Morphological Errors in Monolingual Spanish-Speaking Children With and Without Developmental Language Disorders

Abstract

Purpose

The purpose of this study was to identify which morphological markers have the best diagnostic accuracy to identify developmental language disorders (DLD) in monolingual Spanish-speaking children.

Method

The participants in this study included 50 Spanish-speaking monolingual children with (n = 25) and without (n = 25) DLD. Data collection took place in Mexico. Children were administered a comprehensive elicitation task that set up felicitous contexts to produce morphological structures previously identified as problematic for Spanish-speaking children with DLD: articles, direct object pronouns, adjectives, plurals, verb conjugations, and the subjunctive in Spanish.

Results

Statistically significant group differences between children with and without DLD were found for all morphological structures examined but plurals. Logistic regression analyses suggested that a model that included clitic and verbs was the best model to uniquely predict group membership. This model showed sensitivity of 96% and specificity of 80%.

Conclusion

Clitics and verbs should be considered morphological markers of DLD in monolingual Spanish-speaking children.

Children with developmental language disorders (DLD), also known as “specific language impairment” or “primary language impairment,” experience unexplained limitations in language ability that are not attributable to sensory or neurological deficits (Bishop et al., 2016; Leonard, 1998). These limitations include difficulties in lexical learning, phonology, pragmatics, and particularly morphosyntax. At the morphosyntactic level, children with DLD make errors that appear to be language specific. For example, children with DLD who speak Spanish are known to make morphological errors at the noun phrase level, while English-speaking children with DLD tend to make more errors in the verbal domain (see Leonard, 2014, for a review of morphological errors across languages). This seeming variation in the specific morphosyntactic markers across languages has driven interest in the study of cross-linguistic manifestations of DLD (Leonard, 2013).

In general, Spanish-speaking children with DLD produce shorter, less complex, and more ungrammatical sentences when compared to typically developing (TD) children (Bedore & Leonard, 2005; Restrepo, 1998). Common grammatical errors seen in Spanish-speaking children with DLD include morphological and omission errors with articles, direct object pronouns, verb inflections, auxiliary verbs, adjective agreement, plural inflections, and the subjunctive mood (Bedore & Leonard, 2001, 2005). One important limitation in the Spanish literature is that a lot of what is known about DLD in Spanish-speaking children originates from studies on Spanish-speaking children living in the United States. This is problematic since Spanish-speaking children living in the United States are either active dual language learners of Spanish and English or are growing up with English as an additional ambient language. Since language development in bilingual contexts differs from monolingual language development (e.g., Morgan et al., 2013),it is difficult to know to what extent the characteristics described in these previous studies are specific to either DLD or the bilingual context. Thus, it is crucial to specifically examine the unique grammatical profiles of Spanish-speaking children with DLD growing up in monolingual contexts. The differentiation between typical and atypical language development remains at the forefront of our current priorities. Refining grammatical markers (grammatical characteristic typical of DLD; Rice & Wexler, 1996) that are useful to identify DLD would improve diagnostic accuracy in Spanish-speaking children with DLD and advance diagnostic practices for clinicians working with Spanish-speaking children globally. This study has a clinical focus and aims to find the clinical markers with the best diagnostic accuracy to identify DLD in monolingual Spanish-speaking children.

Grammatical Markers of DLD in Spanish-Speaking Children

The last 20 years have seen a rise in the study of grammatical markers of DLD in Spanish-speaking children. Our review of the literature identifies 16 studies between 2001 and 2017 investigating grammatical markers in Spanish-speaking children with DLD. Most studies describe the grammatical profiles in terms of accuracy rates and provide a description of error patterns (e.g., Bedore & Leonard, 2001). In some cases, the emphasis of the studies is on grammatical errors (Jackson-Maldonado & Maldonado, 2017; Jacobson & Schwartz, 2002). The findings of these 16 studies are summarized by grammatical structure and elicitation method in terms of accuracy rates in Table 1. We calculated effect sizes of the magnitude of the difference using Cohen's (1998) d between children with and without DLD for each grammatical structure to predict the grammatical structures with the best potential to differentiate the groups.

