The study of children with developmental language disorder beyond English: a tutorial

ABSTRACT

The main goal of this tutorial is to promote the study of children with developmental language disorder (DLD) across different languages of the world. The cumulative effect of these efforts is likely to be a set of more compelling and comprehensive theories of language learning difficulties and, possibly, of language acquisition in general. Benefits to children and local societies are also likely to accrue. After presenting some of the initial considerations involved in the cross-linguistic study of children with language disorders, we provide examples of the types of questions that might be asked. The examples are informed by our own collaborative work studying children DLD across the languages of Cantonese, Finnish, German, Hebrew, Hungarian, Italian, Spanish, Swedish, and Turkish, as well as English. Examples from investigators’ work on other languages are also included. We discuss within-language comparisons of children with DLD and their same-age and younger typically developing peers as well as between-language comparisons of children with DLD. Examples concern issues of morphophonology, prosody, syntactic movement, verb paradigm complexity, and underlying mechanisms, among others. These examples—tied as they are to current theories and hypotheses—are necessarily limited to the types of languages already receiving investigative attention. Through the participation of child language scholars from a wider set of disciplines we can expand the number and types of languages studied and, as a consequence, greatly enhance our understanding of childhood language disorders.

1. Introduction

One of the positive developments in the field of child language research is the sharp increase in the number of languages being studied. Given the richness and diversity of languages around the world, these research efforts have greatly expanded our understanding of how children accomplish the amazing feat of acquiring their ambient language with seemingly minimal effort. We will no doubt learn much more as we work toward including the many additional languages in the world that have yet to be included in the child language literature (see Kidd & Garcia 2022). It is a healthy sign that scientific journals such as Language Acquisition are playing an active role in this expansion (Arunachalam et al. 2022).

Yet, conducting child language research around the world is not always easy. The diversity of languages is often accompanied by a diversity of locations, cultures, and logistical obstacles. For the study of children with language disorders, researchers are faced with an even more challenging task. These children represent only a fraction of the larger population of children in each language and identifying them often requires extra effort. However, there are compelling reasons to engage in this work. Here we mention three.

First, there may be children who speak under-studied languages who could benefit from clinical services once the groundwork has been laid. Studying these children could lead to the development of assessment measures for early identification of difficulties and more accurate diagnosis. This work could also help uncover especially vulnerable aspects of the children’s language that might become the focus of intervention efforts.

Second, studying children with language disorders across a wider variety of languages will allow more comprehensive tests of current theories of language impairments. Even within the sphere of morphosyntax, there are more than a dozen theories that continue to be plausible based on the limited number of languages thus far studied (see Leonard 2014, pp. 239-245, for a review). Some of these theories are not likely to withstand further scrutiny once a wider range of language typologies are considered.

Third, children with language disorders can contribute to theories of language development more generally. This is true regardless of whether children with language disorders are considered to be operating at a very low end of a continuum of language abilities or are viewed as having some type of disruption to the language learning system that makes them different from other children. In the case of the former, theories of language development can be tested to see if their basic tenets hold even at an extreme end of ability. In the case of the latter, theories can be tested to see if the deviations that language takes due to a disruption can be accommodated without ad-hoc principles that render the theories less explanatory.

Theories of language development that assume particular perceptual, cognitive, or motoric prerequisites can also be evaluated by studying children with language disorders. If children with a weakness in the area presumed to be a prerequisite also exhibit a language disorder, tentative support for the theory might be assumed. (Note that if some children with language disorders do not show a weakness in the presumed prerequisite area, the theory could still be correct, but not as an explanation for all types of language disorders.)

In this paper, we provide examples of the kinds of questions that might be asked when studying clinical populations in different languages. We are purposely wide ranging in our scope, covering between- as well as within-language comparisons, and questions best answered by studying languages across different typologies as well as similar languages of the same typology. It is our hope to attract new scholars to the cross-linguistic study of children with language disorders. Therefore, the questions behind our examples are empirically-driven but are not limited to a particular theoretical framework. Similarly, our questions can apply in principle to children who are bilingual. Indeed, these children can be a rich source of information, as children with a disorder in one language are likely to show a disorder in the other language, with no significant difference between the two in severity (Paradis et al. 2021). With two languages from the same child, researchers can gain an excellent perspective on how a disorder is manifested as a function of the particular language spoken.

We use as a foundation our studies of children with developmental language disorder (DLD) across different languages. These are children with significant deficits in language ability of unknown origin. There appear to be genetic sources of these deficits, though no variant or mutation of any single gene appears to be responsible. Instead, the language disorders appear to be multifactorial. Some children with DLD have accompanying weaknesses in motor coordination, attention problems, or below average nonverbal IQs, but these do not alter the basic diagnosis of DLD given that it is not yet clear how these other weaknesses, when present, are causing the language disorder (Bishop et al. 2017). Of course, children with DLD are not the only clinical population that can be the subject of crosslinguistic study. However, DLD is the most prevalent of the childhood disorders of language and many of these children’s characteristics are shared by other groups of children.

Our examples of questions that might be pursued across languages are not intended to be comprehensive. Indeed, these questions are necessarily constrained by the types of languages we have studied thus far (in collaboration with many colleagues). These languages (along with English) are: Cantonese, Finnish, German, Hebrew, Hungarian, Italian, Spanish, Swedish, and Turkish. Although far from comprehensive, we hope the examples serve to illustrate how experimental and theoretical questions might be translated into choices of languages to study in the future and linguistic details within those languages that might be most informative to examine. Within the research area of DLD, non-Indo-European languages, especially, are in need of study.

