Acquisition of locative structures across languages

It has been argued that children's acquisition of linguistic structures related to spatial concepts passes through a series of stages. Numerous studies investigating the development of spatial cognition in children independently of their linguistic skills suggest that infants already possess a conceptual representation of a number of spatial relationships at the preverbal stage (cf. Gibson & Spelke, Reference Gibson, Spelke and Mussen1983; Lakusta, Wagner, O'Hearn, & Landau, Reference Lakusta, Wagner, O'Hearn and Landau2007; Mandler, Reference Mandler, Bloom, Peterson, Nadel and Garrett1996, Reference Mandler2004; Piaget & Inhelder, Reference Piaget and Inhelder1956; Pulverman, Sootsman, Golinkoff, & Hirsh-Pasek, Reference Pulverman, Sootsman, Golinkoff, Hirsh-Pasek and Clark2002; Wagner & Carey, Reference Wagner and Carey2005, etc.). For instance, it has been shown that children discriminate between the concepts CONTAINMENT and SUPPORT already before the age of one (see Casasola & Cohen, Reference Casasola and Cohen2002; Quinn, Reference Quinn1994; Quinn, Norris, Pasko, Schmader, & Mash, Reference Quinn, Norris, Pasko, Schmader and Mash1999).Footnote ¹

It has been suggested that children come to the language-learning task already pre-equipped with non-linguistic spatial concepts, and that their task is to look for words that match them (Clark, Reference Clark and Moore1973; Nelson, Reference Nelson1974; Piaget & Inhelder, Reference Piaget and Inhelder1956; Slobin, Reference Slobin, Ferguson and Slobin1973). Furthermore, it has been proposed that spatial concepts are hierarchically organized, and that the acquisition of spatial terms is determined by a universal hierarchy. Thus, Johnston and Slobin (Reference Johnston and Slobin1979) have argued that the notions CONTAINMENT, SUPPORT, and OCCLUSION (cf. Piaget & Inhelder, Reference Piaget and Inhelder1956) are the most salient and conceptually simple, while PROXIMITY and relations such as BETWEEN, FRONT, and BACK are more complex and should therefore be acquired later than terms for the more simple concepts. By and large, this conceptual hierarchy is supported by the acquisition order of locative terms in L1 English, Italian, Serbo-Croatian, and Turkish (Johnston & Slobin, Reference Johnston and Slobin1979; see also Caselli, Casadio, & Bates, Reference Caselli, Casadio and Bates1999).

Multiple studies have shown that, prior to prepositions appearing in production, children produce locative utterances lacking overt Ps (Bowermann, Reference Bowerman1973; Brown, Reference Brown1973; Gagarina & Voieikova, Reference Gagarina, Voeikova, Stephany and Voeikova2009; Leikin, Reference Leikin1998; Nicholas, Reference Nicholas2011; Slobin, Reference Slobin1985). Examples of preposition omission in child English are provided in (1):

1. (1a) draw  paper

   Intended: ‘draw on paper’

   (Tomasello, Reference Tomasello1987, p. 85)

2. (1b) mummy  bathroom

   Intended: ‘mummy is in the bathroom’

   (Felix, Reference Felix, Weissenborn, Goodluck and Roeper1992, p. 30).

In languages with overt morphological case marking, e.g., Russian, these utterances typically contain a noun in a case-marked form. For instance, Leikin (Reference Leikin1998, p. 92) points out that “as a rule in the period 1;10 to 2;4 the prepositions are omitted in obligatory contexts including, in Russian, constructions with governed oblique cases”, and specifically in locative contexts that require overt prepositions in adult Russian (e.g., stole table.prep instead of na stole on table.prep ‘on the table’).Footnote ²

Logically, there are three possibilities regarding the state of children's underlying grammar at the stage of preposition omission. All these have been proposed in the literature, and we discuss them here in turn.

1. a. Absence of overt prepositions in production reflects absence of the corresponding functional category P in the children's underlying grammar;

2. b. The category P is present and adult-like, and preposition omission is a reflex of performance limitations;

3. c. The category P is present in children's underlying grammar, but is not yet target-like.

