the high preva- selves whether the individual has captured all the intri- lence of complex diseases (including language disor- cacies of a specific syntactic construct, from lexical ders) among human populations’’ (Benı́tez-Burraco, syntax to sentential syntax (cf. Roeper, 2007; Schmitt 2016, p. 264). As we said, between genes and patholo- & Miller, 2010, among others). The above-mentioned gies there exist cognitive, neuroanatomical, neurophy- IPS seems to address some of the weaknesses in the siological, endocrine, or biochemical quantifiable MLU. The IPS score is obtained from a corpus of 100 components that can define endophenotypes. Recently, utterances, within which 56 specific language structures Barceló-Coblijn et al. (2015) have put forward an endo- must be found. IPS has been implemented into a new phenotype for language disorders which integrates up tool (Sagae, Davis, Lavie, MacWhinney, & Winter, to four factors: (1) linguistic analysis (syntactic compu- 2010) by combining it with Natural Language tation), (2) information management (communicative Processing techniques. In particular, the new IPS strategies), (3) recent Evo-Devo insights in the nature approach introduces dependency structures to identify of phenotypic variation, and (4) network approaches to grammatical relationships (Hudson, 1990). In general, emergent properties of complex systems. Thus, our these methodologies have sought to capture in one way reflections and conclusions on the results obtained in or another the level of syntactic complexity reached by the present works have to be envisaged from the Evo- the speaker. Devo viewpoint, as syntactic networks are considered In the following section, we present a new approach, here proxies of syntactic endophenotypes. focused on an Evo-Devo perspective, which is able to Before starting the linguistic analysis by means of capture aspects of the complexity of the speaker’s lin- networks, and to better understand a linguistic guistic system which were unapproachable with previ- approach that combines syntactic analyses and com- ous techniques. This approach is born from the plex networks, it is necessary to explain what the con- combination of studies on complexity theory and net- cept of complexity is in network science and how it can work science on one hand and linguistic (syntactic) the- be integrated in psycholinguistics. ory on the other hand, and its results are interpreted under the light of an Evo-Devo approach to language development. 2. Computational approaches to linguistic complexity 1.2.1. An Evo-Devo approach to clinical linguistics. Here, we The term complexity, as several others in science, is not want to propose a new technique seeking to bypass the easy to define. John Locke in his Essay Concerning dichotomous view of language development. This tech- Human Understanding (ch. XII) says that ‘‘Ideas thus nique is based on the emerging viewpoint which builds made up of several simple ones put together, I call upon Evo-Devo approaches and conceptual work Complex’’; and paraphrasing the physicist Neil Fraser developed in biology (e.g. West-Eberhard, 2003). The Johnson (2007), a neutral, acceptable explanation could Evo-Devo perspective avoids the debate between two be that a complex behavior emerges from the interac- apparently opposed viewpoints, the first considering tion of a collection of objects (but see Mitchell, 2009, ontogeny focused on genes only and the counter view- ch. 7, for discussion). We consider that complex systems point considering language acquisition to be something theory offers a promising new approach to language, as that has nothing to do with biology. Between genes it does in many other areas of scientific inquiry (Watts and diseases there exist cognitive, neuroanatomical, & Strogatz, 1998). The most important tenet of network neurophysiological, endocrine, or biochemical quantifi- science (Newman, 2010) is that two networks could able components that can define endophenotypes contain the same number of nodes (or vertices), but a 310 Adaptive Behavior 27(5) 1 different number of edges (or connections) and there- nodes tend to connect to highly connected nodes ), pre- fore, both networks may show a completely different ferential acquisition (early learned words seem to be behavior and radically different scores regarding net- highly connected because they stand out in the learning work complexity measures. In other words, two systems environment and are more noticeable), and the lure of having the same number of elements behave differently associates (‘‘unknown words may be highlighted by due to the different relationships holding between their known words to which they are related’’). In that work elements. An interesting aspect of some networks is the semantic networks were built on the basis of 130 nouns. presence of hubs. Hubs are highly connected nodes, usu- Thus, nodes represented nouns and links represented ally few in number within the network. A network of n relationships between nouns (e.g. both ball and apple nodes can develop the characteristics of a small-world if share the feature round). Interestingly, the authors cov- it has a high cluster coefficient (indicating how many ered a sensitive period for language development, from neighbors of a given node are also neighbors of each 16 to 30 months. Their results called into question that other) and a low average shortest path length (a small the principle of preferential attachment is the key of number of links along the shortest paths that must be language development (and their results also high- traversed when navigating the shortest paths in the net- lighted the relevance of the connectivity of a word in work). In a nutshell, the number of nodes provides an the environment). important descriptor of the size of a network, but even Besides, Beckage, Smith, and Hills (2011) investi- more important is the particular connectivity pattern gated the growth of a semantic network by means of that the network has. Importantly, the network-based word co-occurrence. Their study analyzed both typical viewpoint is highly versatile, as it can be applied to very and late talkers, and their results support the intuition different examples, ranging from urban networks that small-worldness is related to the child’s lexical (Neal, 2018), to biological networks (Arita, 2005), or to development. Until then, most works in semantics were linguistic networks (e.g. among many others, Solé, based on English, German, Dutch, and Spanish, but Corominas-Murtra, Valverde, & Steels, 2010). Kenett, Kenett, Ben-Jacob, and Faust (2011) took a step further working on Hebrew—a Semitic language famous for its morphological roots and patterns (e.g. 3. Networks and language Ravid & Schiff, 2006)—and investigated the networks The application of networks to linguistic studies is not both at a local and global level and how these two levels brand new and has been applied to semantics, phonol- interact. Thus, at the local level, these authors observed ogy, or to the analysis of texts. At the end of this sec- that the network was organized into sub-cliques, based tion, we will show that syntax and the development of on semantic categories. the syntactic ability can also be approached with net- Recent work on semantics and networks has paid work analyses. Phonology is, probably, the area of lin- attention to the ontogeny of the mental lexicon (Stella, guistics that has gotten most attention in network Beckage, Brede, & De Domenico, 2018). These authors studies. One such approach is Vitevitch (2008), where have taken into account several properties of lexicon the target was to capture relevant aspects of the mental (semantics, taxonomy, and phonology), showing that lexicon. A key concept in that work was neighborhood these properties indeed interact detecting a cluster of density, which accounts for the number of words that highly frequent words. Moreover, Stella and collabora- sound as a selected word (thus a word A is a ‘‘phonolo- tors conclude that emergence of this core of the lexicon gical neighbour’’ of B, if a single phonological change occurs abruptly during ontogeny (around age 7). to A converts A into B). In these regards, the similari- On another front, it has already been shown that lan- ties between languages as different as Spanish, guage corpora exhibit a complex system behavior, and Mandarin, Hawaiian, and Basque point into the direc- that they can be represented in terms of a graph of word tion of commonly shared psycholinguistic mechanisms interactions (Corominas-Murtra, Valverde & Solé, 2 used in the architecture of the mental lexicon 2009; Ferrer-i-Cancho & Solé, 2001). In these (Arbesman, Strogatz, & Vitevitch, 2010). Besides, it has approaches, however, words are not connected by syn- to be noted that the usefulness of this type of approach tactic or semantic relations between them, but taking is also being expanded to clinical studies as well into account the ‘‘distance’’ to each other within the (Kenett et al., 2013; Vitevitch & Castro, 2015). sentence. The maximal distance between words— However, the application of the network science is typically between 2 and 3—is assessed by the researcher. not limited to phonology. Semantic networks have also These calculations have been used mainly in English. been studied in the last years, yielding interesting results Nevertheless, if applied to other languages such as and hypotheses. For example, Hills, Mouene, Mouene, German or Dutch, which split many verbs into two Sheya, and Smith (2009) tested three different models components, the result could be that these components of word acquisition, based on preferential attachment apparently have no link or relation. Consider, for (according to which ‘‘rich become richer,’’ that is, new instance, the following sentences: (1) is an example of Barceló-Coblijn et al. 311 split of the verb in present perfect in German and (2) is (2009) developed a first approximation to combining an example of a phrasal verb in Dutch: syntactically hand-annotated sentences with network (1) Heute bin ich mit meinem Bruder um zehn Uhr representations and then tracking the longitudinal mit dem Bus nach Hause gefahren. stages in linguistic development. In that work, two Today am[AUX] I with my brother at ten hour longitudinal English corpora were syntactically ana- with the bus to house ridden[PAST PARTICIPE] lyzed by hand and then the analyzed sentences of each Today I have gone home with my brother at ten file were represented by a network program, where syn- o’clock by bus. tactic dependency relations constituted the basis of the (2) Ik sta elke dag om seven uur op. links or edges in the networks. First, during the syntac- I stay[VERB] every day at seven hour up[PARTICLE]. tic analysis, the linguist syntactically analyzed the sen- I get up every day at seven o’clock. tence deciding which word was connected to which word, using DGA annotator (Popescu, 2003), a pro- Notice that, even if ‘‘mit dem Bus’’ is removed in (1), gram based on Dependency Grammar (Hudson, 1990). the distance between bin and gefahren still is much lon- Both corpora had 10 chronologically ordered files, con- ger than 3. This way to consider connectivity is directly taining spontaneous speech of children. This way, their based upon the linearity a printed text imposes. By line- objective was to follow the child’s linguistic develop- arity, linguists understand that–in oral languages–one ment. Corominas-Murtra and collaborators saw that in word comes after the other, due to the constraints of both cases the ability to combine words represented by the oral channel. In this case, one counts the ‘‘posi- the network developed similarly. In that work, two dif- tions’’ and hence the distance between words. This type ferent stages were detected (a phase characterized by of approach has two sets of main problems: on one tree-like networks and another characterized by scale- hand, it is built on the idea of counting word positions free networks with the small-world characteristic), but as has been argued repeatedly, natural language which were separated by an abrupt transition. That does not seem to count (see i.a.Chomsky, 1975; Moro, transition coincided in time with the emergence of 2008; Newmeyer, 2005), and on the other hand, it is hubs, which interestingly corresponded to functional blind to a core property of natural language which is words, such as ‘‘a,’’ ‘‘the,’’ ‘‘that,’’ and ‘‘this’’ and modal its possibility to