2. Some preliminaries

2.1. Selecting a target language

What language should be chosen in a study of children with language disorders? Especially when details of a theory of language development or disorders are to be tested, the language selected will likely have particular characteristics that offer a new vantage point for scholars. However, practical decisions must also be made in the selection process. Because children with a “disorder” are of interest, researchers will need to ensure that the target language resides within a culture that recognizes limited language abilities as a problem. Experts who are native speakers of the language are also important to ensure proper translation and interpretation of the children’s language production and comprehension.

2.2. Tests of language ability as one means of assessment

Initial assessment of language ability will be greatly aided by the availability of standardized assessment instruments, preferably those showing good diagnostic accuracy. Such accuracy is shown by acceptable levels of sensitivity and specificity. Sensitivity represents how accurate the instrument is in picking out children who truly have a deficit, whereas specificity represents how well the instrument avoids inaccurately identifying a child with typical language development as having a language disorder. However, even in countries with a long history of studying childhood language disorders, the diagnostic accuracy of tests has only recently been the subject of scrutiny; a study by Plante & Vance (1994) may have been the first such study. It is still common in many languages to discover that language assessment instruments have norms in the form of standard scores for each age group but unknown diagnostic accuracy. In these instances, a cutoff score representing more than one standard deviation below the mean, such as -1.25 SD, is often selected, though this type of score says more about a child’s standing relative to same-age peers than about whether the relatively low score is indicative of a disorder. Decisions regarding how to interpret such scores can be aided by whether parents, caregivers, or teachers express concerns about the child’s language abilities. It is important to note in this regard that the commonly used cutoff of -1.25 SD can be traced back to the influential epidemiological study of Tomblin et al. (1997). However, in that study, the investigators ensured that the -1.25 SD cutoff corresponded to the independent clinical judgments made by experts with experience in the diagnosis of children with language disorders.

Unfortunately, for most languages of the world, even standardized tests lacking diagnostic accuracy information are not available. In a later section of this paper, we will turn to this important issue to suggest possible steps that can be taken to conduct research on language disorders in these under-represented languages.

2.3. Within-language comparisons

Most studies beyond English are within-language comparisons between children with language disorders (e.g., a DLD group) and one or more groups of children with typical language development (TLD). Not surprisingly, age matching is often used. Group differences favoring the TLD group are usually expected, in part because to assign children to a language disorder group and a TLD group in the first place, the two groups will have already been found to be different on some type of language measure. Therefore, the comparisons using age matching are usually aimed at determining whether differences between children with DLD and TLD are seen in language measures of interest beyond the original language tests used to identify the children with DLD.

Because differences with age matching are already likely, some researchers make use of an additional group of children with TLD—a group that is younger but matched to the language disorder group on some language measure that could serve as an important control. There are many language measures that can, in principle, serve as suitable ways to match groups. For example, a study of children’s ability to learn novel nouns could employ matching according to the number of different words used by children in a spontaneous language sample of a designated length (Beverly & Estis 2003). Studies of children’s ability to process a particular type of syntactic structure could use scores on a more general sentence comprehension test to match groups (Souto et al. 2016). These matching measures should be relevant to the dependent measure but leave substantial room for group differences to emerge if the dependent measure reflects a special area of difficulty for the children.

For some languages this matching measure is mean length of utterance (MLU) because many linguistic details of interest involve additions of length in order to be used accurately. This matching technique has a long history, beginning with the work of Morehead and Ingram (1973). Children with language disorders invariably produce shorter utterances than their age mates, so inclusion of the younger MLU-matched group can control for this factor. When group differences are seen with this type of matching, it often means an area of extraordinary difficulty is uncovered in the group with language disorders. In English, for example, such comparisons are often made between five-year-old children with DLD and three-year old children with TLD matched according to MLU. Yet in these comparisons, children with DLD are usually found to use fewer tense and grammatical agreement morphemes in obligatory contexts than their younger counterparts (e.g., Conti-Ramsden et al. 2001, Eadie et al. 2002, Hoover et al. 2012, Rice et al. 1995).

Although matching on a language measure can serve as an important control, an age-matched group of children with TLD may also be needed. As a case in point, consider that Turkish noun morphology allows a string of suffixes attached to the noun stem, as in: Noun Stem + Plural + Possessive + Case (e.g., Güven & Leonard 2020), which might be seen in the Turkish equivalent of “The woman saw her-cars.” Here, the stem “car” would be followed by the plural, then the possessive (which in Turkish is used for the object “possessed”) followed by the accusative case maker (indicating the direct object role of the noun). Thus, four morphemes are used just for the single word representing her-cars in this utterance. If children with DLD are suspected to have difficulty producing sequences of suffixes, matching by MLU in morphemes could result in the children with DLD being matched with children with TLD who are very much younger (much younger than MLU controls in other languages) and, consequently, not as capable as the DLD group on other details of language. The inclusion of a second comparison group—an age-matched TLD group—could indicate whether these other details of language are also below age level in the DLD group, albeit less seriously impaired than the production of suffix sequences.

There has been some criticism of matching younger TLD children with children with language disorders on the basis of measures such as MLU (Plante et al. 1993) because younger TLD children differ from older children with language disorders in many ways, such as in social or emotional maturity, or in their raw scores on nonverbal intelligence tests (e.g., a standard score of 100 will reflect a lower raw score for a three-year-old than for a five-year-old). Furthermore, some experimental tasks ideally suited to measure the dependent measure of interest may require levels of attention or testing experience that are beyond the levels of younger control children, despite their seemingly similar abilities with potential language-matching measures.

When the comparison group is limited to same-age TLD peers, one alternative is to use scores on other measures as covariates. The most common of these (e.g., nonverbal intelligence test scores, maternal education level) reflect factors that, if unaccounted for, might complicate interpretation of the study’s findings. The benefit of covariates is a potentially less cluttered interpretation of the findings. For example, children with language disorders might be found to be weaker than age mates in their complex sentence comprehension; it would be an even stronger finding if it is also found that the difference holds even when group differences in nonverbal intelligence test scores are taken into account.