Examining a corpus of an English-learning SLI child, Radford and Ramos (Reference Radford and Ramos2001) propose a maturational account of preposition omission. They attribute the absence of prepositions in production to the lack of the corresponding category P in the child's underlying grammar. Specifically, they argue that in utterances involving preposition omission “the object is used without the transitive preposition which would be required to case-mark it in the adult grammar, suggesting that the object is a caseless noun expression which has no case-feature to be valued” (2001, p. 50; cf. example (4)). According to this approach, the functional category P matures in children's grammars in the course of development and, as a consequence, overt prepositions appear in production.

Nicholas (Reference Nicholas2011) proposed an alternative performance-based account of preposition omission at early stages of language development based on experimental data from child English and child Icelandic. Unlike English, Icelandic exhibits overt case marking on nouns. In Nicholas's experiments, young Icelandic-learning children regularly produced nouns denoting location or direction of movement without prepositions, which are obligatory in adult Icelandic to encode spatial relations. However, Icelandic children consistently marked the nouns with dative or accusative case inflections to distinguish between locative and directional contexts, respectively. Based on these data, Nicholas concludes that individual prepositions must be underlyingly present in child utterances, and that preposition omission is best attributed to performance limitations and not to a difference in grammatical competence.

Finally, Mitrofanova (Reference Mitrofanova2017) argues for the Underspecified P Hypothesis, according to which the category P in children's grammar at the stage of preposition omission is underspecified, and contains a Place head, which encodes a general locative meaning, but crucially no further functional projections and/or features (cf. Koopman, Reference Koopman2000; Svenonius, Reference Svenonius2006, Reference Svenonius, Cinque and Rizzi2010, for a discussion of the underlying structure of locative PPs). A child's grammar at this stage lacks some or all individual locative prepositions, which are acquired gradually at later stages. Mitrofanova (Reference Mitrofanova2017) discusses the results of two experiments with Russian children. On the production side, the results show that utterances involving preposition omission in Russian do not conform to target-like case-assignment patterns. Specifically, in all locative utterances involving preposition omission children uniformly mark the nouns with prepositional case, independently of the specific case requirements of individual locative prepositions (e.g., stole table.prep instead of na stole on table.prep ‘on the table’ or pod stolom under table.instr ‘under the table’), contra the performance-based approach. At the same time, children do not produce nouns in the default (nominative) case form, contra the predictions of the maturational approach. On the comprehension side, the results indicate that children who omit prepositions in production perform significantly worse on a pointing-out task than children from the same age group who reliably produce prepositions. The study concludes that these findings are consistent with the hypothesis that preposition omission reflects a non-target-like underlying structure in children's grammar. The grammar initially involves only one generalized (underspecified) locative category, while finer-grained distinctions within the locative domain emerge later. Mitrofanova assumes that the generalized Place head is phonologically null and universally assigns prepositional case to its complement noun, which accounts for the uniform pattern of case marking in Russian utterances involving preposition omission.

In this paper, we further explore the hypothesis that the transition from preposition omission to target-like use of prepositions reflects an underlying shift from a maximally generalized locative category to more specific locative structures. Specifically, our aim is (i) to identify the acquisition stages of locative utterances in Norwegian, and (ii) to investigate the influence of various lexical properties of individual locative items (such as frequency in the input, morphological complexity, phonological strength, and semantic ambiguity) on the rate at which individual locative Ps are acquired. Norwegian presents an ideal testing ground for the relative importance of such factors, because the language has two distinct groups of locative items. The first group consists of phonologically weak, highly ambiguous, but at the same time morphologically simple and very frequent prepositions. The second group comprises less frequent and (for the most part) morphologically complex, but at the same time unambiguous and phonologically strong locative Ps. In the next section, we discuss the system of Norwegian locative items in more detail.

The system of locative PPs in Norwegian

This study focuses on four locative relations: containment (IN), horizontal support (ON), position lower on the vertical axis / partially covered by the ground object (UNDER), and position further away from the viewer on the horizontal axis / partially covered by the ground object (BEHIND). In this section we review how these relations can be expressed in Norwegian. We focus on the dialect spoken in Tromsø, as this is the location where the experiments were carried out. However, most of the features described here are relevant for other Norwegian dialects as well.

