Complement or adjunct? The syntactic principle English-speaking children learn when producing determiner–noun combinations in their early speech

Introduction

Recently, developmental studies have been offering some new evidence for the importance of the early mastery of grammatical words, and especially determiners, for syntactic development. In several studies it was found that in children acquiring various languages, the early mastery of determiners strongly predicts children’s concurrent and especially subsequent syntactic development. Kedar, Casasola, and Lust (2006) found that 18- and 24-month-old infants acquiring English oriented faster and more accurately to a visual target following sentences in which the referential expression included determiners. They concluded that by 18 months of age, infants use their knowledge of determiners when they process sentences and establish reference. Le Normand, Moreno-Torres, Parisse, and Dellatolas (2013) examined the speech of French-speaking children aged 2–4 years and correlated the diversity of word types in various form-classes with the children’s mean length of utterance (MLU) in words. They found that the diversity of word types in grammatical word types and in particular in the three most frequent grammatical form-classes, namely, subject-pronouns, determiners and prepositions, was the best predictor of MLU. Szagun and Schramm (2016) studied young deaf children acquiring German who received cochlear implants at an early age. They found that the early type and token frequencies of determiners predict MLU two years later more strongly than the early frequency of lexical words.

These results – covering different languages and different child populations – single out determiners as the most important types of words for syntactic development. Obviously, what is relevant for their influence on the mastery of syntax is their syntactic behaviour, not simply their presence in children’s active vocabulary. Most determiners are, in any case, strongly syntactic, for instance, articles do not appear as one-word utterances; as a rule, they occur together with a common noun and form a phrase with it. It follows that when we say that children learn determiners and this correlates with later syntactic development, we are really saying that they learn a phrase consisting of a determiner and a common noun; when such phrases are learned at an early stage, this predicts the concurrent or later level of syntax.

It is a well-established finding that many English-speaking children start to produce combinations of a determiner and a common noun at the very start of word-combinations. For example, Braine (1963) reported on the child Andrew who at 1;8.0 generated many different combinations in which the determiners more, no and other appeared in the first position, for example, more cereal, no water and other pocket. Similarly, the child Jonathan at 2;0.0 produced more raisins, two ducks or other ball, as well as combinations with possessives such as Elliot [’s] boat. The new results of Le Normand et al. (2013) and Szagun and Schramm (2016) indicate that such an early mastery of this particular construction facilitates in some manner the learning of other syntactic combinations, boosting children’s concurrent as well as future sentence complexity more than other types of early combinations.

Szagun and Schramm (2016) explain the importance of determiners for grammar building in German by their being essential for expressing syntactic roles in the sentence. Determiners mark case and hence syntactic roles of the phrases they head. Children who learn early a variety of determiners lay the early foundations for further grammar building. Le Normand et al. (2013) explain in similar terms the importance of determiners for French in that they encode basic grammatical relations in the clause.

Grammatical words, including determiners, having better predictive value for MLU than lexical words appears to conform with the claims of X-bar syntax (Jackendoff, 1977) and the Minimalist Program (Chomsky, 1995), according to which only functional words can be the heads of clauses and phrases. That is, grammatical words may provide the syntactic skeleton of the clause into which lexical words are inserted. As pointed out by Lebeaux (2000), only closed-class elements (namely grammatical words) may govern phrasal and clausal structures, and in order to learn to build sentences, children must acquire this vocabulary and learn its syntactic functioning. Le Normand et al. (2013) and Szagun and Schramm (2016) explain in similar terms the importance of determiners for further grammar building in young children.

The theory of how one particular syntactic construction affects the mastery of others is not yet well established in developmental research. The core idea derived from learning theory is transfer of learning through analogy (Abbot-Smith & Behrens, 2006; Keren-Portnoy & Keren, 2011; Ninio, 2006). Analogy is said to be made possible by the similarity in the learning tasks. What is called the identical element theory of transfer attributes the degree of transfer between the source and the target tasks to the overlap of shared task-components, namely, to the extent to which elements, features, or principles are shared by them (Singley & Anderson, 1985; Thorndike, 1913). In the case of transfer of learning from determiner phrases to other kinds of syntactic constructions such as verb–object, verb–adverb or adjective–noun combinations, the basis for analogy is unlikely to be shared words. Most likely is the possibility that what gets transferred is some general principle of syntactic connectivity that operates in determiner–common noun combinations as well as in various other syntactic constructions. That is, we assume the influence on other syntactic phenomena is not because of an overlapping presence of the common nouns in different kinds of phrases but because the early combinations of determiner and common noun may model some syntactic principle that underlies the combination, which can also be applied in other types of syntactic combinations.