Table 1.

Summary of percentage of correct production across various studies.

Form	Study	Elicitation	Context	n	Age (years;months)	DLD vs. TD	d
Articles	Anderson & Souto (2005)	Spontaneous	Monolingual	11	4;10	85 vs. 98	NSD
	Jackson-Maldonado & Maldonado (2017)	Spontaneous	Monolingual	17	7;3	92 vs. 99	NSD
	Bedore & Leonard (2005)	Spontaneous	Bilingual	15	4;0–5;6	84 vs. 96	1.05
	Simon-Cereijido & Gutierrez-Clellen (2007)	Spontaneous	Bilingual	24	4;6	79 vs. 96	1.05
	Restrepo & Gutierrez-Clellen (2001)	Spontaneous	Bilingual	15	5;9	78 vs. 97	NSD
	Anderson & Souto (2005)	Elicited	Monolingual	11	4;10	64 vs. 95	NSD
	Bedore & Leonard (2005)	Elicited	Bilingual	15	4;0–5;6	47 vs. 83	1.70
	Morgan et al. (2013)	Elicited	Monolingual	7	5;7	54 vs. 81	1.20
	Castilla-Earls et al. (2016)	Elicited	Bilingual	16	6;3	38 vs. 86	2.29
Clitics	Bedore & Leonard (2005)	Spontaneous	Bilingual	15	4;0–5;6	79 vs. 95	0.91
	Simon-Cereijido & Gutierrez-Clellen (2007)	Spontaneous	Bilingual	24	4;6	80 vs. 92	0.09
	De la Mora (2004)	Elicited	Monolingual	10		45 vs. 85	NSD
	Morgan et al. (2013)	Elicited	Monolingual	7	5;7	51 vs. 74	0.88
Plurals	Bedore & Leonard (2005)	Elicited	Bilingual	15	4;0–5;6	38 vs. 80	2.96
	Jacobson & Schwartz (2002) a	Elicited	Bilingual	12	8;1	65 vs. 84	1.33
	Castilla-Earls et al. (2016)	Elicited	Bilingual	16	6;3	34 vs. 64	1.27
	Bedore & Leonard (2005)	Spontaneous	Bilingual	15	4;0–5;6	93 vs. 97	1.13
	Bedore & Leonard (2001)	Elicited	Bilingual	15	4;0–5;6	56 vs. 96	1.96
	Grinstead et al. (2008)	Elicited	Monolingual	9	4;9	89 vs. 94	1.56
Adjectives	Bedore & Leonard (2001)	Elicited	Bilingual	15	4;0–5;6	72 vs. 94	1.27
Subjunctive	Morgan et al. (2013)	Elicited	Monolingual	7	5;7	52 vs. 72	0.78
	Castilla-Earls et al. (2016)	Elicited	Bilingual	16	6;3	45 vs. 75	1.15
Verbs	Simon-Cereijido & Gutierrez-Clellen (2007)	Spontaneous	Bilingual	24	4;6	90 vs. 97	1.20
	Sanz-Torrent et al. (2008)	Spontaneous	Bilingual a	6	3;7–4;1	93 vs. 97	NSD
	Jackson-Maldonado & Maldonado (2017)	Spontaneous	Monolingual	17	7;3	95 vs. 98	NSD
	Grinstead et al. (2013)	Spontaneous	Monolingual	26	5;8	95 vs. 99	NSD
	Grinstead et al. (2009)	Elicited	Monolingual	21	Unknown	82 vs. 95	1.08
	Jacobson (2012)	Elicited	Bilingual	20	7;0	62 vs. 96	1.69
	Contreras González & Soriano Ferrer (2007)	Elicited	Monolingual	27	8;4	54 vs. 84	1.60
	Bedore & Leonard (2001)	Elicited	Bilingual	15	4;0–5;6	65 vs. 89	1.13

Note. Bilinguals here refer to children in the United States who are either bilingual (Spanish–English) or in language contact situation with English. n = sample size for children with DLD; Age = average age is provided with some exception where only age range was available; DLD = children with developmental language disorders; TD = children with typical development; d = Cohen's effect size; NSD = no standard deviations provided in the original article, so d was not calculated.