Norwegian locative items can be divided into four groups, as shown in (2):

1. (2a) monosyllabic locative prepositions, which are typically unstressed and semantically ambiguous between locative and non-locative meanings (e.g., i ‘in’, på ‘on’);

2. (2b) mono- or bisyllabic prepositions that can carry stress and are unambiguously locative in meaning (e.g., under ‘under’, bak ‘behind’, etc.);

3. (2c) complex forms consisting of so-called “short form locative particles” (SFLPs; cf. Taraldsen & Svenonius, Reference Taraldsen and Svenonius2007) combined with basic locative Ps (e.g., inn-i into-in ‘in’, opp-på up-on ‘on’, opp-i up-in ‘in’, ut-i out-in ‘out in(to)’;

4. (2d) complex forms containing basic locative Ps and so-called ‘long form locative particles’ (LFLPs), which are themselves composed of a directional particle and a final -e (e.g., inn + e i inside in ‘in’, opp + e på up on ‘up’, ned + e på down on ‘down’, ut + e i out in ‘out’, etc.).

The four groups of locative items differ with respect to their syntactic properties. As observed by Taraldsen and Svenonius (Reference Taraldsen and Svenonius2007), monosyllabic Ps from group (2a) do not allow omission of the complement noun phrase and do not take her ‘here’ or der ‘there’ as complements, as shown in (3)–(4):

1. (3) *Det  sitter  en  katt  på

   it  sits  a  cat  on

   ‘There is a cat sitting on.’

2. (4) *Hun lekte i der / der i

   She played in there

   ‘She played in there.’

On the other hand, complex prepositions involving SFLPs from group (2c) can occur with a covert complement (a null pronominal element, or pro) and are also compatible with her/der ‘here/there’:

1. (5) Det  sitter  en  katt  oppå

   it  sits  a  cat  up on

   ‘There is a cat sitting on top (of it).

2. (6) Hun  lekte  uti  der

   she  played  out-in  there

   ‘She played out there.’

According to Taraldsen and Svenonius (Reference Taraldsen and Svenonius2007), the reason for the ungrammaticality of (3)–(4) is that group (2a) prepositions (e.g., i ‘in’, på ‘on’) impose restrictions on their complements, barring pro or her ‘here’ and der ‘there’ from this position. On the other hand, they account for the availability of such complements with group (2c) prepositions by proposing the following underlying structure (see Figure 1).

Figure 1.

Internal structure of Norwegian SFLP-P items.

Under this analysis, SFLPs are taken to originate as part of the complement of simple prepositions. This means that in cases such as (6)–(7), pro and her/der ‘here/there’ merge as complements of the SFLP rather than the preposition itself. This plausibly makes these constructions exempt from the restrictions on such complements associated with simple Ps (cf. Figure 2).

Figure 2.

The structure of Norwegian SFLP items lacking overt complements (a) and in combination with her ‘here’ (b).

Finally, Taraldsen and Svenonius (Reference Taraldsen and Svenonius2007) attribute the surface order of the morphemes in group (2c) items (SFLP-P) to leftward movement of the locative particle across the preposition, as illustrated in Figure 3.

Figure 3.

Derivation of the SFLP-P morpheme order.

Group (2d) items (LFLP-P combinations such as inn + e i ‘in’, opp + e på ‘up on’) pattern with group (2a) prepositions with respect to the discussed properties, i.e., they do not license covert complements and do not combine with her/der ‘here/there’:

1. (7) *Det  sitter  en katt  oppe  på / inne  i

   it  sits  a  cat  up  on inside  in

2. (8) *Katten  er  oppe  på / inne  i  der

   cat  is  up  on inside  in  there

Taraldsen and Svenonius (Reference Taraldsen and Svenonius2007) take this to indicate that LFLPs, in contrast to SFLPs, function as modifiers merging above i ‘in’ and på ‘on’. This means that in ungrammatical constructions such as (7)–(8), pro and her/der ‘here/there’ must merge as direct complements of i/på, which is disallowed (see above).