Interestingly, it is not quite clear what the syntactic principle is that children may extract from the combination of a determiner and a common noun and use to facilitate the learning of other syntactic constructions. The reason is that it is a matter of linguistic controversy what kind of syntactic relation exists between the determiner and the common noun. The syntactic principle children learn when they master determiner–common noun combinations is either the principle of head-dependent complementation or else the principle of attribution by an adjunct. There are two basic syntactic relations: complementation, such as the relation of a subject to a verb or an object to a verb; and the adjunct relation, such as the attribution of a noun by a prenominal adjective or of an adjective by an adverb (Andrews, 1985; Crystal, 1997; Lyons, 1968). Complements (or arguments in some texts) satisfy an a priori need of a head-word to become complete, and are in principle required to be present in the sentence, whereas adjuncts are optional and can be omitted without leading to ungrammaticality. Whereas the occurrence in the sentence of complements is determined by the a priori specifications of verbs or other predicate words such as prepositions (namely, their valency), adjuncts can appear in a sentence as long as the attribution is semantically meaningful. Linguistic evidence bears out the centrality of the distinctions: the difference between complementation and the adjunct relation plays a central role in such diverse phenomena as binding and extraction (Hudson, 1990). This dichotomy is fundamental in theories of syntax, cutting across differences of detail (Chomsky, 1970; Huddleston, 1988; Hudson, 1984; Jackendoff, 1977; Matthews, 1981; Tesnière, 1959).

The major difference between complements and adjuncts is in the manner by which the logical predicate–argument relation underlying the syntax is mapped onto a syntactic head-dependent relation. In the case of complements, the head of the syntactic combination is a predicate and the complement is its logical argument. For instance, a transitive verb such as took is a two-argument predicate at the head of the verb–object combination such as took pictures which we can call f(x,y), where the object term pictures fills the logical argument role y for the function defined by the verb. The verb–object combination receives its syntactic category from the predicate, so that took pictures is a type of taking. By contrast, in an adjunct–head combination such as the adjective–noun big dog, the predicate f(x) is the adjective big, and the combination gets its syntactic and semantic category from the logical argument x of big, which is the noun dog. That is, a big dog is still a kind of dog, and big dog is still a kind of nominal expression.

Following Abney (1987), some major linguistic theories hold that determiners are the syntactic heads of common nouns, which are in a complementation relation with them. Among the theoreticians holding this view are Hudson (1984) in Dependency Grammar, Jackendoff (1977) in X-bar theory and Chomsky (1995) in Minimalism. Other theories consider determiners to be similar to adjectives and to be attributive pre-modifiers of common nouns in an adjunct relation. Among them are some other versions of Dependency Grammar such as Meaning-Text Theory (Mel’čuk, 1988), Functional Generative Description (Sgall, Hajičová & Panevová, 1986) and Lexicase Grammar (Starosta, 1988) as well as Head-Driven Phrase Structure Grammar (Pollard & Sag, 1994), and Role and Reference Grammar (Van Valin & Foley, 1980). For a review of the controversy and the linguistic arguments raised for each opinion, see Matthews (2007).

The controversy is part of a more general difference of opinion between linguistic theories that consider functional words such as auxiliaries and prepositions as well as determiners to be the dependants of the semantically rich common nouns and nonfinite verbs they are associated with, versus the view that sees the grammatical words as syntactic heads, despite their semantic emptiness or maybe because of it. In the case of determiners and common nouns, one side is convinced that the semantically rich common nouns are the true semantic arguments of verbs and other predicates, and that determiners merely provide more detailed information on the referents such as their definiteness. The other side considers more the syntactic features of the two kinds of words, seeing the determiner as the functional head of the combination.

In this study, we aim to find out which of the two basic syntactic operations, complementation or adjunct attribution, children learn when they master determiner–common noun combinations in their early speech. Our acquisition data also help settle the linguistic controversy around this question.

We test the two possibilities on a large sample of young English-speaking children’s earliest and shortest word combinations, namely, two-word-long sentences. Among such short word combinations, there are determiner–common noun combinations with an uncertain syntactic relation; some combinations which clearly express complementation, such as verbs with nominals as complements (i.e. subject–verb, verb–object, verb–indirect object relations), and some which clearly express adjunct relations, namely, attributive adjective–noun combinations. We shall correlate determiner–common noun combinations with two-word sentences containing non-controversial complementation and adjunct relations. If determiner–common noun combinations are complement relations, we expect positive correlations with verb–nominal and predicate–nominal combinations as they represent the same principle of combination, and as such are supposed to facilitate each other. If, on the other hand, determiner–common noun combinations are adjunct relations, we expect positive correlations with adjective–noun sentences, with which they share in that case the syntactic principle of attribution. This correlation is expected to be particularly high as on the hypothesis that determiners are not the heads of nouns, they are specifically thought to be a kind of prenominal attributive adjective (Matthews, 2007), therefore the association between these two categories of use is supposed to be very strong. We expect a much lower correlation with the contrasting type of sentences, as they should share, if anything, only some marginal elements of knowledge rather than the syntactic principle underlying the word combination.