^a Bilingual Spanish and Catalan.

As can be seen in Table 1, studies that employed elicited language tend to obtain larger effect sizes than studies employing spontaneous language measures. Some of the studies focus on a particular grammatical structure in detail (e.g., verbs in Grinstead et al., 2013; articles in Restrepo & Gutierrez-Clellen, 2001), while other studies employ a more comprehensive approach testing multiple structures at a time (Bedore & Leonard, 2001, 2005; Morgan et al., 2013). Seven of the studies included in Table 1 were conducted with monolingual children living in monolingual countries, while the other eight studies included children living in bilingual contexts. In what follows, we summarize observations for nominal and verbal morphology.

Nominal Morphology

Among the grammatical markers of DLD, direct object pronouns and articles emerge as robust identifiers of DLD in Spanish-speaking children. Most studies found statistically significant differences for elements of the noun phrases between children with and without DLD in both spontaneous and elicited language (Anderson & Souto, 2005; Bedore & Leonard, 2001, 2005; Castilla-Earls et al., 2016; Jacobson & Schwartz, 2002; Morgan et al., 2013; Restrepo & Gutierrez-Clellen, 2001; Simon-Cereijido & Gutierrez-Clellen, 2007). An exception is the study of Jackson-Maldonado and Maldonado (2017), who did not find statistically significant differences in direct object production using a spontaneous language task. Effect sizes range between 1.20 and 2.29 for articles and between 0.88 and 2.96 for direct object pronouns, indicating large effect sizes (Cohen, 1998). Other nominal elements, such as plurals and noun–adjective agreement, show statistically significant differences with large effect sizes, with the exception of the study of Grinstead et al. (2008), who found no statistically significant differences for plurals, but these findings need to be replicated to ensure their validity.

Verbal Morphology

Children with and without DLD showed over 90% verb accuracy in spontaneous language samples. However, the accuracy rates are lower in elicited language showing children with DLD performing around 60% accuracy while TD children are over 80%. Effect sizes for elicited language ranged between 1.08 and 1.69, which are considered large effect sizes (Cohen, 1998). The subjunctive mood was investigated in two studies with large effect sizes (Castilla-Earls et al., 2016; Morgan et al., 2013).

The results summarized here reflect average group differences between children with and without DLD. However, group differences alone are not enough to establish whether a grammatical structure can differentiate between the two groups. Instead, to determine the viability of a diagnostic procedure for the identification of a disorder, it is crucial to study diagnostic accuracy. In these types of studies, the results of a gold standard measure (reference measure) are compared to the diagnostic data of a measure of interest (index measure). The results of the index measure are compared against the reference measure to calculate the amount of agreement between them. This comparison yields measures of sensitivity (ability to detect a language disorder when it is truly present) and specificity (ability to exclude a language disorder when it was truly not present). These measures are vital when determining the clinical usefulness of a diagnostic test. A test with poor specificity incorrectly classifies children with typical development as having a language impairment. Conversely, a test with poor sensitivity incorrectly classifies children with language impairment as typical. A good measure of diagnostic accuracy should have a balance between sensitivity and specificity and be at least 90% accurate (10% margin of error). Diagnostic measures with 80%–90% accuracy are considered fair and with less than 80% accuracy are considered unacceptable for diagnostic purposes (Plante & Vance, 1994).