In some languages, the available language tests may not allow for placing children into discrete language disorder and TLD categories. In addition, some investigators may be skeptical that there even is a sharp boundary between children with a disorder and those whose language skills are simply weak. When discrete group categories cannot be made with confidence, an option is to conduct analyses using a continuum of language abilities as has been done in some studies in recent years (e.g., Krishnan et al. 2021, Poll et al. 2013).

In languages for which there are no tests of language ability with normative data, the use of a continuum is especially useful. In this case, the scores placed on a continuum would be from a researcher-created language measure that represents a broad measure of language ability. The comparative measures would then be the more focused language measures that the researcher develops to evaluate the theory or hypothesis at hand. Such a strategy might reveal, for example, that children at a certain point on the continuum on the broad measure of language ability show a disproportionate weakness on the more focused measures. A finding of this type might represent a first step toward identifying a true language disorder.

2.4. Comparisons across languages

Within-language comparisons are important for identifying specific areas of linguistic difficulty for children acquiring those languages. It is also common for such studies to refer to findings from studies on other languages to assist interpretation. In the past, such references were usually made in Introduction and Discussion sections and therefore the comparisons between languages were quite indirect. Recently, meta-analyses have appeared in the literature, which allow more direct comparisons across languages even though the data came from separate studies. For example, Krok & Leonard (2015) reported past tense use in children with DLD across several non-English Germanic languages (Swedish, Danish, Dutch) as well as English. The results for the children with DLD when compared to age-matched and younger TLD children were reported for the studies on English combined and for the studies on non-English Germanic languages combined. Analysis of the data from the combined English and the combined non-English Germanic studies revealed greater use of past tense by both younger and age-matched TLD groups than by the children with DLD.

Studies selected to be included in meta-analyses must meet certain criteria, but they rarely involve a collection of studies with identical methods and measures. For research questions requiring even closer correspondence between the target languages in the ages of the children (e.g., five years of age only), the methods used (e.g., sentence repetition in particular), and the analyses used (e.g., mixed effects models), more direct between-language studies can be conducted. Direct between-language comparisons of children with language disorders represent an even stronger test of any cross-linguistic differences. Yet, such comparisons introduce an important factor that needs to be controlled – the severity of the language disorder. If a group of, say, five-year-olds with DLD in Language A is found to be more proficient in using some linguistic feature than a group of five-year-olds with DLD in Language B, we cannot always assume that the differences are related to the language being acquired. The children with DLD in Language B might have had a language disorder of greater severity.

A matching strategy first implemented by Lindner & Johnston (1992) provides a potential solution to this problem. Children with language disorders in the two languages can be matched according to their standard scores on tests that purport to measure similar types of ability (e.g., morphosyntactic production). Given the likely cross-linguistic differences, this matching will not neutralize differences tied to how children with language disorders handle details in their respective languages. For example, English- and Italian-speaking children with DLD might have the same standard scores but the Italian-speaking children will be more successful in using tense and agreement inflections but less successful in using direct object pronouns in finite verb contexts, as these are usually clitics that precede the finite verb in Italian (discussed in further sections).

3. Cross-linguistic study as a natural experiment

By “natural experiment” we refer to how the structure of one language may parse linguistic details in a way that is not possible in another language and, by doing so, can provide insight into potential sources of difficulty in that other language. There is likely a very long list of questions that might be at least partially answered in this way. Here are a few examples relevant to DLD.

3.1. Tense/aspect

Studies of English-speaking children with DLD have tried to determine factors responsible for the past tense difficulties of these children. Linguistic frameworks have provided a useful perspective on how and when tense might be represented in the grammars of these children, and even how such representations may differ from one language to another. However, the basic ingredient of these proposals—the very notion of “tense”—is not always so easy to define. In English, past tense is difficult to isolate from lexical and grammatical aspect. In studies of children with DLD, tasks requiring distinctions between present and past progressive (e.g., Elmo is riding his car vs. Elmo was riding his car) have been used (Leonard et al. 2003), as well as distinctions between past progressive actions that had versus had not been completed (Leonard & Deevy 2010). Accuracy in producing past tense with telic as opposed to atelic verbs has also been examined (e.g., Leonard et al. 2007). Collectively, the findings from these studies suggest that English-speaking children with DLD are slow to master tense in part because they do not avail themselves of lexical and grammatical aspectual information that often accompanies tense. Younger children with TLD, in contrast, are more drawn to completed actions which may serve as a first pass at learning past tense.

Could this insensitivity to aspect be because aspectual information in English is often bound up with tense, or might the problem be with aspect itself? A language such as Hungarian might provide some preliminary answers to this question. Hungarian has a separate perfective aspect morpheme that can co-occur with a past tense morpheme when completion of the past action is expressed. Leonard et al. (2012) found that Hungarian-speaking children with DLD were less accurate than age-matched and younger children with TLD in their use of the perfective morpheme. Instead of using the morpheme to express the equivalent of The monkey had eaten the cake, the morpheme was absent, leaving the imperfective equivalent of The monkey was eating the cake. In addition, the perfective morpheme was sometimes included when the imperfective expression (that lacks the morpheme) was required.

Cantonese provides another vantage point. This language marks aspect but not tense. Fletcher et al. (2005) discovered that Cantonese-speaking children with DLD made less use of both perfective and imperfective markers than younger as well as age-matched children with DLD.

Thus far, it appears that aspect—apart from tense—may represent an additional weak area for children with DLD. This picture will no doubt become clearer when DLD researchers set their sights on other Sinitic languages and languages from other language families that provide a sharp separation between tense and aspect.