Finally, group (2b) prepositions (e.g., under ‘under’, bak ‘behind’) share many properties with group (2c) items involving SFLPs. In appropriate contexts, they are also able to combine with covert complements and with her/der ‘here/there’, as illustrated in (9)–(10).

1. (9) Det  sitter  en  katt  under / bak

   it  sits  a  cat  under   behind

2. (10) Katten  er  under / bak  der

   cat  is  under   behind  there

Taraldsen and Svenonius (Reference Taraldsen and Svenonius2007) do not explicitly discuss this class of prepositions. Given the data in (9)–(10), one possible analysis is that class (2b) prepositions have the same underlying structure as items involving SFLPs, but differ from them in terms of spell-out. While class (2c) items involve two separate morphemes spelling out the relevant functional heads (SFLP and P), class (2b) prepositions lexicalize both functional heads with a single morpheme (cf. Caha, Reference Caha2009; Svenonius, Reference Svenonius2012, on the model of spell-out which allows for a single morpheme to lexicalize a sequence of syntactic heads).

Summing up, group (2b) and group (2c) items pattern alike with respect to the availability of covert ground complements and her/der ‘here/there’. They also share certain phonological and semantic features: they are able to carry stress and are unambiguously locative in meaning. In this paper, we refer to group (2b) and (2c) locative items as strong Ps, in contrast to group (2a) prepositions, which we refer to as weak Ps. Group (2d) items involving LPLFs will not be considered in this study, since their production rate in the responses of our participants was negligible. The properties of Norwegian strong and weak Ps are summarized in Table 1.

Table 1.

Properties of Weak and Strong Classes of Norwegian Locative Ps

Research questions and predictions

The role of lexical factors

Given the system of locative prepositions in Norwegian, we are able to test the relative importance of different factors for the rate of acquisition of locative items. As discussed above, the developmental order in the acquisition of locative terms cross-linguistically is often taken to reflect a hierarchy of conceptual complexity. Furthermore, Johnston and Slobin (Reference Johnston and Slobin1979) have argued that lexical diversity may slow down the acquisition process. For example, English has numerous ways of encoding proximity (near, next to, beside, by, close to, etc.), while Turkish has only one term (yaninda). Johnston and Slobin argue that lexical diversity is responsible for the delayed emergence of English proximity prepositions as compared to Turkish. Similarly, morphological complexity and semantic ambiguity of prepositions are predicted to slow down the acquisition of locative terms (see Johnston & Slobin, Reference Johnston and Slobin1979, p. 531; Slobin, Reference Slobin1985). Another factor that has been argued to influence the acquisition of functional items is input frequency, with highly frequent items being generally acquired earlier (Ambridge, Kidd, Rowland, & Theakston, Reference Ambridge, Kidd, Rowland and Theakston2015; Gülzow & Gagarina, Reference Gülzow and Gagarina2007; Lieven, Reference Lieven2010; Theakston, Lieven, Pine, & Rowland, Reference Theakston, Lieven, Pine and Rowland2004; see, however, Ambridge, Reference Ambridge2010, on the uncertainties connected to estimating frequency based on various types of corpora). Furthermore, acoustic/phonological/prosodic saliency (such as, e.g., the ability to carry stress) has been found to play a facilitative role in acquisition (Gleitman & Wanner, Reference Gleitman and Wanner1982; Slobin, Reference Slobin1985; see Kern, Gayraud, & Chenu, Reference Kern, Gayraud and Chenu2014, for an overview).