establish long-distance syntactic rela- verbs. Such words count with a strong morpho- tions (such as the relation between the auxiliary bin and syntactic load, though they are devoid of a rich denota- the lexical verb gefahren in (1) or the relation between tional semantics. These results are in consonance with the particle and the verb of the Dutch particle verb results from dynamic approaches, showing the rele- opstaan in (2), or the relation between the interrogative vance and non-linear emergence of functional words in direct object ‘‘what’’ and its corresponding verb ‘‘eat’’ general (Ruhland & van Geert, 1998), the particular in an English sentence such as ‘‘What do you normally use of prepositions (van Dijk & van Geert, 2007), or like to eat?’’). However, although it is true that co- determiners (Bassano et al., 2011). occurrence links are not based on true relations but on Very similar results were found by Barceló-Coblijn distance between words, these networks have anyway et al. (2012), who analyzed three longitudinal corpora captured some aspects that have been also found in of Spanish, Dutch, and German. In that study, adopt- syntactic networks, like the central role played by func- ing the perspective of dynamic systems, particular tional items as hubs (Barceló-Coblijn, Corominas- attention was paid to the emergence of determiners as Murtra, & Gomila, 2012; Corominas-Murtra et al., hubs in all three languages, showing that morphologi- 2009). Network-based approaches have evolved to be cal diversity needs time to be mastered (see, in particu- able to capture these particularities, as we will see next. lar, Table 5 on the analysis of German determiners). Another relevant difference was that, while Corominas- Murtra et al. (2009) included corpora with 10 files, 3.1. Longitudinal syntactic networks Barceló-Coblijn et al. (2012) analyzed corpora includ- We have already mentioned longitudinal network- ing up to 17 files. The inclusion of more segments of based approaches to language development which are the child’s first months of life apparently uncovered a focused on semantics. But there are also some studies third phase. Interestingly, all three children underwent arguing that the same can be done in syntax. An out- the same three stadia in very similar periods to those standing example is van Geert’s approach to language found in French by the Labrell et al.’s (2014) study development based on dynamic theory (Ruhland & van based on dynamic theory. Geert, 1998; van Geert, 1994). Van Geert’s approach to In a nutshell, complex networks offer us a very use- longitudinal corpora shows that the language capacity ful way for the representation of the ontogeny of this develops non-linearly. Also in a series of works Ninio process; words (syntactic items) are represented by (2006) has uncovered non-linear effects in syntactic nodes, while edges between nodes represent relations of development. Likewise, Corominas-Murtra et al. syntactic dependencies. In this sense, the analysis that 312 Adaptive Behavior 27(5) we will present in the following is similar to that of different words in different sessions), but also a differ- Sagae et al. (2010), since both use dependency relation- ent number of words (child’s lexicon grows, and her ships to syntactically analyze a sentence. ability to combine words becomes more developed). The present article also adopts the perspective of Here, networks are not always the same in number, and considering the cognitive capacity of language as a hence the approach is rather different, since what we dynamic, developing system. This study seeks to con- observe is not one network in comparison to another tribute to this trend of analysis by enlarging the ana- made of the same elements. The focus of our attention lyzed corpora to include several Romance and isolate is how much ‘‘linguistic signal’’ the speaker produces in languages that were unexplored with any network tech- a limited amount of time (a session of spontaneous nique so far (thus allowing typological (crosslinguistic) speech of a limited number of minutes), and what type comparisons), and also by assessing the suitability of of complexity can be observed in its underlying linguis- the network technique in clinical studies by analyzing tic system. If all TD children produce more or less the the patterns emerging from atypical populations (DS). same linguistic pattern, it will be possible to say that this might be the typical linguistic pattern of develop- ment. However, if a speaker greatly differs from the rest 4. Comparing different languages and TD of her pairs, this opens the possibility for considering and AD populations her linguistic ability as atypically developed. Hence, in the present approach we do not put the same quantity Van Geert (2004) made an appealing suggestion of of text from the transcriptions (we cannot), since the set applying the dynamic systems perspective of language of AD children differ from TD children precisely in the development to Specific Language Impairment or SLI. quantity and quality of speech. Note that including, In this article, like in Barceló-Coblijn et al. (2012), we say, only 100 lines from a recorded session in both typi- also conceive language ontogeny from the perspective cal and atypical networks would mask the fact that per- of dynamic systems. Specifically, we propose a new haps atypically developed speakers need much more implementation of the network analysis with a com- time until they produce the same quantity of words and parative analysis of languages from the Romance fam- sentences than TD children. It will be shown that, in ily (Italian, French, Catalan, and Spanish), the general, and coinciding with previous results in the lit- Germanic family (English, Dutch, and German), and a erature, DS speakers produce less words. Moreover, we language isolate (Basque). A second analysis will con- will show that DS speakers produce less syntactic con- cern two groups of people with DS, 12 English speakers nections and, hence, produce atypical syntactic net- and 20 Dutch speakers. By comparing between differ- works, in spite of having a similar recording session ent speakers of different languages it should be possible length (other differences found relate to the nature of to (1) see whether the type of language has an influence lexical hubs). on the ability to combine words (similarly to Bassano In the first part of our study, we analyze the syntac- et al., 2011). By comparing between different groups of tic development of three children acquiring a (first) speakers (TD vs AD), it should be possible to (2) rein- Romance language: Italian, French (Demuth & force or to discard potential hypotheses about language Tremblay, 2008; Tonelli, Dressler, & Romano, 1995), development like those presented in previous sections. and Catalan (from the corpus Serra-Solé), and then we In van Geert’s (2004) words, if typical development fol- compare them to the German, Dutch, and Spanish cor- lows some general principles (like discontinuity, U- pora in Barceló-Coblijn et al. (2012). Moreover, to shape, S-shape, particular network structures, etc.), overcome a possible Indo-European or typological bias there is no reason that they should not be applied to we also analyze two corpora of Basque (Soto Valle & the development of impairment and disorder, as van Alonso, 2015)—a language isolate. This is the first time Geert (2004) did with SLI. that such a language is analyzed with this technique. Such comparison also shows the welcome conse- The inclusion of Basque allows us to overcome two quences of the application of the network analysis in potential issues: first, that the results we obtain could the crosslinguistic study of language development in be biased by the phylogenetic relationships between the both typical and atypical populations. Typically, languages under study (Romance and Germanic lan- network-based approaches tend to compare control guages belong to the same Indo-European family) and networks in size, so that it is possible to compare differ- second, that this technique is only useful for non- ent systems made of the same number of elements. agglutinative languages. In general, although there exist However, in the case of language ontogeny, such an important differences, Romance and Germanic lan- option loses part of the ecological value of the original guages share a range of typological properties; Basque, source. Analyzing language ontogeny with real data however, is very different broadly speaking (see, for (spontaneous speech) shows an important difference: instance, the World Atlas of Language Structures 3 each period of the child’s lifetime that is analyzed pro- (WALS) database). Basque is unique within the group duces a different graph with different nodes (people say because it is the only language that has a basic SOV Barceló-Coblijn et al. 313 word order and agglutinative properties (see i.a.Laka, Fodorian ‘‘reverse compositionality’’ (see Fodor, 1998; 5 1996; Trask, 1998). It displays a complex morphologi- Fodor & Lepore, 2001)) whereby a learner who has cal system that exploits a lexical process called ‘‘aggluti- mastered the availability of syntactic combinations nation’’ whereby words are formed by joining affix such as [a^b], [a^g], [a^d], and [e^b] and [e^g] (where morphemes to the stem, where to a high degree each a and e pertain to the same syntactic category (say, morpheme contributes one meaning and vice-versa; determiners), and b, g, and d to a different one (say, that is, each affix is a bound morpheme for one unit of nouns)) generalizes in terms of their syntactic category meaning. For example, in Basque ‘‘etxe’’ (house) and goes on to freely combining elements of the {a, e} becomes by affixation ‘‘etxea’’ (the house), ‘‘etxeak’’ (the type and elements of the {b, g, d} type generating novel houses), ‘‘etxeko’’ (from house/homemade), and ‘‘etxe- combinations such as [e^d] and so on. In a nutshell, the tik’’ (from the house/out of house). In Basque it is pos- network reveals the cohesion level of the child’s linguis- sible to agglutinatively chain postpositions, obtaining tic system. In addition, mastering of syntactic combina- clusters of suffixes and words with a complex morphol- tion is related to a proper acquisition of functional ogy and a complex meaning, for example, ‘‘etxe-ra-ko- words: as we said, the most connected nodes of the net- ak’’ (etxe + ra (allative) + ko (genitive locative) + ak work (the ‘‘hubs’’) are typically functional words and (plural determiner); ‘‘those for the house’’). Likewise, the set appears abruptly at this developmental moment. ‘‘gizon’’ (man) can participate in the generation of com- In sum, our results suggest that the three levels, com- plex words like ‘‘gizon-a-ren-a-gatik’’ (gizon + a (deter- mon in all longitudinal analyses, are a reflection of the miner.singular) + ren (genitive) +a developmental path of TD children. (determiner.singular) + gatik (because); ‘‘because of 4 that of the man.’’) Importantly, this typological fact makes word-based MLU-type measurements (MLU-w) 5. Method non-optimal for crosslinguistic comparison, for exam- 5.1. Typical corpora ple, the number or ‘‘words’’ that are required for the expression of the same sentence varies significantly In the first part of the present work, five sets of corpora from language to language. As a consequence, some of longitudinal studies of typical language acquisition researchers prefer the use of morphemes as the count- have been analyzed: one for Catalan, one for French, ing unit for MLU (see Ezeizabarrena & Garcia one for Italian, and two for Basque. Corpora were Fernandez, 2018, for discussion). In our approach, on obtained from the international CHILDES database the contrary, we decompose each sentence in each syn- (MacWhinney, 2000). Each corpus contains at least 17 tactic unit, separating adpositions from nominals and files chronologically ordered. Each file corresponds to thus obtaining a much more accurate crosslinguistic a different recording session, which typically lasted comparison. between 30 and 60 min of spontaneous linguistic inter- We will show that the results of our linguistic analy- action between the child and their parents/caregivers. ses of TD children acquiring Catalan, French, or Children’s age extends from 17 to 32 months of age. Italian are fully convergent with previous works that Age is typically indicated as follows: ‘‘years;months.