From the 16 studies summarized in Table 1, only three studies investigated the diagnostic accuracy of grammatical structures (Grinstead et al., 2013; Morgan et al., 2013; Simon-Cereijido & Gutierrez-Clellen, 2007). Simon-Cereijido and Gutierrez-Clellen (2007) examined spontaneous language and reported that articles, verbs, and direct object pronouns in isolation showed fair specificity but poor sensitivity to identify Spanish-speaking children with DLD living in the United States (articles: sensitivity 74%, specificity 89%; direct object pronouns: sensitivity 68%, specificity 84%; verbs: sensitivity 68%, specificity 89%). However, all three grammatical structures combined yielded a sensitivity of 79% and a specificity of 100%, indicating that a pooled measure might have better diagnostic accuracy than isolated structures. Morgan et al. (2013) investigated the diagnostic accuracy of a combined elicited measure that included articles, clitics, subjunctive mood, and derivational morphemes. They reported that this combined measure had a sensitivity of 100% and a specificity of 72%. However, Morgan et al. did not conduct diagnostic accuracy calculations for grammatical structures in isolation. Lastly, Grinstead et al. (2013) investigated the diagnostic accuracy of elicited verb production. They reported that targeting specific verb forms in isolation achieved sensitivity and specificity of 89%. Their findings are very promising considering that they employed a single grammatical structure suggesting that verb conjugation is particularly important for the identification of Spanish-speaking children with DLD. In summary, the results of the studies of diagnostic accuracy of grammatical markers suggest that the best potential markers of DLD with good diagnostic accuracy are a combination of grammatical structures (Morgan et al., 2013; Simon-Cereijido & Gutierrez-Clellen, 2007) or verbs in isolation (Grinstead et al., 2013).

Current Study

Among the potential problematic grammatical structures, direct object pronouns, articles, and verbs have been identified as potentially good identifiers of DLD in Spanish-speaking children because (a) the differences on accuracy rates between TD children and children with DLD are robust and consistent among studies and (b) there is previous evidence that suggest that these measures have good diagnostic accuracy. Other grammatical structures, such as plurals, noun–adjective agreement, and the subjunctive mood, had previously shown statistically significant differences between children with and without DLD but need to be replicated and tested for potential diagnostic accuracy.

The research questions and hypotheses guiding this investigation are as follows:

1. Are there differences in the accurate production of articles, direct object pronouns, verb agreements, noun–adjective agreement, plurals, and the subjunctive mood between Spanish-speaking monolingual children with and without DLD?

   Hypothesis: We predict that all grammatical structures included in this study will show group differences between monolingual Spanish-speaking children with and without DLD.

2. Which grammatical structures are the best markers of DLD in Spanish monolingual children?

Hypothesis: We predict that a combination of articles, direct object pronouns, and verbs will have the best diagnostic accuracy to identify monolingual Spanish-speaking children with language disorders.

Method

Participants

The participants in this prospective study included 50 monolingual Spanish-speaking children living in Mexico City. There were 25 children with TD language skills (TD children) and 25 children with DLD. Fifty-eight percent of the participants were boys. All children passed an otoacoustic emission test and obtained a score of ≥ 70 on the Nonverbal Scale of the Kaufman Brief Intelligence Test–Second Edition (KBIT-2; Kaufman & Kaufman, 2004; see Table 2 for descriptive information). This was a convenience sample of monolingual children with and without DLD.

Table 2.

Descriptive information.

Variable	Group	n	M	SD	Min	Max	t	df	`p
Age in months	TD	25	68.2	10.5	48	81	1.185	48	.242
	DLD	25	65.0	8.7	50	83
KBIT-2	TD	25	97.5	9.4	84	121	1.278	48	.207
	DLD	25	93.9	10.7	78	118
TVIPs	TD	22	117.7	11.4	102	145	5.035	39	> .001
	DLD	19	93.8	18.6	58	135
BESA	TD	25	111.2	5.5	93	118	7.917	48	> .001
	DLD	25	89.0	12.9	58	115
CELF Preschool-2 Spanish	TD	25	110.0	7.9	95	123	5.918	48	> .001
	DLD	25	91.1	13.6	58	125

Note. TD = typically developing; DLD = developmental language disorder; KBIT-2 = Kaufman Brief Intelligence Test–Second Edition; TVIP = Test de Vocabulario en Imagenes; BESA = Bilingual English–Spanish Assessment–Morphosyntax Scale; CELF Preschool-2 Spanish = Core Language Score of the Clinical Evaluation of Language Fundamentals Preschool–Second Edition–Spanish Edition.