Based on this previous research, we investigate the influence of six potential factors on the acquisition of locative Ps in Norwegian: conceptual hierarchy, frequency, morphological complexity, phonological strength, semantic ambiguity, and lexical diversity. Johnston and Slobin's (Reference Johnston and Slobin1979) conceptual hierarchy predicts that terms for simpler concepts, i.e., IN, ON, and UNDER, will be acquired earlier than terms for more complex concepts, i.e., BEHIND, irrespective of their weak/strong status. In other words, strong and weak versions of prepositions denoting IN and ON in Norwegian would be expected to appear approximately at the same time. On the other hand, if overall frequency is the central factor, we expect weak preposition, i.e., i ‘in’ and på ‘on’, to be acquired earlier than their strong counterparts inni ‘in’, oppi ‘in’, and oppå ‘on’, since weak prepositions are substantially more frequent in the input than strong Ps (as we discuss below, weak prepositions denoting IN and ON are at least eight times more frequent in child-directed speech than their strong counterparts). A similar result is expected if morphological complexity is the key factor: weak prepositions expressing the relations IN and ON are predicted to appear earlier in production than their morphologically complex strong counterparts. Finally, if the rate of acquisition is mainly influenced by semantic ambiguity and/or phonological strength, it is predicted that strong prepositional items should emerge earlier than weak Ps. These predictions are summarized in Table 2. Furthermore, we can explore the role of lexical diversity by comparing the acquisition of locative Ps in Norwegian with Russian, since Norwegian has multiple ways of expressing the relations IN and ON (i ‘in’, oppi ‘in’, inni ‘in(side)’, and på ‘on’, oppå ‘on’), while Russian employs only one locative preposition for each of these relations (v ‘in’ and na ‘on’). If lexical diversity is an important factor in acquisition, we expect locative items expressing IN and ON to appear earlier in the speech of Russian children than in the speech of their Norwegian peers (see Table 2).

Table 2.

Predictions for the Acquisition of Norwegian Locative Items

Notes. It should be noted that our investigation of Norwegian is not able to distinguish between all of these factors, as some of them make the same predictions. Thus, we will group together semantic ambiguity and phonological strength on the one hand, and morphological complexity and overall frequency on the other hand.

To sum up, the research questions that we ask with respect to Norwegian are the following:

1. (I) Do Norwegian children omit preposition at the earliest stages of acquisition at non-negligible rates?

2. (II) Is the rate of acquisition influenced by the lexical properties of the locative items? If so, what is the relative contribution of the following factors:

   1. (a) conceptual hierarchy

   2. (b) overall frequency

   3. (c) morphological complexity

   4. (d) semantic ambiguity

   5. (e) phonological strength

   6. (f) lexical diversity

To answer these questions, we conducted an elicited production study (Study 1) with a group of Norwegian children aged two to five. The elicited production study was modeled on the experiment with Russian children presented in Mitrofanova (Reference Mitrofanova2017). Additionally, a subset of children from Experiment 1 participated in a small-scale follow-up study (Puzzle-board task). The goal of this task was to assess the influence of pragmatic context on children's production. We also performed an analysis of longitudinal corpus data (Study 2) in order to investigate the acquisition of strong and weak prepositions over time. We present these studies in the next section.

Study 1: Elicited production

Study 2: Corpus analysis

We analyzed a corpus of 27 files of spontaneous conversations with one typically developing child (Ina) acquiring the Tromsø dialect of Norwegian. The corpus spans the period from age 1;8 to 3;3 (see Anderssen, Reference Anderssen2005, for further information about the corpus). We analyzed utterances produced by the child (Ina) and compared them to the child-directed speech (CDS) in the same corpus. The CDS comprised utterances produced by the investigator, the child's parents, and, in the last file, the child's grandmother. Table 5 summarizes the details of the corpus.

Table 5.

Overview of the Ina Corpus (Anderssen, Reference Anderssen2005)

We counted the occurrences of strong items (oppi ‘in’, inni ‘in(side)’, oppå ‘on’, under ‘under’, and bak ‘behind’) and their weak counterparts (i ‘in’, på ‘on’) for both the child and the adults. For the child, we excluded uses that were a direct repetition of an adult's or the child's own utterance (up to two utterances before the actual production). Table 6 summarizes the time of first occurrence and overall frequency of individual items throughout the corpus for the child and the adults.

Table 6.

First Occurrence and Overall Frequency of Individual Ps in the Corpus: INA and CDS

Note. * Child's age (file No.).

Table 7.

Evaluation of the Predictions

As Table 6 shows, the child's first use for all items except inni ‘in(side)’ and bak ‘behind’ occurs before age two (the first occurrence of inni ‘in(side)’ is at 2;1, and the first occurrence of bak ‘behind’ is at 2;6). Furthermore, the overall frequency of weak Ps is at least eight times higher than that of their strong counterparts in child-directed speech. The frequency of strong Ps in child and adult speech is comparable, while the frequency of weak Ps is much lower in the child's speech as compared to that of the adults.