- took into account the perspective of dynamic systems days,’’ for example 2;06.25 means 2 years, 6 months and complexity. Despite procedural differences in the and 25 days of life. Age has then been represented in original composition of the corpora and the heteroge- averaged calculation. neity of the data, a robust common pattern seems to The Basque corpora we employed are partially dif- emerge, apparently characterized by three stages of lin- ferent from the rest, since we could not find corpora ear development separated by two non-linear phase extending exactly for the same age periods as the other transitions. In particular, it is intriguing that around ones; in fact, there were just these two cases available 28 months of age a small-world network emerges, with at the CHILDES database. Both Basque corpora start a ratio of word/edges close to 1:2. Such is the case in at 2.6 years of life (30 months). To see the connectivity each corpus, regardless of the language analyzed (this of Basque postpositions, we syntactically decomposed fact could suggest that the child has reached a particu- complex morpho-phonological words and we analyzed lar developmental threshold, and that the pattern of how many times a postposition appears, taking into word combination here represented by a small-world account that repetitions of the same combinations of network was not dependent of the type of language the lexical items are excluded. For example, if the target is child acquired). This type of network indicates in all the postposition -tik (equivalent to the preposition cases the maturity of the cognitive capacity for combin- ‘‘from’’ in English) and it appears in the text 10 times ing words. It reveals a stage where the child has already with the noun etxe (‘‘house’’), this counts as only one discovered the unbounded combinatoriability of the appearance type (10 tokens of the same type), but if it syntactic items in her language via some sort of syntac- also appears with herri (‘‘town’’), then it counts as two tic ‘‘reverse combinatoriality’’ (in the spirit of the appearance types. 314 Adaptive Behavior 27(5) 5.2. Atypical corpora The first of these graphs corresponds to a date of the The set of clinical corpora are also obtained from child’s life of around 18 months, and the last graph cor- responds to about the 30th month. This way, the evolu- CHILDES. The first set contains data of 12 subjects of tion through time of the network—that represents how different age, all of them diagnosed with DS and speak- words are combined—can be readily visualized. In the ing English (Rondal, 1978a, 1978b). There are two files case of Basque, two small corpora (representing two per individual, with the same age regarding ‘‘year’’ and children) of five files each have been analyzed. ‘‘month,’’ and no information regarding the ‘‘day.’’ We One graph was obtained for each session (i.e. 17 decided to average the values of both files to get the graphs per child, 5 in the Basque cases). Each word in final score. The second set contains data of 20 subjects a graph corresponds to a node in the network, while of different ages affected by DS and speaking Dutch edges reproduce the syntactic dependencies between (Bol & Kuiken, 1990). However, these files correspond words. It is important to keep in mind that the depen- to one session; hence, just one network for each partici- dency arrow reproduced by Cytoscape respects the pant could be construed in this case. The group is het- direction of the arrow created by the DGA annotator. erogeneous regarding ages, ranging from 3 to 18 years. Cytoscape also allowed to visualize the arrow point. Nevertheless, the analysis of a multiplicity of cases However, in spite of the careful linguistic analysis, (cross-sectional) of each type reveals a common pattern information about the kind of syntactic relationships which we would like to argue is characteristic of the could not be integrated. Therefore, since it was not pos- impairment in question (see section 6.2). sible to see the kind of link between nodes, it was decided to be cautious and conservative and the analy- 5.3. Procedure sis of edges was made as undirected edges. We would like to stress that our approach is not fre- A team of linguists analyzed each corpus sentence by quentist, but a different, complementary way of obser- sentence, following the basic guidelines of Dependency ving linguistic data. In a graph, one can see how many Grammar (Hudson, 1990). This kind of grammatical different words and how many different dependency analysis is known and used in computational relationships have been employed by the child. This approaches to language because it provides essential fact implies that repetitions—of either lexical items or information about the syntactic structure of the sen- dependency relationships—are not taken into account tence, and the coded sentences are easy to export for in this approach. If the child repeats ‘‘Pluto barks’’ 10 further analyses to a network-analysis program. To times, it counts as only one type of dependency relation: carry out the syntactic analysis, data were prepared fol- one node for ‘‘Pluto,’’ one node for ‘‘barks,’’ and a lowing the same procedure as in previous studies on dependency relationship from ‘‘Pluto’’ to ‘‘barks.’’ It is syntactic networks (Figure 1; Barceló-Coblijn et al. the breath of vocabulary and the word combinations 2012; Corominas-Murtra et al., 2009). effectively produced which are used as a measure of lin- Corpora were downloaded from the CHILDES guistic development (i.e. we measure the knowledge database, each file was converted into .xml so that it that a speaker has obtained of the combinatorial prop- could be read by the syntactic-annotation program, by erties of the items in her lexicon, as represented in her means of a Perl script. The script also filtered the file, speech). Hence, it is the number of different linguistic pruning the data that were not produced by the child expressions—and not the number of productions—the (e.g. comments of the transcriber, phonetic transcrip- focus of this study. However, it is true that the high tions, what the parents say, etc.). The new .xml file was number of connections of the most connected words— then sent to DGA annotator (Popescu, 2003), a program the hubs—also reflects their condition as the most used specifically created for syntactic analysis by means of words. dependency relationships. In a graph one can see several disconnected net- The analysis was carried out by a linguist manually works, depending on how well connected the words line by line, sentence by sentence, determining the syn- were. Each piece is called a ‘‘connected component.’’ tactic dependency relationships between words by The typical graph of a network analysis of a corpus means of a dependency arrow. Other criteria followed contains one large network (sometimes two), several by the linguists’ team are gathered in a public protocol small networks with two or three nodes, and a number (Corominas-Murtra, 2007) applied in previous studies. of isolated nodes (words that were uttered in isolation After analyzing all files, another Python script spe- and hence were not connected with other words, such cifically created for this kind of studies turned them into as, for example, ‘‘no’’ in a plain denial answer, or ‘‘cat’’ .txt files, which is a format accepted by the network- when pointing to a cat). analysis program Cytoscape (Shannon et al., 2003). Cytoscape (or any other network program) also Once Cytoscape reads the data, it produces a graph. makes it possible to calculate multiple mathematical In this study at least 17 graphs were obtained for each measures for each network; the most relevant for our typical child acquiring an Indo-European language. purposes are Barceló-Coblijn et al. 315 Figure 1. Steps and procedures in order to obtain a syntactic network from a corpus of spontaneous speech. First, the corpus of spontaneous speech is created (on the basis of a spontaneous speech session; in this work, corpora have been dowloaded from CHILDES database). Then each corpus is analyzed (word by word) with the DGA annotator, by establishing the syntactic relations between words. Once the corpus is analyzed, it is adapted by a script in order to be readable by the network program. Finally, the network program reads the annotated corpus and reflects the data as a network, where nodes are words and edges are syntactic relations. 1. the relative size of the largest connected component observed in German, Dutch, and Spanish (Barceló- (or LCC), which indicates the size of the largest com- Coblijn et al. 2012). Thus, so far, all TD children exhibit ponent regarding the whole graph; the same pattern of syntactic development, regardless of 2. the average clustering coefficient (C), which indi- the particular language they are learning. This pattern is cates the extent to which two connected nodes are apparently characterized by three different phases: also connected to a third one, creating a triangle (this is a real number between 0 and 1 (when there is 1. In the first phase, the networks emerging from their no clustering the value is 0, while 1 indicates maxi- speech have a tree-like structure, characterized by a mal clustering)); higher number of nodes (words) than edges (word 3. the average shortest path length (L), which indicates combinations), and which appears with a relatively the mean number of intermediate nodes that connect small LCC; any two of them; and finally 2. In the second phase, the child’s pattern is character- 4. the average degree of the LCC (\k . ), which refers ized by networks in which edges overcome the num- to the average number of connections per node; in ber of words and, as expected, some hubs appear our study, it indicates the average number of depen- (these are ‘‘promiscuous’’ words that combine with dency relations for each particular word. several others, thus providing cohesion to the net- work, and increasing the relative size of the LCC); 6. Results 3. In the third phase, a final phase, the LCC almost equals the whole graph. The characteristic path 6.1. Typical development length decreases and a high clustering appears, The study of TD corpora by analyzing longitudinal cor- showing a strong cohesion in the network. An inter- pora of different languages aims to answer the follow- esting aspect is that at this phase the tendency is ing questions: (1) are there differences between German evolving to a ratio of nodes (words)/edges (syntactic and Romance languages? and (2) what would happen links) of 1:2, that is, two connections for each word. in the case of typologically different languages? Once analyzed and compared, the three new corpora Interestingly, that final topology is reached more or (Catalan, Italian, and French) show a similar develop- less around the same age, 800 days of life, in spite of mental pattern (Table 1 and Figure 2) to the one the heterogeneity of the original data in terms of 316 Adaptive Behavior 27(5) Table 1. Values of the indicators extracted from the syntactic networks from the TD corpora: mean path length (L), average degree of the giant connected component or GCC \k., clustering coefficient (C), relative size of the GCC (R size), number of nodes of the GCC, number of edges of the GCC, and ratio nodes/edges. Age days L \k. C R size Nodes GCC Edges GCC Ratio nodes/edges Catalan 593 1 1 0 0.25 2 1 2 613 0 0 0 0.037037037 1 0 – 634 2.894 1.8 0 0.27027027 10 9 1.11111111111111 640 1.364 1.5 0 0.19047619 4 3 1.33333333333333 678 2.626 2 0 0.20289855 14 14 1 712 1.893 1.6 0.091 0.166666667 6 5 1.2 785 1.821 2 0.161 0.25 7 7 1 799 4.145 2.44 0.022 0.43269231 45 55 0.818181818181818 879 3.292 2.788 0.038 0.44594595 66 93 0.709677419354839 938 3.223 2.565 0.067 0.73015873 46 61 0.754098360655738 976 3.46 3.692 0.076 0.79194631 240 449 0.534521158129176 1022 3.78 3.176 0.045 0.83783784 187 298 0.62751677852349 1067 3.827 2.807 0.029 0.8071066 161 229 0.703056768558952 1127 5.111 2.133 0.006 0.61971831 45 48 0.9375 1265 4.207 2.883 0.011 0.76243094 145 209 0.69377990430622 1309 3.896 3.4 0.057 0.75 180 311 0.578778135048232 1405 3.647 3.827 0.038 0.78059072 185 356 0.519662921348315 French 552 1 1 0 0.111111111 2 1 2 571 0 0 0 0 1 0 – 584.5 0 0 0 0 1 0 – 600.5 0 0 0 0 1 0 – 621 0 0 0 0 1 0 – 662.5 0 0 0 0 1 0 – 694 0 0 0 0 1 0 – 728.5 3.106 2.424 0.025 0.402439024390244 33 41 0.804878048780488 748 2.426 2 0.058 0.25 10 10 1 819 2.928 2.627 0.073 0.377777777777778 51 69 0.739130434782609 832.5 3.28 2.769 0.046 