3.2. Articles and definite/indefinite reference

When English-speaking children with DLD omit articles, are the difficulties confined to prosodic challenges or do problems with definite/indefinite reference contribute to the problem? There are already insightful studies of typically developing children showing that articles that immediately follow stressed syllables are more likely to be produced than articles that precede stressed syllables (e.g., Gerken 1994, 1996). Data from English-speaking children with DLD show a similar pattern (McGregor & Leonard 1994). Although the evidence for prosodic factors is clear, there is the remaining question of whether problems with indefinite and definite reference also play a role. Swedish provides evidence that, at least definite reference may not be especially problematic when prosodic challenges are reduced. In Swedish, indefinite articles are unstressed monosyllables that precede the noun, but the definite counterparts are expressed as an unstressed syllable attached at the end of the noun. In spontaneous speech, children with DLD are more likely to omit the former than the latter (Hansson et al. 2003), and do so more than same-age peers with TLD. However, they seem as capable as their TLD age mates in using the syllabic suffix to express definite reference. Also in Swedish, in the specific context of article + adjective + noun combinations, articles are used to express both definite and indefinite reference. In these contexts, children with DLD omit both forms of articles to a greater degree than their peers. Thus far, then, prosody seems to be an important factor underlying the difficulty, but surely this is not the enitre story.

3.3. “Wh-” questions

Cross-linguistic study can also illuminate some of our assumptions about why children with DLD have great difficulty with particular syntactic constructions. For example, in many languages, children with DLD appear to have difficulty with syntactic constructions requiring movement. One example is seen in “wh-”object questions (e.g., Italian: Arosio & Guasti 2019; English: Deevy & Leonard 2004; Hebrew: Friedmann & Novogrodsky 2003, 2007). But as research is extended to still other languages, it seems that movement is not the only factor. In Cantonese, the wh-word remains in situ, so wh-subject questions appear as in “Who chases Piglet?” and wh-object questions appear as in “Piglet chases who?” Cantonese-speaking children with DLD are as accurate as same-age and younger MLU-matched control children on the former but are much less accurate than both comparison groups on the latter (Wong et al. 2004). Some factor beyond issues with movement seems to be at work, affecting children with DLD to a greater degree than younger children with TLD. It would be important to identify this factor and determine whether it compounds children’s difficulty in languages that also require movement in wh-object questions.

3.4. Relative clauses

Relative clauses constitute another element of language in need of study. In languages such as Greek (Stavrakaki et al. 2015), Hebrew (Friedmann & Novogrodsky 2004), Italian (Contemori & Belletti 2014), Danish (De Lopez et al. 2014), and Russian (Rakhlin et al. 2016), as well as English (Frizelle & Fletcher 2014), object relative clauses are more likely than subject relative clauses to be more difficult for children with DLD than for younger typically developing children. Factors frequently cited as responsible for this difference include working memory ability, the distance between the subject or object and its interpreted position in the sentence, and whether there are also grammatical case cues (e.g., nominative versus accusative) to resolve the subject versus object relative clause interpretation. However, there appear to be additional factors at play.

In Mandarin, children with DLD also have more difficulty with object relatives than with subject relatives yet the order of the constituents places object relatives, not subject relatives, in the canonical subject-verb-object order of the language (Yu et al. 2023). In Cantonese, a different picture emerges. As in Mandarin, Cantonese object relatives more closely resemble the language’s canonical subject-verb-object word order but children with DLD produce subject and object relative clauses with comparable accuracy and at accuracy levels similar to those seen for younger children with typical development (Chan et al. under review). Input frequency may play a role, but another factor seems to be the role of classifiers (CL) in Cantonese which make object relative clauses (as in “piggy push that CL doggy”) essentially identical to simple subject-verb-object sentences expressing the same meaning (Chan et al. 2021).

The aforementioned examples reflect very specific questions that need further exploration. However, there are broader questions dealing with entire classes of languages that must be answered to achieve a more adequate understanding of the language difficulties experienced by children with language disorders. We discuss some of them in the subsections that follow.

3.5. Paradigm complexity and language typology

There are some longstanding assumptions that typology can influence the relative ease by which young children can acquire particular details of their language. The distinction between languages with an agglutinating morphology and those with a fusional morphology is one such example. In particular, agglutinating languages might offer an inflection-learning advantage over fusional languages. If so, we can ask if this also applies to children with DLD. Pinker (1984) has proposed that children’s discovery of the grammatical dimensions relevant in a language’s inflection system might be aided if each inflection reflects a single dimension. For example, children might hypothesize that an inflection marks number, or person, or gender. If the ambient language fails to support the assumption of a single dimension after several such hypotheses, children will consider the conjunction of dimensions, such as number and person, or number and gender. This would give a learning advantage to agglutinating languages where each inflection reflects a single dimension. Children would arrive at the correct interpretation sooner than in a fusional language where the correct interpretation would have to wait until children consider two or more dimensions together. Judging from some of the descriptions of typical language development in agglutinating languages, this proposal seems quite accurate. For example, Aksu-Koҫ and Slobin (1985) observed that much of the noun and verb morphology of Turkish is acquired at near-mastery levels by 24 months of age.

If this proposal is accurate, it might apply to children with DLD in two different ways. First, the magnitude of the differences between children with DLD and their typically developing peers might be reduced in agglutinating languages. Second, children with DLD in agglutinating languages might show greater accuracy with inflections than their DLD counterparts learning a fusional language. However, two factors have the potential to subvert the presumed “one-dimension-at-a-time” advantage. When several dimensions must be expressed with a noun or verb, strings of inflections are required in an agglutinating morphology, which add length and knowledge of the correct order in which the inflections must appear in the sequence. In addition, when more inflections must be attached to each other along with attachment to a noun or verb stem, there are likely to be morphophonological rules that apply to these combinations. Such rules can apply to fusional languages as well, but such challenges are compounded when two or more inflections in a sequence must undergo these operations.