Children with DLD met both of the following criteria:

1. Failed the Tamiz de Problemas del Lenguaje (TPL; Auza et al., 2018), which is a screening test standardized with monolingual Mexican children. The TPL includes a subtest of noun phrase morphology and a sentence repetition subtest. We used the prespecified cutoff scores in the TPL by age group. The sensitivity and specificity of the TPL for children between 4 and 6 years of age are over 80%, which is considered acceptable for studies of diagnostic accuracy (Plante & Vance, 1994; sensitivity: age of 4 years 85%, age of 5 years 83%, age of 6 years 96%; specificity: age of 4 years 83%, age of 5 years 81%, age of 6 years 81%).

2. Showed evidence of low grammatical skills in naturalistic language (percentage of grammatical utterances below 80% during a story retell task; Auza & Castilla-Earls, 2017; Restrepo, 1998). There were a small number of children who failed the TPL but had a percentage of grammatical utterances of above 80%. In this case, their utterance length (e.g., mean length of utterance in words) in a story retell task was examined to determine whether the high grammaticality was an effect of little elaboration at the utterance level as suggested by Simon-Cereijido and Gutierrez-Clellen (2007). If their sentence length was below 1 SD from the mean in comparison with the Systematic Analysis of Language Transcripts samples reference database (J. Miller & Iglesias, 2018) for monolingual children, then these children were considered to have DLD.

To further describe the language profiles of these children, we also administered the Spanish Morphosyntax subtest of the Bilingual English–Spanish Assessment (BESA; Peña et al., 2014), the core language subtests of the Clinical Evaluation of Language Fundamentals Preschool–Second Edition–Spanish Edition (CELF Preschool-2 Spanish; Semel et al., 2009), and the Test de Vocabulario en Imágenes (TVIP; Dunn & Dunn, 1997). The TVIP was normed with monolingual children in Mexico, while the CELF Preschool-2 Spanish and the BESA were normed with bilingual children in the United States. When language tests normed in bilingual children are used with monolingual children, there is an inflation effect on standard scores: Monolinguals often score higher than bilinguals in tests standardized with bilingual children (Castilla-Earls et al., 2019; Morgan et al., 2013), although group differences between the TD and DLD groups are still evident (Morgan et al., 2013). Therefore, traditional clinical scores for the identification of children with DLD in bilingual children are not applicable for these measures, but group differences are expected. In this study, we used measures standardized with bilingual children to further describe the language performance of the participants and not for identification purposes.

Descriptive information is presented in Table 2. There were no significant statistical differences on scores of the KBIT-2 or on age between the TD and DLD groups. As expected, the group of TD children outperformed the group of children with DLD on the Spanish BESA Morphosyntax subtest, the CELF Preschool-2 Spanish, and the TVIP. Figure 1 shows the performance on the BESA Morphosyntax and CELF Preschool-2 Spanish by group. At the individual level, all children in the TD group had relatively high scores (above 93 on the BESA and 95 on the CELF Preschool-2 Spanish). Children in the DLD group showed more variable performance on these measures as suggested by the dispersion of their scores. However, most of the children in the DLD group (93%) scored below 100 on both the CELF Preschool-2 Spanish and the BESA. These scores are in alignment with studies reporting a need to increase the cutoff score of bilingual tests by 1 SD to identify monolingual children with DLD (Morgan et al., 2013), although our purpose in using these bilingual test is purely descriptive.

Figure 1.

Plot of individual scores in the Bilingual English–Spanish Assessment (BESA) and the Clinical Evaluation of Language Fundamentals Preschool–Second Edition–Spanish Edition (CELF) by group. TD = typically developing group; DLD = developmental language disorder group.

In terms of vocabulary, children with DLD, as a group, performed significantly poorer than TD children. Due to a test administration error, there was vocabulary data available for 22 children in the TD group and 19 children in the DLD group. All TD children had scores that were equal or higher than 100 on the TVIP, suggesting that their vocabulary skills were within normal limits. In the DLD group, most children performed within normal limits (above 85), although only eight children with DLD had scores above 100 on the TVIP. In addition, there were five children with DLD who had scores below 85 on the TVIP. These results suggest that some of the children in the DLD group had additional deficits in vocabulary development.