0.456140350877193 52 77 0.675324675324675 846.5 3.46 2.571 0.038 0.563758389261745 84 110 0.763636363636364 861 3.208 2.867 0.072 0.535714285714286 60 89 0.674157303370787 875 3.33 2.831 0.073 0.550847457627119 65 95 0.68421052631579 895.5 3.238 3.25 0.077 0.618181818181818 136 231 0.588744588744589 909.5 3.486 3.41 0.093 0.674033149171271 122 216 0.564814814814815 Italian 522.5 1.4 1.5 0 0.042105263157895 4 4 1 536.5 1.977 1.778 0 0.1 9 8 1.125 551 1.45 1.6 0 0.027472527472528 5 4 1.25 571 3.421 2.0625 0 0.228571428571429 32 33 0.96969696969697 585.5 3.031 2.095 0.006 0.106598984771574 21 22 0.954545454545455 598.5 2.012 1.846 0 0.053719008264463 13 12 1.08333333333333 613 3.231 2.074 0.006 0.094405594405594 27 28 0.964285714285714 627 1.167 1.333 0 0.090909090909091 3 2 1.5 641.5 1.711 1.667 0 0.024 6 5 1.2 655.5 4.148 1.917 0 0.07843137254902 24 23 1.04347826086957 669.5 2.268 1.882 0 0.057046979865772 17 16 1.0625 683 4.885 2.08 0 0.152439024390244 50 52 0.961538461538462 697 4.932 2.078 0.016 0.184115523465704 51 53 0.962264150943396 717.5 4.881 2.174 0.009 0.347432024169184 115 126 0.912698412698413 732 5.13 2.4 0.009 0.421286031042129 190 228 0.833333333333333 746 5.346 2.312 0.015 0.495176848874598 154 180 0.855555555555556 759 4.578 2.486 0.016 0.615 245 314 0.780254777070064 773.5 4.145 2.764 0.027 0.715492957746479 254 355 0.715492957746479 789.5 4.446 2.426 0.019 0.622516556291391 94 114 0.824561403508772 794 4.036 2.905 0.022 0.705093833780161 263 383 0.68668407310705 825.5 4.692 2.525 0.004 0.705202312138728 244 311 0.784565916398714 838.5 4.477 2.788 0.03 0.77521613832853 269 379 0.70976253298153 852.5 3.955 2.994 0.034 0.772234273318872 356 538 0.661710037174721 867 4.266 2.719 0.022 0.744471744471744 303 413 0.73365617433414 (continued) Barceló-Coblijn et al. 317 Table 1. Continued Age days L \k. C R size Nodes GCC Edges GCC Ratio nodes/edges Basque A 939 6.022 2.065 0.002 0.442857142857143 62 65 0.953846153846154 952.5 2.667 1.882 0 0.129770992366412 17 16 1.0625 967.5 3.494 2 0 0.184873949579832 22 22 1 981 3.767 1.92 0 0.233644859813084 25 24 1.04166666666667 1015.5 3.866 3.029 0.082 0.73015873015873 138 220 0.627272727272727 1029.5 4.571 2.133 0.015 0.441176470588235 60 65 0.923076923076923 Basque M 935 3.959 2.458 0.023 0.588652482269503 83 105 0.79047619047619 949.5 3.974 2.364 0.015 0.582781456953642 88 106 0.830188679245283 963.5 4.194 2 0.002 0.367647058823529 50 50 1 984 3.829 2.286 0.021 0.538461538461538 42 51 0.823529411764706 998 5.704 2.127 0.003 0.508064516129032 63 67 0.940298507462687 1025.5 4.665 2.452 0.023 0.67906976744186 146 179 0.815642458100559 TD: typically developing. Figure 2. Summary of results for the relevant measures of the resulting networks for eight typically developing individuals. Each point displayed represents a single moment and not a mean score. We have integrated the five new corpora of this study with our previous results, in German, Dutch, and Spanish (Barceló-Coblijn, Corominas-Murtra, & Gomila, 2012). The same developmental pattern was found by Corominas-Murtra, Valverde, and Solé (2009), with two English learners. sampling time, procedures, periodicity, and so on. The of the evolution of hubs in Dutch and for a sample of Basque cases (shorter in number of recordings) show a the evolution of connectivity of determiners in slight delay, in part, due to the lack of data from previ- Catalan), a result that coincides with the previous ous ages, but it could also be related to its more agglu- results. This pattern is found in all cases late and tinative character. Nevertheless, networks with a high abruptly, coinciding with an important change in the clustering and a low path length are also obtained in network topology (see Table 1 in comparison to Table this language. 5). Determiners (e.g. ‘‘the,’’ ‘‘a’’) emerge as hubs in all Furthermore, the hubs that abruptly give rise to the languages so far analyzed following this late, abrupt small-world networks tend to belong to the same lexical transition pattern. In the previous work on networks in class: these are the so-called functional words (mostly English (Corominas-Murtra et al., 2009), the hubs were determiners; see Supplementary Material for a sample ‘‘the,’’ ‘‘a,’’ ‘‘that,’’ and ‘‘it.’’ These nodes have at 318 Adaptive Behavior 27(5) -tzeko(14) Under each suffix form, a number within parenthesis refers to the class of suffix: (0) bare form with no morphological suffix; (1) singular article, absolutive; (2) plural article, absolutive; (3) ergative; (4) possessive; 30 months of life between 30 and 40 links, almost 80 in (5) comitative; (6) inessive; (7) allative; (8) ablative; (9) auxiliary verb; (10) suffix for the generation of manner adverbs, akin to English ‘‘-ly’’; (11) result-denoting suffix (akin to English ‘‘-ed’’); (12) participial; (13) the case of ‘‘a.’’ Articles in Dutch, German, and 1 Spanish also obtained similar numbers, producing around 30 and 40 syntactic dependency links. Some language-specific differences are always expected: the -ten/-tzen(13) use of ‘‘it’’ in English does not correspond with the use of the equivalents in Dutch (het) or in German (es). Spanish and Catalan have three and two demonstra- tives respectively (some Catalan dialects show three 5 demonstratives, but they are not represented in the -tu(12) sample), while Dutch and English have two and German has only one. However, German has a declen- 2 sion system, complicating a little bit the proper use of each form. The Romance languages here analyzed, -ta(11) Italian, French, and Catalan, follow the very same pat- tern, which is the pattern previously observed in 3 Spanish. One of the benefits of making a longitudinal analysis -ik(10) is that one can observe the evolution of the graph over 1 a long period of time. As said, each graph contains sev- eral ‘‘components’’ (or networks of one or more nodes). A. V.(9) Interestingly, in all languages analyzed until now (Catalan, Italian, French, Spanish, Basque, German, 39 English, and Dutch) the number and size of compo- nents behaved in a very similar way: when the child -dik(8) began to speak (tree-like network stage), there were 1 many small components of two and three nodes, and imperfect; (14) future. The bottom row indicates the number of other nodes each of the suffixes combines with. Table 2. Affixes in Basque from the Maider corpus (each one is treated as an independent node). sometimes a couple of larger networks of four or five -ra(7) nodes (plus a large number of isolated nodes, represent- ing unconnected words which were uttered in isolation). 1 As the child grew and developed her linguistic ability to -n(6) combine words, we observed the emergence of a LCC 14 of a dozen or more nodes, sometimes followed by one or two small networks of five to seven nodes (plus the -(r)(e)kin(5) mini networks of one, two, or three nodes). When the child got to the second phase, small networks larger than four nodes seldom appeared. Instead, there was a clear and dominant LCC, and a number of networks of 1 one and two nodes, followed by a smaller number of -(r)en(4) components with three nodes. Finally, when children got to the last stage, we observed an enormous LCC. In this stage, there was no independent network larger 8 than three nodes, and even the presence of three-node networks is not guaranteed. -k(3) 10 The second main question that concerned typical development and network analysis is related to the agglutinative features of languages like Basque, and -(a)k(2) whether this represents an obstacle for its application. 6 From the two corpora of Basque that we analyzed, we obtained networks very similar to those of the non- -a(1) agglutinative languages. We also performed a separate 19 analysis of postpositions to see their connectivity pat- tern (Table 2). Bare form(0) Even in Basque, where determiners can be both free lexical units (e.g. bat ‘‘one/a,’’ as in etxe bat ‘‘a house’’) or suffixes attached to the nominal stem (e.g. -a like in 73 etxea, ‘‘the house’’), these elements play the role of Barceló-Coblijn et al. 319 Table 3. Number of links of the determiners and other functional words in English, from networks of subjects affected by DS. Age, months Subject MLU a that the it there and I 54 Abby 1.55 9 7 4 0 2.5 0 0 38 Baxter 1.04 2 0.5 1.5 1 1.5 0 0 56 Bob 1.72 4.5 2.5 4 2.5 0 1 2 57 Cheryl 1.79 5 2.5 3.5 3.5 1.5 2.5 11 84 Carrie 2.09 8.5 2.5 11.5 3.5 0 4 7.5 56 Dan 1.96 19 11 17 8.5 2.5 2.5 9.5 55 Janet 1.75 13 5 5 8 3 2.5 3.5 52 Kevin 1.12 3 2.5 0 0 0 0 0.5 146 Mat 2.93 19.5 4 30.5 2.5 1 24.5 13.5 100 Rhoda 2.70 13.5 12 17 5 2 10.5 8 37 Selma 1.11 0 0.5 0 0 0 0 0 94 Silvia 2.03 15 14 15 2.5 1 1.5 11.5 MLU: Mean Length of Utterance. In this table the number of links of the typical hubs in English has been summarized. In the first study with two 30-month-old unimpaired children, hubs where ‘‘a’’, ‘‘the’’, ‘‘that,’’ and ‘‘it’’ and the scores reached between the 30 and 40 links. Instead, in DS cases of English the scores are much reduced. In each case, values come from averaging the scores of two networks (see Supplementary Material). Table 4. Number of links of the determiners and other functional words in Dutch, from networks of subjects affected by DS. Age, months Individual MLU De Het hetpron een Deze Dit Dat Die En Ik the the it a this this that that and I 196 Anita 3.3 2 0 0 5 0 0 1 3 4 3 125 Antoniette 3.2 7 2 4 10 0 0 2 1 4 7 221 Bert 2.4 6 0 0 0 0 0 0 0 4 11 182 Chantal 2.8 7 0 0 0 1 1 2 4 2 2 181 Eline 2.5 12 2 0 7 1 0 0 1 25 3 219 Erik 1.8 2 1 0 3 0 0 0 1 3 1 162 Fiona 3.4 13 0 4 9 0 1 6 3 7 14 156 Iris 3.0 0 0 1 4 1 1 3 2 10 12 129 John 2.0 6 0 0 3 0 0 1 1 7 1 187 Katrijn 1.9 7 3 2 6 0 0 1 0 1 2 165 Lisa 1.8 6 2 0 10 0 0 1 2 2 2 202 Marcel 1.2 0 0 0 1 5 0 1 0 1 0 98 Martijn 4.7 9 3 3 6 0 0 3 10 3 1 227 Marius 3.3 8 2 1 7 2 1 1 5 10 8 205 Michael 3.0 17 0 2 8 0 0 1 1 3 2 100 Pieter 3.5 14 2 3 7 2 1 2 5 21 7 100 Ronnie 5.2 14 0 3 21 0 0 8 7 16 17 224 Sanne 3.0 4 0 2 2 0 0 4 2 6 10 200 Sven 2.4 12 2 0 7 3 0 3 1 6 2 53 Tineke 2.4 0 2 0 5 5 0 3 3 6 0 MLU: Mean Length of Utterance. In a single conversation in Dutch, a 900-day-old individual typically produces about 30 links for each of the hubs that are determiners (cf. Barceló- Coblijn, Corominas-Murtra, & Gomila, 2012). hubs in the small-world network. Table 2 displays the is divided into two subgroups, depending on the lan- patterns of connectivity of different suffixes showing guage they spoke, Dutch or English. The networks that that determiners are much more connected than characterize the syntactic performance reached by DS postpositions. children of older ages are notably dissimilar. Results are shown in Tables 3–5. English DS children became able to produce a very low number of words and to 6.2. Atypical development barely combine them (the speaker ‘‘Mat’’ stands out in The third main question was whether there are differ- comparison to the group, though he is still far from ences between typically and atypically developed chil- TD younger speakers; see Figure 3). dren. As expected, the answer is positive. One large Then, the number of words increased over time, but group of 32 speakers with DS was analyzed. The group very slowly, given the ages of the speakers. In the case 320 Adaptive Behavior 27(5) Table 5. Information about the Dutch and English DS speaker’s networks. Months Nodes Edges Ratio N/E Clustering Path length Rel_Size \k. Dutch Ronnie 196 223 356 0.6264044944 0.042 4.039 0.8479087452 3.175 Fiona 125 150 236 0.6355932203 0.034 3.907 0.7075471698 3.107 Marius 221 130 194 0.6701030928 0.025 4.083 0.6914893617 2.969 Pieter 182 128 165 0.7757575758 0.027 4.843 0.6844919786 2.5625 Iris 181 119 158 0.753164557 0.034 4.814 0.650273224 2.639 Michael 219 116 138 0.8405797101 0.033 5.737 0.5979381443 2.379 Antoniette 162 98 133 0.7368421053 0.061 4.407 0.5414364641 2.673 Sanne 156 110 126 0.873015873 0.028 5.217 0.597826087 2.273 Sven 129 72 84 0.8571428571 0.002 4.68 0.5217391304 2.306 Eline 187 74 80 0.925 0.068 3.351 0.4431137725 2.135 Bert 165 62 77 0.8051948052 0.05 4.141 0.4881889764 2.484 Anita 202 57 61 0.9344262295 0 5.085 0.5643564356 2.14 John 98 47 56 0.8392857143 0.128 4.075 0.4234234234 2.298 Katrijn 227 54 56 0.9642857143 0.06 6.361 0.357615894 2.074 Martijn 205 45 56 0.8035714286 0.015 4.017 0.6 2.4 Tineke 100 51 56 0.9107142857 0 4.69 0.504950495 2.196 Chantal 100 41 42 0.9761904762 0.013 3.979 0.4270833333 2.049 Lisa 224 35 37 0.9459459459 0 3.376 0.4117647059 2.114 Erik 200 10 9 1.1111111111 0 2.533 0.0862068966 1.8 Marcel 53 6 5 