Evidence thus far suggests that agglutinating languages can pose problems for children with DLD, just as is seen in fusional languages. Evidence comes from languages such as Turkish (Güven & Leonard 2021), Finnish (Kunnari et al. 2011), and Hungarian (Lukács et al. 2009). For example, in the Güven & Leonard (2021) study, the use of verb inflections by Turkish-speaking children with DLD (M age = 5;3) was compared to that of younger typically developed children (M age = 2;10) matched according to MLU in words. Across verb inflection types, the mean use in obligatory contexts was 97% for the typically developing children and only 72% for the children with DLD. In the Kunnari et al. study of Finnish-speaking children with DLD, these children (M age = 5;2) showed less use of verb inflections (67%) than a group of younger children (M age = 3;8) with typical language development (89%). MLU was not examined, however, length could not have been the sole factor, given that some errors by the children with DLD contained longer strings of inflections than the correct form.

The errors seen in agglutinating languages vary somewhat depending on the types of stimuli used and the focus of the study. Evidence to date from Hungarian, Finnish, and Turkish reveals no errors in the order in which suffixes appear following the stem. Errors of one case replacing another case can occur. For example, Leonard et al. (2014) found that Finnish-speaking children with DLD sometimes confused the partitive case and the accusative case – producing either one for the other.

One common pattern is a tendency for children with DLD to leave out some of the suffixes required. Consider the following forms for kutya “dog” in Hungarian shown in (1). The form kutya is the stem but serves as nominative singular.

Table

(1)	Nominative singular Plural (only) Accusative (only) Plural + Accusative
	kutya kutyák kutyát kutyákat

From these examples, it can be seen that a change from singular to plural involves the addition of a suffix, as does a change from nominative to another case, such as accusative. The agglutinating nature of nouns can be seen when, to express the word “dog” in an accusative plural context (a context equivalent to “I fed the dogs”), a plural suffix is needed, followed by an accusative suffix. Using a probe task, Lukács et al. (2010) found that children with DLD were more likely than younger control children to produce only the plural (kutyák) or only the accusative (kutyát) when both suffixes were required to be produced in sequence. They differed most from the younger controls on this type of item. Interestingly, bare stems—representing the nominative singular form—were only produced when the item required only the plural or only the accusative suffix. When both suffixes were required in sequence, only one of the suffixes was omitted.

In Hungarian, Finnish, and Turkish, when a single suffix is required and children with DLD fail to include it, the error is almost always the production of the nominative singular, a bare stem rather than a failed attempt to articulate the suffix. Evidence for this view is seen when the form that would remain if the suffix was not successfully articulated differs from the nominative singular. To use an example reported by Leonard et al. (2014) for Finnish, when children with DLD failed to use the accusative forms ove-n (“door”) and lapse-n (“child”), they produced the nominative singular ovi and lapsi, instead of showing evidence in the vowel of an approximation of the accusative, as would be seen in the productions of *ove and *lapse.

Do the morphophonological requirements of an agglutinating language contribute to the difficulties of children with DLD? Here we refer to the phonological changes required when suffixes must be combined with the stem and suffixes must be combined with each other. Morphophonological changes can be required in the stem itself, or in the suffix. In the Kunnari et al. (2011) study of Finnish DLD described earlier, this issue was avoided through the use of specific probe items that required no special morphophonological changes. Even in this instance, children with DLD were less accurate that younger as well as older children with typical language development. However, when morphophonology is allowed to play out, it clearly influences the findings. In some instances, the children with DLD add the suffix without making a required change in the suffix. For example, in Finnish, the nominative singular (bare stem) for “coat” is takki and the accusative singular is taki-n. Productions such as takki-n occurred, entailing the addition of the accusative suffix -n to the stem takki. Another example is the production of noita-n (“witch” in the genitive singular) instead of noida-n where the stem is noita. Güven & Leonard (2020) found a difference in accuracy for Turkish-speaking with DLD between suffixes whose addition required a morphophonological change (60%) and suffixes that did not (78%). For younger MLU control children, even changes requiring morphophonological changes were produced with high accuracy (91%). Similar difficulties with morphophonological changes have been reported for Hungarian-speaking children with DLD (e.g., Lukács et al. 2009). Collectively, these studies suggest that differences favoring younger children with typical language development over children with DLD are magnified when scoring requires not only use of the correct suffixes but also correct application of the morphophonological rules

Although we have reviewed evidence showing that children with DLD acquiring agglutinating languages are weaker in morphology than younger children with typical language development, much more research is needed to address the proposal that even for children with DLD, agglutinating morphology provides an advantage over fusional morphology. Direct comparisons are extremely difficult to find in the literature on DLD. What are needed are comparisons between languages with agglutinating morphology and those with fusional morphology that require the same number (and preferably the same types) of dimensions that, in the case of the fusional language, must be expressed in a single noun or verb inflection.

One step in this direction might be the study of children with DLD who are acquiring Hebrew. This language has a fusional morphology which can require the combination of up to four dimensions. Specifically, within past tense there are also distinctions based on number, gender, and person. Dromi et al. (1999) found that inflections within past tense in particular were more problematic for children with DLD relative to younger MLU controls. Errors were not limited to problems with tense; errors represented a mix of dimensions that were not properly included in the children’s productions. As we will discuss below, in Romance languages such as Spanish and Italian, fewer dimensions are fused in the verb inflections. In those languages, accuracy by children with DLD is more on par with that of younger MLU control children unlike in Hebrew. However, much more research is needed before we have definitive answers to the question of whether there is a tipping point in the number of dimensions – fused or agglutinating – that children with DLD can successfully manage.