Because the children in the sample are at the very start of word combinations, with their best MLU in words being below 2.0 (see Method), we examine their two-word-long word combinations of the different syntactic types as the representation of their emerging competence. The number of tokens of each type of combination that they are able to produce within the observed time-period should best reflect the facilitation each type receives from the others during this early learning period.

Method

Results

Determiner, complement and adjunct constructions selected for analysis

We identified all sentences of two words in which there was syntax of three types of relations:

Determiner–common noun combinations;

Complement relations headed by a verb or another predicate, that is, subject–verb, subject–predicate, verb–object and verb–indirect object relations;

Adjunct relations headed by a noun, that is, adjective–common noun prenominal attributive combinations.

These sentences form the corpus analysed in this study. Table 1 presents the corpus by type of syntactic relation. The average contribution of different children to the corpus was 24.6 utterances, SD = 34.1; the range was 1–187.

OPEN IN VIEWER

Table 1.

The corpus by type of syntactic relation.

Syntactic relation	Tokens	%
Complementation	4823	48.1
Adjunction	748	7.5
DT–CN combination	4458	44.4
Total corpus	10,029

Determiners in determiner–common noun combinations belonged to seven different detailed form-classes. There were 4458 sentences of this type; Table 2 presents the distribution of types of determiners in the corpus by form-class. Almost half were articles, a, an and the. The next in frequency were distributive pronouns (sometimes called quantifiers) all, another, any, anymore, many, more, next, no, other and some. The next category is of the demonstrative pronouns that, these, this and those. Possessors were mostly proper nouns with or without expressed possession clitic ’s. e.g. Jonathan’s. Possessive pronouns consisted of her, his, me, mine, my, our, own and your. Numbers were five, four, hundred, once (meaning one), one, three, twenty and two. Lastly, interrogative pronouns in the corpus serving as determiners were what and which.

OPEN IN VIEWER

Table 2.

Distribution of types of determiners in the corpus by form class.

Types of determiner	Tokens	%
Article	2173	48.7
Distributive pronoun	799	17.9
Demonstrative pronoun	557	12.5
Possessor (noun)	380	8.5
Possessive pronoun	345	7.7
Number	173	3.9
Interrogative pronoun	31	0.7
Total	4458

Complement relations headed by a verb or predicate were four subtypes: subject–verb, subject–predicate, verb–object and verb–indirect object relations. There were a total of 4823 tokens in the corpus; Table 3 presents the distribution of the subtypes. Among the complement relations of verbs, the most frequent were verbs with a direct object such as want juice, cut it, change diaper, get ball, hear tractor, hold it, wear boots, beat drum, build tower and eat cookies. The next frequent were subjects with verbs such as is it?, I did, I want, you can, birdie eat, he cries, you see?, I fall, me play and it hop. There were a few sentences of verbs with indirect objects such as tell me, give me, give her, show me and ask Maura. The most frequent type of complement combination was of a subject and nonfinite predicate such as where dolly?, baba cold, that big?, it purple, what that?, that mine, I done and baby hurt.

OPEN IN VIEWER

Table 3.

Distribution of complement relations headed by a verb or predicate in the corpus by syntactic relation.

Syntactic relation	Tokens	%
Subject–predicate	1971	40.9
Verb–object	1779	36.9
Subject–verb	1038	21.5
Verb–indirect object	35	0.7
Total	4823

Adjunct relations headed by a noun, that is, adjective–common noun prenominal attributive combinations, were relatively infrequent, only 748 tokens were observed. The sentences children produced with this relationship were, for example, big truck, little lamb and pink cup.

Correlations

In order to check if sentences expressing clear complementation are associated with sentences with clear adjunction, we computed Pearson correlation coefficients between the number of sentences of the two types. Unambiguous complementation was represented, as detailed earlier, by all sentences with subject–verb, subject–predicate, verb–object and verb–indirect object relations. Unambiguous adjunction was represented by all sentences with a prenominal adjective attributing a common noun. Although the correlation was significant, r(405) = .37, p < .001, it accounted for a mere 13.9% of the variance. This level of positive correlation is expected to occur between the number of tokens children produce of any kind of syntactic combination at this early stage, reflecting the effect of the general developmental level of the children. We might conclude that complementation and adjunction represent different syntactic principles not only linguistically but also in children’s syntactic system.

To test the alternative hypotheses of the study that combinations of determiner and common noun express either the syntactic relation of complementation or of adjunction, we computed Pearson correlation coefficients between the number of sentences of the type determiner–common noun with number of sentences with clear complementation, and compared that to the correlation of determiner–common noun with number of sentences with clear adjunction. Table 4 presents the results.