1.2 0 1.867 0.0789473684 1.667 English Mat 146 199.5 263 0.7585551331 0.041 4.3885 0.772826087 2.6425 Silvia 94 116.5 178 0.654494382 0.025 4.3545 0.5907146524 2.4795 Dan 56 111.5 160 0.696875 0.0385 3.82 0.523478836 2.859 Rhoda 100 110.5 137 0.8065693431 0.0595 4.055 0.699741435 2.993 Cheryl 57 79 98.5 0.8020304569 0.051 4.184 0.5575624886 2.399 Carrie 84 70.5 77 0.9155844156 0.031 5.3375 0.4257745999 2.161 Abby 54 33.5 38.5 0.8701298701 0.0285 3.389 0.3578307728 2.3095 Janet 55 34.5 35.5 0.9718309859 0.0145 3.931 0.2365130219 2.0185 Bob 56 38.5 33.5 1.1492537313 0.0295 4.098 0.3779050736 2.093 Baxter 38 10 9 1.1111111111 0 2.8085 0.1423001949 1.6875 Kevin 52 8.5 8.5 1 0.0775 2.474 0.1214285714 1.6665 Selma 37 2.5 1.5 1.6666666667 0 1.1665 0.0275012444 1.1665 The table has been ordered on the number of edges/syntactic links produced by the speaker. Age has also been calculated in months. Rel_Size indicates the relative size of the giant connected component (GCC) or largest network within the graph; \k. indicates the averaged number of edges/syntactic links of the GCC; and ratio N/E indicates the averaged number of nodes/edges. of Dutch DS children, their vocabulary is still small but outcome of our study, which may go unnoticed to the there seems not to be an analogous difficulty for com- naked eye, but was uncovered by the network analysis. bining words syntactically (even though only one case Furthermore, it has been mentioned that an indica- reached a small-world topology), and in spite of possi- tor of the cohesion of networks is the clustering coeffi- ble morpho-syntactic deficits (Bol & Kuiken, 1990). cient (C). It is interesting to see whether networks that Interestingly, network measures clearly showed that are similar in the number of nodes and edges differ their respective developmental paths differ substantially regarding C, because this would be the clearest indica- from the one followed by neurotypicals. tor that there is a substantial difference in the cohesion Crucially, our data showed that up to eight DS of the network (see Table 5). For example, only one of speakers produced a larger number of different words the DS Dutch speakers, ‘‘Ronnie,’’ reached the thresh- than the neurotypical children (it has to be noted that old of the small-world characterized by a ratio of word/ our DS speakers are significantly older than our TD syntactic relations of 1:2. Does this mean that there is speakers) but the number of syntactic connections was no difference between this DS speaker’s linguistic com- noticeably smaller for DS speakers, and only three DS binatorial ability and that of TD speakers? Let’s com- speakers produced as many edges as TD younger chil- pare data: the networks of the TD Dutch speaker dren did. That is, vocabulary size does not directly map ‘‘Daan’’ at 926 days have the following values: onto higher network complexity: in our case TD chil- N = 223, E = 368, C = 0.063, while the DS Dutch dren with smaller vocabularies showed a richer com- Ronnie at the much older 5963 days reaches N = 223, mand of their combinatorial properties than AD E = 356, C = 0.036; thus, two networks quite similar children with larger vocabularies. This is a clear regarding the n of nodes and edges show a notable Barceló-Coblijn et al. 321 difference regarding the C. By contrast, the TD available for both groups of DS individuals in the German speaker ‘‘Simone’’ (in purple in Figure 3) CHILDES database and in the original works (Bol & reaches the DS Dutch Ronnie’s C with barely half of Kuiken, 1990; Rondal, 1978a, 1978b), we further inves- the elements: at 873 days (N = 139, E = 206, tigated the relationship between this measure and our C = 0.035). In other words, the 3-year-old TD basic ones: the number of nodes (words) and edges German speaker, with much less lexical elements, shows (dependency relationships) of the LCC, the largest net- a pattern of linguistic cohesion which is similar to the work in each graph (Figure 3). First of all, we checked pattern attested in the big DS linguistic network. These for normality and homoscedasticity of the data. Both empirical data confirm again what Watts and Strogatz normality and homoscedasticity were met. Then we (1998) theoretically proposed: in networks, nodes are wanted to know whether both MLU and linguistic net- important, of course, but connectivity is a crucial factor works measured the same linguistic aspect. Hence, a that can sharply distinguish between two apparently Pearson correlation was performed and it revealed a similar networks. Again, a network smaller in the num- similar level of correlation between MLU and both net- ber of nodes can be more cohesive (i.e. have a higher C work measures (r = .704 for nodes and r = .709 for and a lower L) than a network with a larger number of edges, in both cases with p \ .001), suggesting that nodes; the size matters, but is not the only important both measures keep track of the same growth process. indicator. In our study, we interpret this as suggesting An important question was whether the type of lan- that a more connected network shows a richer com- guage could affect the network analysis. For this rea- mand of the combinatorial properties of lexical items; a son, three ANOVAs (analyses of variance) were more disconnected one, on the other hand, seems to performed setting language as independent variable, treat combinations more as extensional exceptions than and MLU, nodes, and edges as dependent variables. as a general intensional rule that can be recursively Analyses showed that for MLU, the language variable applied. accounts for 27% of the differences between groups Summarizing, our network analysis proves to be a (F(1, 30) = 11.03; p = .002; h = .27; CI (h)95%: useful clinical tool and clearly shows the difference .025–.971). On the contrary, the type of language only between population groups, providing objective and accounts for 1.4% in the case of nodes (F(1, formal indicators of complexity in performance reached 30) = 0.44; p = .512; h = .014; CI 95%: .025–.487) by different individuals/groups. and 1.3% in the case of edges (F(1, 30) = 0.38; On a different note, the comparison of the hubs p = .542; h = .013; CI 95%: .025–.458)—even if the found provides also an important source of clinical latter are non-significant. This outcome suggests that information (Tables 3 and 4). While hubs in the atypi- complex networks can help overcoming the language- cals largely belong to the same functional class of hubs specific problems of morphology and closed-class in TD networks, the number of combinations in AD is words that the word-based MLU measurement faces, much smaller. Although it can be said that these are becoming a useful complementary methodology to still hubs, the number of links is considerably smaller assess the linguistic competence of DS speakers. in comparison to that of healthy individuals. English speakers affected by DS produced a few more than a dozen syntactic links with the word ‘‘the.’’ Only the 7. Discussion oldest individual (‘‘Mat’’) produced 30 combinations. In the case of ‘‘it,’’ this item is seldom produced (almost The present work has explored incipiently the dynami- disappeared) and it does not constitute any hub. In the city of language ontogeny in both typical and atypical study of DS in Dutch, results also show striking differ- conditions. This has been possible thanks to the combi- ences in the scores of hubs. The number of links is in nation of linguistic theory and network science. The the best case (‘‘Ronnie’’) half of that of TD children. sentences produced by the speakers have first been ana- However, there appears a different hub, represented by lyzed syntactically following syntactic theory: syntactic the word en (‘‘and’’), suggesting that these speakers items were detected and their dependency relationships, exploit coordination (parataxis) as a linguistic strategy if any, labeled with a dependency link. In contrast to maybe to avoid the complexities of structural embed- other network approaches based on word co-occur- ding (hypotaxis). Interestingly, the number of links of rence, here networks have been created on the basis of the set of demonstratives is severely reduced in both large collections of sentences, each of them syntactically DS groups, which in several cases amounts to zero analyzed by a linguist, word by word, item by item. links. See Table 3 for the DS performance in English, These analyses have been transmitted to a network pro- Table 4 for the Dutch cases, and Table 5 for the net- gram, which interpreted words as nodes and syntactic work values of DS syntactic networks. links as edges, creating a network from each file (Figure Finally, an additional, interesting question is 1). This procedure takes a considerable amount of time, whether or not this network-based technique has con- though we think it is worth it, given the results we vergent validity. Given that the MLU measure was obtained, which uncovered a pattern of crosslinguistic 322 Adaptive Behavior 27(5) Figure 3. Number nodes (top) and edges (bottom) produced by typical and atypical individuals. Age is presented in days. Typical individuals (on the left of each graphic) are represented by three linked points, letting see their progress through time. Each point represents a different kind of network. The Down syndrome cases (12 English speakers represented by dots and 20 Dutch speakers represented by triangles) represent different individuals of different ages. unity and cross-population variation (we observed simi- paths in DS populations). Admittedly, the procedure lar developmental paths traversing the same phases at followed during the analyses of corpora could have similar ages across genetically and typologically differ- affected or influenced in the resultant structure (see ent languages, and radically diverging developmental Gruenenfelder & Pisoni, 2009). Barceló-Coblijn, Barceló-Coblijn et al. 323 Duguine, and Irurtzun (2019) analyzed twice the same traversed during ontogeny, and that the system changes language corpora of three languages (Basque, French, when important, structural elements are incorporated, and Dutch), adopting different linguistic theoretical filling structural gaps, has already been put forth in hypotheses each time (related to the role of determiners recent works on semantics and lexicon ontogeny (Stella, as governors or as dependent elements of nouns). Beckage, & Brede, 2017). More recently, Stella et al. Results clearly show that the same corpus produces two (2018) have shown how the lexicon could evolve during different networks and that the degree of edges of deter- ontogeny through an abrupt transition at age 7, and miners is severely altered depending on the adopted that the key of that developmental abruptness is the hypothesis. However, clustering (C) was also affected in acquisition of the core set of vocabulary. Interestingly, an important way. In the present work, we have the results of network-based approaches on both syntax adopted and applied always the same theoretical deci- and semantics suggest that the language capacity devel- sions to both typical and DS corpora. Differences ops following quite similar patterns regarding syntax between these two groups of children are clear, separat- and semantics, something that nicely fits Elisabeth ing typical children from DS children. The absence of Bates’ classic hypothesis about the ‘‘inseparability of mixed scores between typical and atypical reinforces grammar and the lexicon’’ and that ‘‘the emergence of our intuition that these first, explorative results could grammar is highly dependent upon vocabulary size’’ be on the right track. (Bates & Goodman, 1997). A preliminary conclusion is that the apparent exis- With respect to atypical development, our study tence of several developmental phases crosslinguisti- might have uncovered a pattern that is unobservable cally could imply that linguistic development is not a by other means even if the paucity of linguistic struc- process that follows either a simple linear or a non- tures in AD children is considered symptomatic and linear progress, but a ‘‘saltational’’ process that com- related to their atypical condition. The present bines both kinds of progresses (and thus overcoming approach provides a framework to change our concep- the black-or-white scenario mentioned above): when tion of AD: in agreement with previous findings, atypi- the system reaches a particular level, it takes some time cal