3.6. Can differences in typology summon different underlying mechanisms?

A greatly understudied area of DLD is the degree to which different language typologies call on different underlying mechanisms, some of which may be quite inefficient in children with DLD. For languages in which these particular mechanisms play a role, the areas of language most relying on them will be most impaired. For example, across many languages, children with DLD have been found to be quite weak in their ability to repeat multisyllabic nonwords. This weakness has been correlated with a range of weaknesses in lexical learning (e.g., Gathercole et al. 1992), and syntactic comprehension (e.g., e.g., Montgomery & Evans 2009), though pin-pointing the nature of this relationship to word learning tasks or particular syntactic operations has been difficult (e.g., Melby-Lervåg et al. 2012). In contrast, the ability involved in nonword repetition may be more centrally involved in learning the intricacies of morphophonology in agglutinating languages. For example, in a study of the agglutinating language, Hungarian, Lukács et al. (2009) asked whether nonword repetition might relate to children’s ability to use sequenced tense and agreement inflections in their verb usage. They initially found that children with DLD were less accurate than a group of younger controls in their use of these inflections, often leaving out one of the inflections or failing to apply the correct morphophonological rule. However, when the children’s nonword repetition scores were used as a covariate, the group differences disappeared. Lukács et al. (2009) proposed that this effect was meaningful because in agglutinating languages, children’s control of sequences of sounds is especially important.

Cantonese may be at the opposite end of the continuum. Stokes et al. (2006) found that Cantonese-speaking children with DLD resembled same-age peers and exceeded younger peers in their nonword repetition ability. Cantonese is a tone language with a more restricted range of consonant-vowel combinations than in many non-tone languages. Furthermore, in multisyllabic words, tones are only rarely neutralized unlike in some other tone languages. These characteristics likely contributed to the greater ease with which the children with DLD could produce the nonwords. As Stokes et al. (2006) put it, languages such as English and Swedish “have complex phonotactic structures, variable stress patterns, and difficult-to-articulate consonants. Cantonese has none of these.” (p. 230).

There may well be other underlying weaknesses that exacerbate deficits in some languages more than others. Even social communication skills such as awareness of the listener’s needs could influence sentence use. For example, in Cantonese, a sentence that is appropriately specific might include an aspect marker but there are contexts in which the same sentence without the marker is acceptable. Similarly, because Cantonese does not employ tense, the relationship between speech time and event time can be expressed by a temporal adverb. Without the adverb, the sentence can have a default interpretation. In both of these instances, a child with poor social communication awareness might produce sentences that are communicatively inadequate but these sentences may or may not signal a grammatical time-related deficit.

3.7. A closer look at inflection errors

When grammatical inflection errors occur in fusional languages, they can be characterized in several different ways. And children with DLD have been described as having extraordinary weaknesses from each of these perspectives. The first perspective is one that has received the greatest scholarly attention. According to this view, children with DLD are unusually slow in acquiring an underlying linguistic principle related to tense and grammatical agreement (Rice & Wexler 1996, Wexler et al. 1998). This delay prolongs the period when, in finite verb contexts, English-speaking children with DLD produce bare stem infinitives, German-speaking children with DLD produce overt infinitives in sentence-final position (Rice et al. 1997), Italian-speaking children with DLD omit pre-verb direct object clitics (Wexler, 1998, 2003), and (with some additional assumptions), Dutch-speaking children with DLD produce either bare stems or overt infinitives (Wexler et al. 2004).

A second perspective shares with the first the view that many verb errors are productions of nonfinite verbs, though a more accurate characterization would be that the verb errors are derived from nonfinite verb forms. In languages such as English, Swedish, Danish, German, and Dutch, these errors match nonfinite sequences that appear in larger structures in the input, such as questions (e.g., English Can [Jennifer drink wine]?; Swedish Kan [Ingrid dricka vin]? Danish Kan [Emma drikke vin]? German Kann [Hans Wein trinken]? Dutch Kan [Jan wijn drinken]?) and other constructions. If children with DLD do not appreciate the structural dependencies involved in these larger structures, they might extract the nonfinite sequences and use them as stand-alone utterances (see, e.g., Leonard et al. 2015). Errors of this type seem to be closely dependent on the type of language being acquired, though more languages need to be studied to determine the range of languages in which the dominant error can be described as a nonfinite verb. English seems to fall at one extreme, with Germanic languages showing somewhat less of this type of use than English, and Romance languages showing less than Germanic languages. Despite English falling at the extreme end of the continuum, one might argue that the strongest evidence for the errors-as-nonfinite-productions view comes from Germanic languages that have overt infinitive inflections (such as those mentioned previously). The fact that infinitives in English are bare stems makes interpretation a little less straightforward, as we describe in the following section.

A third characterization of errors is that children with DLD often default to a more frequent form of the verb. This view can, in principle, coexist with the nonfinite production view. For example, children may inappropriately extract nonfinite verb forms from larger structures (e.g., Can [she see me]?) and mistakenly assume they can use them in simple sentences by virtue of hearing (what are actually finite) bare stems in simple sentences in the input (e.g., They see me). This might occur when children are both weak in understanding structural dependencies and limited in their command of the inflection paradigm of their language. Even without nonfinite verbs appearing in larger structures, children with DLD might, according to the defaulting view, “bail out” by simply selecting the form of the verb that is most familiar.

Because bare stems are the most frequent in English, they are the most likely choice to be adopted as a default. However, the defaulting choice might be verb-specific, that is, dependent on the input frequency of the bare stem relative to other forms of the same verb. For example, Kueser et al. (2018) found that English-speaking children with DLD were more likely to use third singular -s in obligatory contexts with verbs that most frequently appear in this form in the input; conversely, for verbs that rarely appear with third singular -s in the input, the bare stem was produced instead.