Measures

The experimental measure used in this study consisted of an elicitation task that set up felicitous contexts to produce grammatical structures that had been previously identified as problematic for Spanish-speaking children with DLD: articles, direct object pronouns, adjectives, plurals, verb conjugations, and the subjunctive in Spanish. The elicitation task consisted of 48 items that were elicited by asking questions about a picture (direct object pronouns, articles, adjectives, and the subjunctive mood) or asking the child to complete a sentence related to a picture (verbs and plurals). Children were provided three training items per task.

Direct Object Pronouns

The elicitation of direct object pronouns follows the classic elicitation method described in the study of Schaeffer (1997) and used previously in the studies of Castilla (2008) and Castilla-Earls and Pérez-Leroux (2010). To elicit direct object pronouns, we presented a picture and asked a question that obliged the use of a direct object pronoun. For example, Qué le hace la mamá a la niña? (“What does the mom do to the girl?”), with an accompanying picture of an adult female combing the hair of a girl. In this case, the target/correct response is the pronoun la (SgFem) with any accompanying verb in either the pronominal or the postnominal position. Correct responses for this item included la peina (“he/she combs her”), la arregla (“he/she gets her ready”), and peinándola (“combing-her”). Incorrect responses included object omissions peinando (“combing”) and arreglando (“getting ready”), gender error Esta peinándolo (“he/she is combing him”), and number error Está peinándolas (“he/she is combing them-fem”). Other incorrect responses included unscorable responses such as unintelligible statements, no responses, and other unrelated responses.

Articles and Adjectives

To elicit noun phrases that included both articles and adjectives, we presented pictures with two alternatives (e.g. red birds and yellow birds) and asked a question that forced the child to choose one of the alternatives. For example, Estos animales iban volando. Mira, estos se le pararon al niño. Cuáles se le pararon al niño? (“These animals were flying towards the child. Look, these ones landed on the child. Which ones landed on the child?”). If the child answered los pájaros (“the birds”), a follow-up question was introduced Cuáles? (“which ones?”) to facilitate the production of an adjective. Correct responses to articles included any article in agreement with the noun produced by the child los pájaros rojos (“the red birds”), los pájaros (“the birds”), el pájaro (“the bird”), and los rojos (“the red ones”). Incorrect responses for articles included article omissions pájaros (“birds”), gender errors la pájaro (“the-Sgfem bird”), and number errors el pájaros (“the-MasSg birds”). For adjectives, correct responses included any adjective in agreement with the noun los pájaros rojos (“the red birds”) and los rojos (“the red ones”), and incorrect responses included omissions el pájaro (“the bird”) and gender and/or number substitution el pájaro roja (“the red-fem bird”) and los pájaros rojo (“the red-Sg birds”). Unscorable responses (unintelligible utterances, no responses, and other unrelated responses) were also considered incorrect.

Verbs

To elicit verbs in this study, we employed a sentence completion task using pictures. In this task, we presented a coordinated sentence in which the second clause needed to be completed by adding a verb. In the first clause of the sentence, we provided a subject and a verb to set the tense, and in the second clause, we provide a noun to be conjugated to complete the sentence. For example, we showed a picture of a boy running and two girls dancing with the following sentence: Nos estamos divirtiendo mucho. Juan corre y nosotras… (“We are having fun. Juan runs and we…”). Correct responses included any verb conjugated in the second-person plural and in the present tense bailamos (“dance-plural”) and caminamos (“walk-plural”). Incorrect responses included tense or person errors such as bailar (“to dance”) and bailó (“danced-sing”). Unscorable responses, as described previously, were also considered incorrect. This task included five items in the first person (two singular, three plural), four items in the third person (two singular, two plural), and one item in the second/singular person. In terms of tense, seven items were presented in the preterite, and three were presented in the present tense.

Subjunctive

The elicitation of the subjunctive mood included very short stories to force the production of the subjunctive in object complement and temporal clauses. For example, there is a picture of a mother visibly upset with a girl in front of a mirror with a toothbrush in her hand. Qué quiere la mamá que haga la niña? (“What does the mom want the girl to do?”). Correct responses included any use of the subjunctive mood (Que se lave sus dientes/“that she washes-subj her teeth”). Incorrect responses included any response using the indicative mood lava (“washes-ind”), person/tense errors lavar (“to wash”), and unscorable responses.