OPEN IN VIEWER

Table 4.

Correlations of tokens of determiner–common noun combinations with number of sentences expressing complementation and adjunction relations, in two-word sentences of young children (N = 407).

Predictor	Number of tokens	Correlation coefficient	Significance level (405 df)	Percent variance explained
Complementation	4823	0.79	p < .001	63.0%
Adjunction	748	0.48	p < .001	22.7%
Total clear syntax	5571

Although both correlations are significant at the p < .001 level, the percent variance accounted for is almost three times as high when we correlate determiner–common noun combinations with complementation (63%) than with adjunction (22.7%). It appears that determiner–noun combination is much more similar in distribution to complementation than to adjunction. This despite the fact that on the hypothesis that determiners are not the heads of nouns, they are specifically thought to be a kind of prenominal attributive adjective, therefore the association between these two categories of use is supposed to be particularly strong. We did find a significant positive correlation but, relative to the correlation with complementation, it was considerably lower. As said above, this amount of positive correlation is expected between any kind of syntactic combination, generated by the general developmental level of the children. By contrast, the correlation with sentences expressing complementation accounts for a much larger amount of variance, much above the random correlation that would probably be generated by any two kinds of word combinations.

Not only were the correlations with adjunction much lower than with complementation, but, in addition, adjective–common noun combinations were rare while determiner–common noun combinations were very frequent in the children’s speech. Were determiner–common noun combinations a subtype of adjective–common noun, we would expect a similar rare occurrence. Instead of this, determiner–common noun is as frequent as complementation of verbs and predicates by nominals. The pattern of results implies that determiner–common noun is a type of complementation, not adjunction.

To further examine the relationship of determiner–common noun combinations with verbal complementation, we correlated them with the three main subtypes of verbal complementation. The correlation with subject–verb combinations was 0.69, the correlation with subject–nonfinite predicate combinations was 0.77, and the correlation with verb–object combinations was 0.57 (combining direct and indirect objects). The correlation coefficients are similar in magnitude to the total of complementation sentences, which was 0.79.

Discussion

The results of this study indicate that determiner–noun combinations are a type of complementation. As far as the linguistic controversy is concerned, our results support the approach favoured by generative theories such as Chomsky (1995) and by Word Grammar, a major dependency grammar (Hudson, 1984). These theories support a strict separation of syntax and semantics and employ purely formal argument for the definition of syntactic relations. The view that functional words such as determiners head word combinations rather than the content words of the phrase such as common nouns, fits the theories’ presuppositions regarding the autonomy of syntax and its independence from semantics.

As for syntactic development, our results help explain why the early mastery of determiners is the best predictor of concurrent and future syntactic development, as found by Le Normand et al. (2013) and Szagun and Schramm (2016). When children learn determiners, which means by necessity also learning to combine them with their common noun complements, they apparently gain at the same time the syntactic principle underlying such combinations, which in turn facilitates the acquisition of other, crucially important syntactic relations. What is called the sentence core is composed exclusively of complementation relations; the three core grammatical relations that serve as the basis to the syntactic structure of clauses in English are subject–verb, verb–object and verb–indirect object relations (Foley & Van Valin, 1984). Determiner–noun combinations are not considered part of the causal core but, interestingly, they provide a way to learn more easily the syntactic, combinatory principle underlying the most important part of clausal structure. As we can see from the high correlations with each individual syntactic relation of the complementation type, the facilitation is wide ranging, and applies to different types of word combination as long as they are built on complementation. There is a good chance that the results of this study apply to other languages as well as English, since Rispoli (1991) demonstrated that the acquisition by children of the core syntactic relations in different languages is carried out through very similar processes.

It needs further study to understand why these combinations of determiners and their complement nouns provide the best exemplars for learning the principle of complementation. Children do acquire at an early age such verb-based combinations as sentences combining a transitive verb and its object, such as the combinations observed in our corpus, e.g. get ball, make hole, build tower and want juice, or a subject with a verb as in baby eat, I kick, or she sleep. On the face of it, such combinations exemplify the complementation relation no less than such determiner–common noun combinations as a baby, the lorry, two books, or this box. Nevertheless, it is the determiners and not the verbs that best predict a child’s syntactic development. The only explanation that can be offered is that, paradoxically, the rich semantics of verbs hinders the extraction of an abstract principle from the sentences in which they participate. Grammatical words such as articles serve to express grammatical content and not semantics; as such they are more suitable to serve as the source of learning and transfer of grammatical content. It appears that this also applies to one of the primary combinatory operations by which syntactic structure is built, namely, head-dependent complementation. Future studies should explore how general this finding is with respect to other grammatical rules and principles. It is possible that grammatical words are in general the best channels for the acquisition of formal linguistic knowledge.