Fortunately, the idea of defaulting has been explored in languages beyond English. Spanish is one such language. The present tense third person singular form is a very frequent form in Spanish and also appears to be the most frequent substitute for other finite forms in the productions of children with DLD (Grinstead et al. 2009). Grinstead et al. (2013, 2018) have taken this finding to another level in proposing that in these instances the children are using present third person singular as a type of nonfinite form.

A fourth characterization, especially applicable to languages with a rich fusional inflectional morphology, is that children with DLD possess the notions of tense, number, person, and gender, but are more prone than their peers to produce fused forms that approximate but not fully match the combination of features required in the context. Even in Spanish where there is a dominant substitute, other forms are sometimes produced in error. These errors might be viewed as “near-misses” as when a child produces present first singular in place of present first plural, or present first singular in place of past first singular (see Bedore & Leonard 2001). Similar findings have been reported for children with DLD acquiring Italian (Bortolini et al. 1997) and Hebrew (Dromi et al. 1999).

Near-miss errors have also been reported for agglutinating languages that also have a fusional component in their morphology. In Hungarian, the verb must agree with the subject in person and number and with the object in definiteness. When combined with tense, verb inflections make 24 different distinctions. Lukács et al. (2009) presented Hungarian-speaking children with DLD with sentences for which the children had to provide the correct verb form. Incorrect productions were examined in terms of whether the errant form differed from the correct form in one dimension (e.g., an error only of person or definiteness), two dimensions (e.g., an error of both person and definiteness), three dimensions (e.g., an error of person, number, and definiteness) or all four dimensions. Given these dimensions, there were 23 different types of substitution errors that were possible; five differed from the target on only one dimension, nine differed on two dimensions, and seven differed on three dimensions. (Two differed on all four dimensions but no such errors occurred in the data.) Strikingly, errors were disproportionately one-dimension “misses” despite representing a relatively small proportion of the errors that could have been made. No one substitute stood out as the most frequent, including the form considered to have the simplest morphology, the present tense third person singular indefinite form.

Near-miss errors warrant further study because they imply a level of grammatical knowledge not bestowed on children with DLD when errors are viewed as the children’s grammars lacking a linguistic principle or the children automatically resorting to a frequently heard form. It is possible that near-miss errors reflect the child’s growing awareness of each dimension in an inflection paradigm (number, tense, person, gender) while still lacking the ability to consistently choose the form that represents the correct combination of these dimensions.

More study is also required of languages with “mixed” errors—languages in which a frequent substitute can be regarded as a default while the remaining substitutes have a near-miss appearance. For example, along with using infinitives in place of finite forms, Dutch-speaking children with DLD often use a bare stem (Blom et al. 2014, de Jong 1999, Spoelman & Bol 2012, Verhoeven et al. 2011, Wexler et al. 2004). In Dutch, first person singular in present tense is a bare stem. The bare stem is also used in second person singular questions, as in Drink je koffie? “Drink you coffee?” where the declarative would employ the second person singular drinkt. Because the bare stem does appear in first person singular and interrogative second person singular contexts in Dutch, it is plausible that it is a near-miss error when appearing in contexts requiring an overt inflected form. The alternative argument would be no more parsimonious—that there are two default forms, perhaps one with nonfinite status and the other with finite status. Other languages will surely contribute toward answering these questions.

4. Interpreting the difficulty with object clitics

In Romance languages, object clitics are probably the most reliable grammatical details that distinguish children with DLD from their peers, though, as will be seen, difficulties with object clitics can also be found in other languages. The prototypical difficulty is seen when the language under study has a subject-verb-object canonical word order and when discourse or the physical setting makes the direct object clear. In these instances, an object clitic is typically expected instead of the noun as direct object. If the verb is finite, the clitic precedes rather than follows the finite verb. An example from Italian can be seen when instead of La ragazza apre il libro “The girl opens the book,” the speaker produces La ragazza lo apre “The girl it opens” with the clitic lo. Considering that the clitic often constitutes an obligatory argument in the sentence, and adds less length to the utterance than the element it replaces (il libro in this example), it is surprising that direct object clitics are so difficult for children with DLD. These children make significantly less use of object clitics in appropriate contexts than both their same-age peers and, in many studies, children two or three years younger. Such differences are seen in Italian (e.g., Arosio et al. 2014, Bortolini et al. 2006, Cipriani et al. 1993, Dispaldro et al. 2013, Leonard & Dispaldro 2013); Spanish (e.g., Bedore & Leonard 2005, Bosch-Galceran & Serra-Raventós 1994, Castilla-Earls et al. 2020, De la Mora Gutiérrez 2004, Morgan et al. 2012, Simon-Cereijido & Gutiérrez-Clellen 2007); and French (e.g., (Jakubowicz et al. 1998, Jakubowicz & Tuller 2008, Maillart & Schelstraete 2003, Paradis et al. 2003). Similar problems have been reported for children with DLD acquiring Greek, a null-subject language with subject-verb-object order from the Hellenic family of languages (e.g., Chondrogianni et al. 2015, Dosi & Koutsipetsidou 2019, Smith et al. 2008, Stavrakaki & van der Lely 2010, Tsimpli & Mastropavlou 2008).

Although most studies are within-language comparisons between children with DLD and their same-language peers with typical language development, a few between-language comparisons have appeared. For example, there have been comparisons between Italian-speaking children’s use of object clitics (e.g., lo in La ragazza lo spinge “The girl him pushes” and English-speaking children’s use of the most similar forms in their language, direct object pronouns (e.g., The girl pushes him). Although English-speaking children with DLD have many grammatical weaknesses, the use of object pronouns is not one of them. Not surprisingly, then, clear group differences are seen favoring the DLD group acquiring English (Bortolini el al. 1998, Leonard et al. 1987). Likely reasons for this difference will be apparent in our discussion below.