Plurals

To elicit plurals, we showed pictures with single and multiple objects and asked children to complete a sentence. For example, we showed a single train picture while saying aquí hay un tren (“here there is a train”) and a picture of three trains with y aquí hay tres… (“and here there are three…”). The target response was trenes (“trains”). Incorrect responses included trens (“trains-incorrect plural”), tren (“train”), and unscorable responses.

Procedure

This study was approved by institutional review boards in both the United States (State University of New York–Fredonia) and Mexico (Hospital General Dr. Manuel Gea González). Parents provided consent for their child's participation, and children provided assent. Data collection took place between 2016 and 2017. Testing consisted of two sessions of approximately 50 min each, within a 3-week period. Testing took place in the children's school or at a convenient location for the parents. The first session included the administration of the hearing screening, the KBIT-2 Nonverbal Scale, the CELF Preschool-2 Spanish, the Morphosyntax subtest of the BESA, and the TVIP. The second session consisted of the Spanish retelling task and the experimental measures. The order of tasks within a session were administered randomly, with the exception of the hearing screening and the KBIT-2, which were always administered first. Native Spanish-speaking research assistants administered all the testing in Spanish, and responses to the elicitation and story retell tasks were audio-recorded.

Research assistants who were blind to the language status of the children transcribed and coded all data to minimize subjective bias (Dollaghan & Horner, 2011). Systematic Analysis of Language Transcripts (J. Miller & Iglesias, 2018) was used to transcribe and code the language samples. Regarding the elicitation task, all responses were audio-recorded during assessment and later transcribed by a research assistant. We calculated transcription reliability for the transcription task at the item level. Two different research assistants transcribed the responses for 10 randomly selected audios (20%). Transcription agreement per item of the elicitation task was established if both transcriptions had the same coding. This was considered a stricter protocol than traditional transcription at the word level, which might underestimate errors related to the grammatical structure of interest. The transcription reliability for the elicitation task was 96%. The coding of all responses was done in two steps. First, a research assistant coded all responses as correct or incorrect. Incorrect responses were classified by type of error. Second, the first author of this study reviewed all coding in its entirety. Disagreements between the two raters were discussed and solved.

Results

To examine group differences, we compared the correct production of articles, direct object pronouns, verb agreement, noun–adjective agreement, plurals, and the subjunctive mood in monolingual children with and without DLD. Accuracy percentages are presented in Table 3. Table 3 also includes omissions, substitutions, and other responses for informative purposes. However, all analyses in this study were based on accuracy. Children without DLD outperformed those with DLD on all grammatical structures but plurals. We used an adjusted p value of .0083 to control for Type 1 error (see Table 4). In examining the effect sizes as indexed by Cohen's d, we found large mean differences, with the smallest mean difference being 1.23 for the subjunctive mood and the largest being 2.41 for direct object pronouns.

Table 3.

Response categories by group.

Morphological structure	TD				DLD
	Correct	Substitutions	Omission	Unscorable	Correct	Substitution	Omission	Unscorable
Clitics	84.4 (14.5)	6.0 (9.6)	1.2 (4.4)	2.0 (4.1)	32.5 (26.9)	12.0 (13.4)	20.4 (24.4)	22.4 (28.7)
Articles	95.7 (11.4)	2.0 (5.9)	0.5 (2.5)	2.5 (8.1)	69.6 (22.0)	8.5 (11.8)	8.6 (10.1)	13.3 (14.6)
Adjectives	90.5 (13.6)	0.5 (2.5)	5.5 (9.6)	3.5 (8.5)	66.2 (22.4)	3.0 (5.4)	12.5 (13.5)	18.3 (15.6)
Plural	83.2 (17.9)	8.0 (10.0)	6.4 (11.1)	2.4 (8.8)	63.2 (33.5)	10.4 (13.1)	19.2 (26.8)	7.2 (14.0)
Subjunctive	72.0 (31.1	16.0 (23.8)	NA	12 (16.3)	33.6 (31.5)	27.2 (23.0)	NA	39.2 (31.3)
Verbs	78.4 (14.9)	8.8 (15.1)	NA	12.8 (14.0)	49.6 (21.5)	12 (31.7)	NA	38.4 (42.8)

Note. Standard deviation in parentheses. TD = typically developing; DLD = developmental language disorder; NA = not applicable.