Given that the files in the corpus are of different lengths, we calculated the proportion of Ps relative to the number of words produced per file for the child and the adults (coded as INA and CDS, respectively). The change in the proportions over the corpus is illustrated in Figures 6–7. Note that Ina's proportion of strong Ps is at an equal or higher rate than that of the adults, while her proportion of weak Ps is at a lower rate and exhibits a gradual increase throughout the corpus.

Figure 6.

Proportion of weak prepositions per 100 words over the 27 recordings.

Figure 7.

Proportion of strong prepositions per 100 words over the 27 recordings.

In order to establish whether the difference between the child and the adults is significant, we fit a generalized linear (Poisson) model where the number of uses was predicted given the interaction of two explanatory variables: speaker (child vs. adult) and age (in months). The log of number of words per file for the child and the adults, respectively, was used as offset (the generalized linear Poisson model is commonly used when the denominators for the two groups vary, e.g., the child and the adults produced a different number of words per file; see Gelman & Hill, Reference Gelman and Hill2006). The results of the analysis show that there was no significant difference between the child and the adults with respect to the use of strong Ps. There was no effect of age either: the child's use of strong Ps did not change significantly between ages 1;8 and 3;3. We conclude that the child was not significantly different from the adults with respect to the use of strong Ps throughout the corpus (see Table 12 in the ‘Appendix’ for details of the model).

In contrast, the child was not target-like with respect to the use of weak prepositions: the adults used significantly more weak Ps than the child (ß = 2.29, SE = 0.26, p < .001). There was also a significant positive correlation of the child's weak P production and age, as the proportion of weak Ps increased significantly from age 1;8 to 3;3 (ß = 0.06, SE = 0.01, p < .001). The use of weak Ps by the adults did not change throughout the corpus. The details are summarized in Table 12 in the ‘Appendix’.

As pointed out above, weak prepositions are ambiguous between locative and non-locative uses. This means that the low overall rate of weak prepositions in Ina's speech does not necessarily imply that she is non-adult-like with respect to the locative uses of these prepositions. Indeed, it could be the case that the difference observed between Ina and the adults is due to the child under-using weak prepositions in non-locative contexts, while the proportion of locative uses is target-like. To test this, we manually classified all of the child's uses of weak prepositions throughout the corpus as either locative or non-locative. We counted the cases where the complement of the preposition indicates a location (e.g., i barnehagen ‘in daycare’, i stua ‘in the living room’, på bordet ‘on the table’, etc.) as locative. In contrast, we categorized cases where the preposition is selected by the verb in verb-P constructions (e.g., se på ‘look at’, kle på ‘put on (clothes)’, glad i ‘fond of’, etc.) as non-locative. We then ran the same type of statistical analysis (generalized linear Poisson model) to estimate whether there was an effect of age for locative and non-locative uses of weak Ps separately. The results show that the proportion of both locative and non-locative uses of weak Ps increased significantly with age (ß = 0.08, SE = 0.01, p < .001 for locative uses, and ß = 0.05, SE = 0.01, p < .001 for non-locative uses; see Table 13 in the ‘Appendix’). This suggests that Ina is still in the process of acquiring weak Ps, both in their locative and non-locative uses, in the time span covered by the corpus. We further compared only the locative uses of weak and strong Ps. As Figure 8 clearly illustrates, the use of strong Ps does not change significantly throughout the corpus, while the use of weak Ps increases significantly with age.

Figure 8.

Regression lines representing a change in proportion of (a) weak and (b) strong locative prepositions per 100 words over time (age_in_months) for Ina.

To sum up, the results of our corpus analysis show that the child is not significantly different from the adults with respect to the use of strong Ps. Furthermore, the proportion of strong Ps does not change significantly throughout the corpus. At the same time, she uses significantly fewer weak Ps than the adults, and the use of weak Ps increases significantly towards the later files of the corpus. We conclude that the child's use of strong Ps becomes adult-like earlier than that of weak Ps, and that strong locative prepositions present fewer difficulties for children in the course of first language acquisition. We return to these results in the ‘Discussion’.