cases are not just delayed stages in the course of until the next phase transition; during this period a pat- typical development. And in coincidence with previous tern of linear progress is also observed in the network works (e.g. Chapman, Hesketh & Kistler, 2002; growth. These results are in agreement with Labrell et Chapman, Sindberg, Bridge, Gigstead, & Hesketh, al. (2014) who found a developmental path character- 2006), a large degree of variability has been found ized by several spurts and plateaus and also with recent within the DS group. We would like to stress that sev- findings in the ontogeny of English syntactic networks eral DS children could reach a high score in their lexi- (Corominas-Murtra, Sànchez Fibla, Valverde, & cal production and that notwithstanding their syntactic Ricard Solé, 2018). However non-linearity is evident performance—their capability to syntactically combine when the network radically changes its topology, from lexical items—clearly differed from their typically one type of network into another of a completely dif- developed peers. This suggests that focusing on word ferent type. Until now, in the present work this kind of counting only could mask important information dual progression has been confirmed in seven longitudi- related to the individual’s language capacity. In our nal corpora of seven different languages. Although the study, most of the times the DS speakers did not reach group size is still too small for a statistic generalization, the TD speakers’ topology (only 1 speaker out of 32 and therefore the nature of the present work remains did, and at a significantly older age), nor their use and explorative, we think that our results suggest that this connectivity of functional words was the same, which is not happenstance. Interestingly, our results and the could be interpreted as being due to a specific difficulty results found in Barceló-Coblijn et al. (2012) have both in grasping the syntactic rules and mastering the combi- found non-linearity of developmental patterns, coincid- natoriability of functional, closed-class words in combi- ing with other works based on dynamic systems theory nation to the general difficulty to learn new vocabulary (Corominas-Murtra et al., 2018; Labrell et al. 2014; (Assen & Bol, 2012). This strongly suggests that the Ruhland & van Geert, 1998). connectivity in atypical cases like DS is essentially dif- Changes in topology coincide with the increase of ferent. Note that, for example, up to eight DS speakers the degree of functional words, as if they were they core have a number of nodes (words) larger than several of the syntactic capability. This is not something new, neurotypicals. The clearest example is the German TD as it has been pointed out in empirical works in linguis- speaker (‘‘Simone,’’ in purple, in Figure 3)—a child tics (Bassano et al., 2011; Ruhland & van Geert, 1998; qualified as ‘‘exceptionally gifted’’ according to IQ tests van Dijk & van Geert, 2007), though the present work, (see CHILDES online manuals). If it were all about the in spite of its exploratory aim, provides a global per- number of words, the TD German child should have a spective of how these words emerge and occupy their less developed linguistic skill than the group of DS place within the system. The perspective that the lin- speakers that have produced more words than her (or guistic system evolves so that several stages must be their production should be more complex than hers). 324 Adaptive Behavior 27(5) Figure 4. Four graphs from Dutch corpora—three samples of linguistic performance affected by Down syndrome: top left (age 04;04.21), top right (age 10;05.03), and bottom left (age 18;05.02). Bottom right, typical child’s performance (age 02;06.25). The Dutch determiner ‘‘de’’ has been highlighted. But, as repeatedly said in this work, the difference is connections—allegedly the promoter of small-world- not the number of words only, because two structurally ness—is considerably reduced, if not completely absent. different networks could have a similar number of Moreover, the hubs of the network also show an atypi- words/nodes. To us, the crucial difference is in the way cal behavior, with very low scores or with lexical differ- nodes are interconnected, the number of edges (syntac- ences. These results coincide again with Bates & tic connections), and in the connectivity of hubs, most Goodman (1997), who pointed out that some structures of the times represented by functional words. Here, the like function words are especially vulnerable to brain performance of Simone, reflected in the network, damage in neurological intact speakers under percep- undergoes an explosion of syntactic edges and they are tual degradation (see Barceló-Coblijn et al. 2017 for a connected in such a way that the network develops the first syntactic network analysis of linguistic develop- structure and topology of a network with a high clus- ment of a couple of twins, one of them affected by a tering coefficient and a low average path length focal brain injury). (thus suggesting small-worldness). Like the rest of TD Regarding the use and applicability of networks in children, Simone’s hubs were functional words and the psycholinguistics, statistics also indicate that this tech- relation between words and syntactic links is near to nique has convergent validity and hence that it can 1:2 (for each word, two syntactic relationships). complement other measures such as the MLU. As a Differences in the network topology are the reflection complement to MLU or other techniques, a network of their different linguistic performance (see Figure 4 analysis is not their substitute. A network analysis for comparison). In DS networks, there are less connec- extracts formal indicators from a corpus that otherwise tions in number, but also the type of long-distance cannot be obtained. Other techniques are more focused Barceló-Coblijn et al. 325 on the analysis of sentences as independent entities. Supplementary Material in the present work). From a tree-like network, one could track the path of Importantly, notice that acknowledging syntactic rela- one of the sentences, but it is certainly a challenge for tions as something real—that is, not as mere theoretical humans to track a sentence in a small-world network labels—would open the door to analyses of directed with hundreds of nodes and edges. Although the cre- networks. ation of specific software could perhaps track a sen- Finally, we would like to also frame our approach tence within a small-world network, this would be of within contemporary debates in cognitive science. Here little use for the macroscopic view we took in this work. the psychological ability for language is considered a Tracking the original sentences is not the function of complex dynamic system. Complexity emerges when networks. In our view, what they are really useful for is different elements within a system interact with each for providing formal indicators of the pattern of con- other. In the case of language, indicators of complexity nectivity extracted from a linguistic corpus. An out- had already been detected in phonology (e.g. Vitevitch, standing example of this central idea, applied to 2008) or semantics (Bilson, Yoshida, Tran, Woods, & semantic association, is Kenett, Gold, and Faust’s Hills, 2015; Hills, Maouene, Riordan, & Smith, 2010; (2015) approach to semantic organization of people Hills et al., 2009), and the present work shows that the with Asperger syndrome (AS). Their networks suggest ability to combine words also shows clear indicators of that AS speakers have a particular (hyper-modular- a complex behavior. By comparing TD and AD speak- ized), more compartmentalized, mental lexicon. ers, we have seen that the latter do develop, although With respect to syntactic theory, we think that the following a different developmental path. Our studies present work contributes to solving at least part of the fit that third Evo-Devo approach to human ontogeny issues raised by Ninio (2014). We think that some para- mentioned in the introduction, according to which the meters are more useful than others, depending on the individual development follows both linear and non- nature behind the network that is being analyzed. The linear progressions (e.g. see the classic Evo-Devo biol- small-world topology is a first indicator of efficiency ogy, for example, Alberch & Alberch, 1981; Alberch, (Martin, 2012), but in our case, the linguistic context is Gould, Oster, & Wake, 1979). An Evo-Devo approach really important (Čech, Mačutek, & Liu, 2016). It is a also provides a conceptual framework to make sense of cognitive context, and it is a (developing) organism our results on atypical development: they are to be what is under study and hence a dynamical process viewed not as delays, but as different phenotypical (van Geert 1994; Barceló-Coblijn, Corominas-Murtra, paths (hence meeting Karmiloff-Smith’s, 2007, and van & Gomila 2012). Nonetheless, this is not enough. Geert’s, 2004, viewpoint). Therefore, we think that syn- Importantly, a crucial characteristic of AD speech is tactic networks can be used as psychiatric endopheno- the atypical, low connectivity of the typical hubs, or types (Barceló-Coblijn et al., 2015) since the network even the emergence of other words as hubs, something and its formal, objective indicators meet the criteria of that has already been pointed out in other network- a biomarker bridging genotype and phenotype. When a based approaches to clinical studies (Crossley et al., biologically driven disturbance constraints develop- 2014). Moreover, in our approach, syntactic connec- ment, like in DS, the linguistic phenotype stabilizes dif- tions/edges are not random but established following ferently. Such divergent development due to the effect syntactic theory. The linearity imposed by speech is not of the trisomy in DS populations is also supported by a problem for our approach, since the analysis is based studies of the connectome that show that the connectiv- on syntactic dependency relations and syntactic theory, ity pattern of DS brains resembles a random network, which are not limited by linear distance between words while the connectivity pattern of neurotypicals is small- within a sentence. The absence of any linearity con- word (Ahmadlou, Gharib, Hemmati, Vameghi, & straints is more important than it may seem at first Sajedi, 2013). Future research will deal with linguistic glance, because, among other reasons (e.g. remind samples of other syndromes that also affect language, Zweig’s (2016) observations regarding the true nature to see whether or not it is possible to visualize a pheno- of co-occurrence networks), our syntactic analysis goes typic morpho-space, as hypothesized by Barceló- beyond the transcription and takes benefit of advances Coblijn and Gomila (2014). On another front, more in contemporary syntactic theory. In this aspect, we control is necessary during corpus creation. If session agree with Čech, Mačutek, and Liu’s (2016) observa- time is selected as a crucial feature, then it should be tion that answers about the role of syntax in networks controlled both during recording sessions (all sessions should take into account linguistic theory, and we think with the same duration) and between sessions (the same that the speaker’s syntactic ability affects several rele- intervals in days between recordings). Other aspects, vant aspects, like the nature and connectivity of hubs. like the intervention of other actors (relatives, friends, However, theoretical syntax itself can also affect hubs’ etc.) should also be strongly controlled. This could be connectivity, depending on the theoretical choice one solved with the development of a specific protocol. The applies—for example, the degree \k . (n of edges) of same can be said regarding the hypothesis the linguist determiners (see Barceló-Coblijn et al., 2019, and follows during the syntactic analysis. As in any other 326 Adaptive Behavior 27(5) discipline, syntactic theory contains several hypotheses. Supplemental material Syntactic network approaches would greatly benefit Supplemental material for this article is available online. from a detailed syntactic protocol created and shared to the community. ORCID iD Lluı́s Barceló-Coblijn https://orcid.org/0000-0002-8765-6314 8. Limitations and future directions Notes The present work contributes to the issue of the suitabil- ity of network analyses vis-à-vis language variation. It 1. See Barabási and Albert’s (1999) seminal work. 2. For example, a text like Moby Dick shows precisely the argues for such a suitability, provided that the network connectivity pattern of the small-world network (Solé modeling is combined with a proper linguistic analysis. et al., 2010). However, there still are some aspects that could be 3. http://wals.info/ improved in the future. One of them is the integration of 4. Besides, the Basque verb is highly inflected, showing syntactic information from the syntactic links inside agreement in number and person with subjects, direct edges. Until now, the available software does not allow objects, and indirect objects (plus optionally even the such integration. Ideally, information from the syntactic addressee) to the point that a sentence can be completely link such as ‘‘subject,’’‘‘complement,’’ and so on (Figure grammatical composed of just a synthetic verbal form 1) should be encapsulated inside the edge so that it can such as ba-d-akar-zki-zu-e-ke-t [Pol-X-Obring- 3PlAbs-2sgDat-2Pl-Pot-1sgErg] ‘‘I may bring be analyzed further. As mentioned before, this would them to you (pl.).’’ make possible an analysis of directed networks, since the 5. ‘‘Reverse compositionality’’ is introduced by Fodor and information ‘‘inside’’ the edge could be checked anytime. Lepore (2001) as such. Information about lexical categories (i.e. noun, verb, adjective, determiner, etc.) is also lost during file transfor- Compositionality says, roughly, that its syntax and its mation in current work. Ideally, in future developments lexical constituents determine the meaning of a complex such information should also be saved to see automati- expression; it’s thus part of the explanation of why practi- cally which lexical categories vertebrate the network, for cally everybody who understands ‘‘dogs’’ and ‘‘bark’’ example, using a specific program for these matters (e.g. understands ‘‘dogs bark.’’ But it also needs explaining Barceló-Coblijn et al., 2017). Another important aspect that you practically never find people who understand ‘‘dogs bark’’ but don’t understand ‘‘dogs’’ or ‘‘bark.’’ to reinforce is the statistical information that can be What we’ll call ‘‘reverse’’ compositionality explains this obtained from this kind of approaches. The present work by assuming that each constituent expression contributes is exploratory due to the size of corpora. Future work the whole of it’s meaning to its complex hosts. If that’s including a larger number of participants and languages right, then if you understand ‘‘dogs bark,’’ it follows that could draw a stronger conclusion on the developmental you know everything you need to determine the meanings trajectories between TD and AD children. In relation to of ‘‘dog’’ and ‘‘bark’’: in effect, the meanings of the parts this, and as observed in the previous section, gathering of a complex expression supervene on the meaning of that new data from spontaneous speech controlling time ses- expression. (pp. 365–366) sion, intervals, and the number of intervening actors will discard potential biases affecting the analysis. References Abbedutto, L., Warren, S. F., & Conners, F. A. (2007). Lan- Authors’ Note guage development in Down syndrome: From the prelin- guistic period to the acquisition of literacy. Mental Cristina Real Puigdollers is now affiliated with Universitat Retardation and Developmental Disabilities Research Pompeu Fabra, Barcelona, Spain. Reviews, 13, 247–261. Ahmadlou, M., Gharib, M., Hemmati, S., Vameghi, R., & Acknowledgements Sajedi, F. (2013). Disrupted small-world brain network in children with Down syndrome. Clinical Neurophysiology, We thank Paco Calvo and Manuel G Bedia for their support 124, 1755–1764. in the development of this work. Alberch, P., & Alberch, J. (1981). Heterochronic mechanisms of morphological diversification and evolutionary change in the Funding neotropical salamander, Bolitoglossa occidentalis (Amphibia: Plethodontidae). Journal of Morphology, 167, 249–264. The author(s) disclosed receipt of the following financial sup- Alberch, P., Gould, S. J., Oster, G. F., & Wake, D. B. (1979). port for the research, authorship, and/or publication of this Size and shape in ontogeny and phylogeny. Paleobiology, article: This work has been supported by the following proj- 5, 296–317. ects and grants: IT769-13 (Basque Government), FFI2017- Arbesman, S., Strogatz, S. H., & Vitevitch, M. S. (2010). The 87140-C4-1-P (MINECO), BIM (ANR), UV2 (ANR-DFG), structure of phonological networks across multiple languages. PSI2014-62092-EXP, TIN2016-80347-R and PGC2018- International Journal of Bifurcation and Chaos, 20, 679–685. 096870-B-100 (MiCIU, AEI & FEDER, EU). Barceló-Coblijn et al. 327 Arita, M. (2005). Scale-freeness and biological networks. Benı́tez-Burraco, A., & Murphy, E. (2016). The oscillopathic Journal of Biochemistry, 138, 1–4. nature of language deficits in autism: From genes to lan- Assen, N., & Bol, G. (2012). Fast mapping bij jongeren met guage evolution. Frontiers in Human Neuroscience, 10, het syndroom van Down [Fast mapping in juveniles with Article 120. doi:10.3389/fnhum.2016.00120 Down syndrome]. Stem-, Spraak- En Taalpathologie, Bilson, S., Yoshida, H., Tran, C. D., Woods, E. A., & Hills, 17(4), 1–19. T. T. (2015). Semantic facilitation in bilingual first lan- Baker, M. C. (2005). The innate endowment for language: guage acquisition. Cognition, 140, 122–134. Overspecified or underspecified? In P. Carruthers, Bol, G., & Kuiken, F. (1990). Grammatical analysis of devel- S. Lawrence, & S. Stich (Eds.), The innate mind: Structure opmental language disorders: A study of the morphosyn- and contents (pp. 156–174). Oxford, UK: Oxford Univer- tax of children with specific language disorders, with sity Press. hearing impairment and with Down’s syndrome. Clinical Balari, S., & Lorenzo, G. (2013). Computational phenotypes: Linguistics and Phonetics, 4, 77–86. doi:10.3109/02699 Towards an evolutionary developmental biolinguistics. 209008985472 Oxford, UK: Oxford University Press. Brock, J. (2007). Language abilities in Williams syndrome: A criti- Barabási, A.-L., & Albert, R. (1999). Emergence of scaling in cal review. Development and Psychopathology, 19, 97–127. random networks. Science, 286, 509–512. doi:10.1126/ Čech, R., Mačutek, J., & Liu, H. (2016). Syntactic complex science.286.5439.509 networks and their applications. In A. Mehler, Barceló-Coblijn, L., Benı́tez-Burraco, A., & Irurtzun, A. A. Lücking, S. Banish, P. Blanchard, & B. Frank-job (2015). Syntactic networks as endophenotype of develop- (Eds.), Towards a theoretical framework for analyzing com- mental language disorders: An Evo-Devo approach to plex linguistic networks (pp. 167–183). Berlin, Germany: clinical linguistics. Biolinguistics, 9, 43–49. Springer-Verlag. Barceló-Coblijn, L., Corominas-Murtra, B., & Gomila, A. Chapman, R. S., Hesketh, L. J., & Kistler, D. J. (2002). Pre- (2012). Syntactic trees and small-world networks: Syntac- dicting longitudinal change in language production and tic development as a dynamical process. Adaptive Beha- comprehension in individuals with Down syndrome: Hier- vior, 20, 427–442. doi:10.1177/1059712312455439 archical linear modeling. Journal of Speech, Language, and Barceló-Coblijn, L., Duguine, M., & Irurtzun, A. (2019). The Hearing Research, 45, 902–915. emergence of hubs in complex syntactic networks and the Chapman, R. S., Sindberg, H., Bridge, C., Gigstead, K., & DP hypothesis: The relevance of a linguistic analysis. In Hesketh, L. J. (2006). Effect of memory support and eli- A. Bastardas, À. Massip, & G. Bel-enguix (Eds.), Applica- cited production on fast mapping of new words by adoles- tions of complexity theory in language and communication cents with Down syndrome. Journal of Speech, Language, sciences (pp. 273–288). Dordrecht, The Netherlands: and Hearing Research, 49, 3–15. Springer. doi:10.1007/978-3-030-04598-2_15 Chomsky, N. (1975). Reflections on language. New York, NY: Barceló-Coblijn, L., & Gomila, A. (2014). Syntactic develop- Pantheon Books. ment in pheno-typic space. In E. A. Cartmil, S. Roberts, Corominas-Murtra, B. (2007). Network statistics on early H. Lyn, & H. Cornish (Eds.), The Evolution of Language: English Syntax: Structural criteria. arXiv. Retrieved from Proceedings of the 10th International Conference (EVO- http://arxiv.org/abs/0704.3708 LANG10, pp. 33–40). Singapore: World Scientific. Corominas-Murtra, B., Sànchez Fibla, M., Valverde, S., & Barceló-Coblijn, L., Serna Salazar, D., Isaza, G., Castillo Solé, R. (2018). Chromatic transitions in the emergence of Ossa, L. F., & Bedia, M. G. (2017). Netlang: A software syntax networks. Royal Society Open Science, 5, Article for the linguistic analysis of corpora by means of complex 181286. doi:10.1098/rsos.181286 networks. PLoS ONE, 12(8), e0181341. Corominas-Murtra, B., Valverde, S., & Solé, R. V. (2009). Bassano, D., Maillochon, I., Korecky-Kröll, K., van Dijk, The ontogeny of scale-free syntax networks: Phase transi- M., Laaha, S., Dressler, W. U., & van Geert, P. (2011). A tions in early language acquisition. Advances in Complex comparative and dynamic approach to the development of Systems, 12, 371–392. determiner use in three children acquiring different lan- Crossley, N. A., Mechelli, A., Scott, J., Carletti, F., Fox, P. guages. First Language, 31, 253–279. T., McGuire, P., & Bullmore, E. T. (2014). The hubs of the Bates, E., & Goodman, J. (1997). On the inseparability of human connectome are generally implicated in the anat- grammar and the lexicon: Evidence from acquisition, omy of brain disorders. Brain, 137, 2382–2395. aphasia and real-time processing. Language and Cognitive Demuth, K., & Tremblay, A. (2008). Prosodically-condi- Processes, 12, 507–586. tioned variability in children’s production of French deter- Beckage, N., Smith, L., & Hills, T. (2011). Small-world net- miners. Journal of Child Language, 35, 99–127. work growth in typical and late talkers. PLoS ONE, 6(5), Ezeizabarrena, M.-J., & Garcia Fernandez, I. (2018). Length e19348. doi:10.1371/journal.pone.0019348 of utterance, in morphemes or in words? MLU3-w, a reli- Benı́tez-Burraco, A. (2016). A biolinguistic approach to lan- able measure of language development in early Basque. guage disorders: Towards a paradigm shift in clinical lin- Frontiers in Psychology, 8, Article 2265. doi: guistics. In K. Fujita, & C. Boeckx (Eds.), Advances in 10.3389/fpsyg.2017.02265 biolinguistics: The human language faculty and its biological Fenson, L., Dale, P., Reznick, S., Thal, D., Bates, E., Har- basis (pp. 256–272). London, England: Routledge. tung, J. P., & . . . Reilly, J. S. (1993). MacArthur Commu- Benı́tez-Burraco, A., & Boeckx, C. (2014). Universal Gram- nicative Development Inventories: Technical manual. San mar and biological variation: An Evo-Devo agenda for Diego, CA: Singular Publishing Group. comparative biolinguistics. Biological Theory, 9, 122–134. 328 Adaptive Behavior 27(5) Ferrer-i-Cancho, R., & Solé, R. V. (2001). The small world of Leonard, L. B., & Finneran, D. (2003). Grammatical mor- human language. Proceedings of the Royal Society B, 268, pheme effects on MLU: ‘‘The same can be less’’ revisited. 2261–2265. doi:10.1098/rspb.2001.1800 Journal of Speech Language and Hearing Research, 46, Fodor, J. (1998). In critical condition. Cambridge: MIT Press. 878–888. Fodor, J., & Lepore, E. (2001). Why compositionality won’t MacWhinney, B. (2000). The CHILDES project: Tools for go away: Reflections on Horwich’s ‘‘deflationary’’ theory. analyzing talk. The database. Mahwah, NJ: Lawrence Ratio, 14, 350–368. doi:10.1111/1467-9329.00169 Erlbaum. Galeote, M., Soto, P., Sebastián, E., Checa, E., & Sánchez- Martin, G. (2012). Network analysis and the connectopathies: Palacios, C. (2014). Early grammatical development in Current research and future approaches. Nonlinear Spanish children with Down syndrome. Journal of Child Dynamics, Psychology, and Life Sciences, 16, 79–90. Language, 41, 111–131. Mehler, J. (1974). Connaı̂tre par déssaprentisage. In Gibson, G. (2009). Decanalization and the origin of complex E. Morin, & M. Piattelli-Palmarini (Eds.), L’unité de disease. Nature Reviews Genetics, 10, 134–140. doi: l’homme II: le cerveau humain [Knowing by forgetting] 10.1038/nrg2502 (pp. 25–37). Paris, France: Editions du Seuil. Gottesman, I. I. & T. D. Gould. (2003). The endophenotype Mitchell, M. (2009). Complexity: A guided tour. Oxford, UK: concept in psychiatry: Etymology and strategic intentions. Oxford University Press. American Journal of Psychiatry, 160, 636–645. Moro, A. (2008). The boundaries of babel. Cambridge: MIT Gruenenfelder, T. M., Pisoni, D. B. (2009). The lexical Press. restructuring hypothesis and graph theoretic analyses of Müller, N., & Ball, M. J. (Eds.) (2013). Research methods in networks based on random lexicons. Journal of Speech clinical linguistics and phonetics: A practical guide. Oxford, Language and Hearing Research, 52, 596–609. UK: Wiley-Blackwell. Hills, T. T., Maouene, J., Riordan, B., & Smith, L. B. (2010). Mundy, P., Kasari, C., Sigman, M., & Ruskin, E. (1995). Non- The associative structure of language: Contextual diversity verbal communication and early language acquisition in chil- in early word learning. Journal of Memory and Language, dren with Down syndrome and normally developing children. 