Three kinds of productions occur when children with DLD fail to supply the appropriate object clitic. The most common is clitic omission. This appears to be true for all of the languages just discussed, especially during the preschool ages. For somewhat older children with DLD, the direct object noun is sometimes produced, which conforms to the canonical word order but is not pragmatically appropriate. For example, in their study of Greek-speaking children with DLD, Dosi & Koutsipetsidou (2019) found frequent use of responses such as Pleni stin kamilopardali “(She) is washing the giraffe” in contexts where the object clitic tin would be more appropriate, as in Tin pleni “(She) it washing.” Substitutions of one clitic for another have also been reported. Such reports are more frequent for children with DLD acquiring Spanish (e.g., Bedore & Leonard 2001), especially substitutions of the singular in place of the plural, as in Lo compró “(She) it bought” instead of Los compró “(She) bought them.”

There have been both domain-general and domain-specific proposals for why object clitics present such extraordinary difficulty for children with DLD. The possible explanations cover a wide range. One explanation centers on prosody: object clitics are usually unstressed monosyllables that in their most common sentence contexts are vulnerable to omission. In null subject languages such as Greek and several of the Romance languages, this means that the object clitic might ordinarily appear in sentence-initial position and therefore could be vulnerable to initial weak syllable deletion. One piece of evidence consistent with the prosody view is a finding by Leonard and Bortolini (1998). In that study, Italian-speaking children with DLD were more proficient in using clitics in post-infinitive position (an acceptable position for clitics, as in lo in Voglio spingerlo “(I) want to push it”) than when the clitic preceded the finite verb (e.g., Lo spingo “(I) push it”).

However, it is clear that prosody alone is not a sufficient explanation. In several languages with object clitics, reflexive clitics are also used and these are less difficult than object clitics for children with DLD (Arosio et al. 2004, Jakubowicz et al. 1998). Like object clitics, reflexive clitics are weak monosyllables that precede the verb: Compare the Italian reflexive si in Il bambino si lava “The boy himself washes” with the Italian object clitic lo in La mamma lo lava “The mom him washes” (Arosio et al. 2014). Similarly, definite articles in Greek, Spanish, Italian, and French are less problematic than object clitics for children with DLD (Arosio et al. 2014; Bedore & Leonard 2005, Jakubowicz et al. 1998, Smith et al. 2008). Along with their status as weak syllables that are often in prosodically vulnerable sentence positions, some articles are phonetically identical to clitics. Compare, for example, the article le and object clitic le in French: Elle lave le petit garҫon “She washes the little boy” and Elle le lave “She him washes” (Jakubowicz et al. 1998). Despite the similar prosodic characteristics of these different morphemes, the deficits in object clitics are clearly greater.

The differences between the languages in the features reflected in articles are helpful in ruling out some possible sources of difficulty with object clitics. For example, in Romance languages, object clitics express grammatical case but articles do not. However, in Greek, articles do express grammatical case and they are less problematic than clitics for children with DLD.

There are various accounts of the difficulty with object clitics that go beyond the issue of prosody. Some are within linguistic frameworks and relate to syntactic movement, feature interpretability, or grammatical agreement. Others emphasize processing, such as the degree of movement from the clitic’s interpreted position, not the movement itself. Although the accounts are very different, all revolve around the fact that object clitics often require a deviation from canonical word order. And, in their very different ways, each account provides a means of explaining why both reflexive clitics and definite articles are less challenging for children with DLD than object clitics. For example, definite articles are usually adjacent to the corresponding noun and therefore they can have local agreement with the noun in features such as number and gender. Explanations for the higher accuracy for reflexive clitics than for object clitics are somewhat different. Most notably, reflexive clitics might be directly inserted in the preverb position rather than the result of movement from an interpreted position as is often assumed for object clitics (see Arosio et al. 2014).

There is some irony in the fact that measures of object clitic use show excellent diagnostic accuracy in both their sensitivity and specificity (e.g., Arosio et al. 2014, Bortolini et al. 2006) yet we have not yet isolated the factor or factors responsible for children’s extraordinary difficulty with this element of grammar. It seems likely that a clearer understanding of the nature of this difficulty could lead to a more effective intervention to help children with language disorders overcome this weakness.

5. Conclusion

In this paper, we have only touched on the types of questions that might be asked about childhood language disorders in different languages. These questions were obviously shaped by the particular languages we have examined thus far. Insightful scholars will no doubt see other, perhaps more important questions that can be asked within these languages. The main goal of this paper was a broader one, that is, to promote the cross-linguistic study of childhood language disorders across the globe. Although the study of clinical populations across different languages will present challenges, the infusion of child language researchers into this line of work can lead to significant advances in our understanding of these disorders. Child language researchers are nothing if not highly creative and skilled at finding ways to isolate particular aspects of a language for careful study. Through their increasing involvement we can learn more about disorders in each target language and, by synthesizing data across languages, we can move forward in the development of theories with broader explanatory value. We also believe that greater knowledge about language disorders in each language can equip parents, teachers, and professionals with information they can use to help children navigate through a linguistically complex world. Our cross-linguistic endeavors, then, can have practical as well as intellectual rewards.

Declaration of Interest

The authors report no financial or non-financial conflicts of interest.

Acknowledgments

Most cross-linguistic studies are collaborative, and the same is definitely true of the authors’ work on DLD. The authors’ cited studies of DLD—in Cantonese, Finnish, German, Hebrew, Hungarian, Italian, Spanish, Swedish, Turkish, and even English—could never have been accomplished without the expertise and leadership of the authors’ many colleagues. Their contribution to our understanding of DLD cannot be overstated.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

The authors’ studies included in this review were funded by research grants R01 DC00458 and R01 DC009574 from the National Institute on Deafness and Other Communication Disorders of the National Institutes of Health, USA.