63, 259–273. Journal of Speech and Hearing Research, 38, 157–167. Hills, T. T., Mouene, M., Mouene, J., Sheya, A., & Smith, L. Neal, Z. (2018). Is the urban world small? The evidence for B. (2009). Longitudinal analysis of early semantic net- small world structure in urban networks. Networks and works. Psychological Science, 20, 729–739. doi:10.1111/ Spatial Economics, 1–17. doi:10.1007/s11067-018-9417-y j.1467-9280.2009.02365.x Newman, M. E. J. (2010). Networks: An introduction. Cam- Hudson, R. A. (1990). English word grammar. Hoboken, NJ: bridge, MA: Oxford University Press. Blackwell Publishing. Newmeyer, F. (2005). Possible and probable languages: A gen- Johnson, N. F. (2007). Two’s company, three is complexity: A erative perspective on linguistic typology. Oxford, UK: simple guide to the science of all sciences. Oxford, UK: Oxford University Press. Oneworld. Ninio, A. (2006). Language and the learning curve: A new the- Kaplan, E., Goodglass, H., & Weintraub, S. (1983). Boston ory of syntactic development. Oxford, UK: Oxford Univer- Naming Test. Philadelphia, PA: Lea & Febiger. sity Press. Karmiloff-Smith, A. (2007). Atypical epigenesis. Developmen- Ninio, A. (2014). Syntactic networks, do they contribute valid tal Science, 10, 84–88. information on syntactic development in children? Com- Kenett, Y. N., Gold, R., & Faust, M. (2015). The hyper- ment on ‘‘Approaching human language with complex net- modular associative mind: A computational analysis of works’’ by J. Cong and H. Liu. Physics of Life Reviews, 11, associative responses of persons with Asperger syndrome. 632–634. Language and Speech, 59(Pt. 3), 297–313. Piattelli-Palmarini, M. (1989). Evolution, selection and cogni- Kenett, Y. N., Kenett, D. Y., Ben-Jacob, E., & Faust, M. tion: From ‘‘learning’’ to parameter setting in biology and (2011). Global and local features of semantic networks: in the study of language. Cognition, 31(1), 1–44. doi: Evidence from the Hebrew mental lexicon. PLoS ONE, 10.1016/0010-0277(89)90016-4 6(8), e23912. Popescu, M. (2003). Dependency Grammar annotator. In Kenett, Y. N., Wechsler-Kashi, D., Kenett, D. Y., Schwartz, H. Florentina, & M. Popescu (Eds.), Building awareness in R., Ben-Jacob, E., & Faust, M. (2013). Semantic organiza- language technology (pp. 17–34). Bucharest, Romania: Edi- tion in children with cochlear implants: Computational tura Universitătxii din Bucuresxti. analysis of verbal fluency. Frontiers in Psychology, 4, Arti- Ravid, D., & Schiff, R. (2006). Roots and patterns in Hebrew cle 543. doi:10.3389/fpsyg.2013.00543 language development: Evidence from written morphologi- Labrell, F., van Geert, P., Declercq, C., Baltazart, V., Caillies, cal analogies. Reading and Writing, 19, 789–818. S., Olivier, M., & Le Sourn-Bissaoui, S. (2014). ‘‘Speaking Rice, M. L., Smolik, F., Perpich, D., Thompson, T., Rytting, volumes’’: A longitudinal study of lexical and grammatical N., & Blossom, M. (2010). Mean length of utterance levels growth between 17 and 42 months. First Language, 34, in 6-month intervals for children 3 to 9 years with and 97–124. without language impairments. Journal of Speech Lan- Laka, I. (1996). A brief grammar of Euskara, the Basque lan- guage and Hearing Research, 53, 333–349. doi:10.1044/ guage. Bilbao, Spain: University of the Basque Country 1092-4388(2009/08-0183) UPV/EHU. Retrieved from http://www.ehu.eus/eu/web/ Roberts, J. E., Price, J., & Malikin, C. (2007). Language and eins/a-brief-grammar-of-euskara communication development in Down syndrome. Mental Lenneberg, E. H. (1967). Biological foundations of language. Retardation and Developmental Disabilities Research New York, NY: Wiley. Reviews, 13, 26–35. Barceló-Coblijn et al. 329 Roeper, T. (2007). The prism of grammar: How child language Stella, M., Beckage, N. M., Brede, M., & De Domenico, M. illuminates humanism. Cambridge: MIT Press. (2018). Multiplex model of mental lexicon reveals explosive Rollins, P. R., Snow, C. E., & Willett, J. B. (1996). Predictors learning in humans. Scientific Reports, 8, Article 2259. doi: of MLU: Semantic and morphological developments. First 10.1038/s41598-018-20730-5 Language, 16, 243–259. doi:10.1177/014272379601604705 Tomasello, M. (2003). Constructing a language: A usage-based Rondal, J. (1978a). Maternal speech to normal and Down’s theory of language acquisition. Cambridge, MA: Harvard syndrome children matched for mean length of utterance. University Press. Monograph of the American Association of Mental Defi- Tonelli, L., Dressler, W. U., & Romano, R. (1995). Frühstu- ciency, 3, 193–265. fen des Erwerbs der italienischen Konjugation: zwei Fall- Rondal, J. (1978b). Patterns of correlations for various lan- studien [First steps in the acquisition of Italian guage measures in mother-child interactions for normal conjugation: two case studies]. Suvremena Lingvistika, 21, and Down’s syndrome children. Language and Speech, 21, 3–15. 242–252. Trask, R. L. (1998). The typological position of Basque: Then Ruhland, R. & van Geert P. (1998). Jumping into syntax: and now. Language Sciences, 20, 313–324. doi: Transitions in the development of closed class words. Brit- 10.1016/S0388-0001(98)00006-0 ish Journal of Developmental Psychology, 16, 65–95. van Dijk, M., & van Geert, P. (2007). Wobbles, humps and Sagae, K., Davis, E., Lavie, A., MacWhinney, B., & Winter, sudden jumps: A case study of continuity, discontinuity S. (2010). Morphosyntactic annotation of CHILDES tran- and variability in early language development. Infant and scripts. Journal of Child Language, 37, 705–729. Child Development, 16, 7–33. Scarborough, H. S. (1990). Index of productive syntax. van Geert, P. (1994). Dynamic systems of development: Change Applied Psycholinguist, 11, 1–22. doi:10.1017/S014271 between Complexity and Chaos. New York, NY: Prentice 6400008262 Hall. Schmitt, C., & Miller, K. (2010). Using comprehension meth- van Geert, P. (2004). A dynamic systems approach to diag- ods in language acquisition research. In E. Blom, & nostic measurement of SLI. In L. Verhoeven, & H. Van S. Unsworth (Eds.), Experimental methods in language Balkom (Eds.), Classification of developmental language acquisition research (pp. 32–56). Amsterdam, The Nether- disorders (pp. 327–348). Mahwah, NJ: Lawrence Erlbaum. lands: John Benjamins. Vitevitch, M. S. (2008). What can graph theory tell us about Shannon, P., Markiel, A., Ozier, O., Baliga, N. S., Wang, J. word learning and lexical retrieval? Journal of Speech Lan- T., Ramage, D., & . . . Ideker, T. (2003). Cytoscape: A guage Hearing Research, 51, 408–422. software environment for integrated models of biomolecu- Vitevitch, M. S., & Castro, N. (2015). Using network science lar interaction networks. Genome Research, 13, 2498–2504. in the language sciences and clinic. International Journal of doi:10.1101/gr.1239303 Speech-Language Pathology, 17, 13–25. Solé, R. V., Corominas-Murtra, B., Valverde, S., & Steels, L. Watts, D. J., & Strogatz, S. H. (1998). Collective dynamics of (2010). Language networks: Their structure, function, and ‘‘small-world’’ networks. Nature, 393, 440–442. doi: evolution. Complexity, 15, 20–26. 10.1038/30918 Soto Valle, R., & Alonso, G. A. (2015). Apuntes sobre la West-Eberhard, M. J. (2003). Developmental plasticity and adquisición de la morfosintaxis del euskera en niños de 2 y evolution. Oxford, UK: Oxford University Press. medio a 5 años [Notes on the acquisition of Basque mor- Yorkston, K. M., & Beukelman, D. R. (1981). Assessment of pho-syntax in children between 2 and a half and 5 years]. Intelligibility of Dysarthric Speech. Tigard, OR: CC Revista de Logopedia, Foniatrı´a y Audiologı´a, 35, 30–49. Publications. doi:10.1016/j.rlfa.2014.06.002 Zweig, K. A. (2016). Are word-adjacency networks networks? Stella, M., Beckage, N. M., & Brede, M. (2017). Multiplex In A. Mehler, A. Lücking, S. Banish, P. Blanchard, & lexical networks reveal patterns in early word acquisition B. Frank-job (Eds.), Towards a theoretical framework for in children. Scientific Reports, 7, Article 46730. analyzing complex linguistic networks (pp. 153–163). Ber- doi:10.1038/srep46730 lin, Germany: Springer-Verlag. About the Authors Lluı́s Barceló-Coblijn is head of the Laboratory of Research on Complexity and Experimental Linguistics at the University of the Balearic Islands (UIB). He has got a bachelor’s degree in Catalan philology (2006, UIB) and another one in general linguistics (2007, University of Barcelona (UB)). After that, he made a master’s degree in cognitive science and language (2008, UB) and another mas- ter’s in human evolution and cognition (2010, UIB). His PhD thesis (2012) is on both language evolu- tion and language evolution in ontogeny (first language acquisition). Currently, he is developing a research agenda on language ontogeny from the viewpoint of complex, dynamic systems, analyzing corpora by means of complex networks. 330 Adaptive Behavior 27(5) Aritz Irurtzun is chargé de recherche (research associate) in linguistics at the CNRS, working at the IKER lab (UMR 5478) in Bayonne, France. Before that, he was adjunct professor at the University of the Basque Country (UPV/EHU), the same institution where he obtained his doctoral degree with a dissertation titled ‘‘The grammar of focus at the interfaces.’’ He works on the general architecture of grammar and the nature of the syntax-semantics and syntax-phonology interfaces, with particular interest on phenomena such as focus, interrogatives, event semantics, or prosody and intonation. Cristina Real Puigdollers is a postdoctoral researcher in the Department of Translation and Language Sciences at the Universitat Pompeu Fabra (UPF) in Barcelona, from 2016. She holds a PhD in linguistics from the Universitat Autònoma de Barcelona (2013). Her research interests are mainly in linguistic theory, specifically, in the semantics–syntax interface, with particular interest in the semantic and syntactic primitives that compose words. She is currently a researcher in the Group of Formal Linguistics (GLiF) at UPF. Her current work focuses on the lexicalization of events and locations in Romance languages and the theoretical models to derive crosslinguistic variation. She has published in journals including Acta Linguistica Hungarica, The Linguistic Review, and Linguistic Analysis. Emilio López-Navarro is a member of Human Evolution and Cognition Research Group (EVOCOG), an associated unit to Spanish CSIC. He got his graduate in psychology in the University of Oviedo, followed by a master’s degree in human evolution and cognition in the UIB. In 2013, he began his scientific career collaborating with EVOCOG and with clinical research groups from the UIB to address the challenges of third-wave therapies for psychotic disorders treatment. His PhD the- sis titled ‘‘Person-centered approximation to psychoses: effectiveness and outcomes’’ was advised by Enric Munar and Paul Chadwick, within the Doctoral Program of Human Cognition and Evolution. Currently, he is working in the attentional foundations of mindfulness applied to emotional distrac- tors. Antoni Gomila is a full professor of psychology at the psychology department of the UIB. His back- ground is in philosophy (PhD in 1990, ‘‘After the computational mind’’), and his interests have cen- tered on issues at the ‘‘no man’s land’’ of the frontiers between philosophy of mind, psychology, and cognitive science: representation and meaning, expression and intersubjectivity, intentional explana- tion and rationality, always taking an evolutionary perspective into account. He is a researcher of the human evolution and cognition research group. His last book is Verbal Minds: Language and the Architecture of Cognition (Elsevier, 2011).