Table 4.

Group comparisons.

Variable	Difference	t	df	p	Cohen's d
Verbs	28.80	5.50	48	< .0001	1.56
Articles	25.36	5.12	48	< .0001	1.45
Clitics	51.47	8.53	48	< .0001	2.41
Subjunctive	38.40	4.34	48	< .0001	1.23
Adjective	24.29	4.63	48	< .0001	1.31
Plurals	20.00	2.63	48	.0114	0.74

To find the best grammatical markers of DLD in Spanish monolingual children, we fit a series of logistic regression models to find a predictor set that most accurately predicted group membership. Therefore, our analyses were exploratory. The models tested included (a) clitics only; (b) articles only; (c) verbs only; (d) verbs and clitics; (e) verbs and articles; (f) article and clitics; (g) verbs, articles, and clitics; (h) verbs, articles, clitics, and subjunctive; (i) verbs, articles, clitics, subjunctive, and adjectives; and (j) verbs, articles, clitics, subjunctive, adjectives, and plurals. There were no missing data. To determine the best fitting model, we considered two criteria: statistical and clinical. Because Models 1–10 involved the inclusion or exclusion of predictor variables, all models could be considered to evaluate the same model, but certain models constrain a parameter estimate to be zero. As such, these models are nested. Therefore, the statistical criteria for best model selection included the significance of the predictors, the χ² test of model fit, and the Akaike information criterion (AIC). The clinical criteria included the best possible sensitivity and specificity with the simpler/fewer grammatical elements possible.

The results of these models can be seen in Table 5. In Models 1–5, we found verbs, articles, and clitics all to be significant. However, when articles and clitics were entered together (Model 6), articles were no longer significant, suggesting that articles did not uniquely predict group membership. This pattern of article nonsignificance was consistent in Models 7–10. The addition of subjunctives in Model 8, adjectives in Model 9, and plurals in Model 10 did not make further changes in the significance of the predictors: The only two predictors that uniquely predicted group membership were verbs and clitics. In conducting the nested model comparisons, we found significant χ² test when comparing Model 4 to Models 1–3. These significant χ² tests indicate that the additional parameters estimated in Model 4 improve model fit. Additionally, we found all χ² tests for Model 4 with Models 7–10 to be nonsignificant, indicating that the additional predictor variables in Models 7–10 did not significantly improve model fit. Models 5 and 6 could not be compared to Model 4 because they estimated the same number of parameters; therefore, we compared the −2 log likelihood of models: Model 4 had the lowest −2 log likelihood indicating better fit. Thus, based on this criterion, we found Model 4 to be the best fit to the data. Next, we examined the AIC, which is an index of absolute fit. Again, Model 4 had the lowest AIC value indicating that, according to this index, Model 4 was the best fit to the data.

Table 5.

Model comparisons.

Model	Parameter	Estimate	SE	χ2	p	Sensitivity	Specificity	2LL	LL ratio	AIC	Prob χ2
1	Age	–0.03	0.05	0.27	.603	100	76	28.92	40.39	34.92	.990
	Clitics	0.10	0.03	12.75	.000
2	Age	–0.01	0.04	0.04	.847	80	80	48.4	20.91	54.4	.999
	Articles	0.08	0.02	11.94	.001
3	Age	0.00	0.04	0.01	.919	92	64	44.4	24.91	50.4	.999
	Verbs	0.09	0.03	13.02	.000
4	Age	–0.11	0.09	1.56	.211	96	80	22.32	47.0	30.32	NA
	Verbs	0.09	0.04	4.77	.029
5	Clitics	0.10	0.04	7.28	.007
	Age	–0.03	0.05	0.52	.472	96	68	37.24	32.08	45.24	NA
	Verbs	0.08	0.03	8.17	.004
	Articles	0.07	0.03	5.65	.018
6	Age	–0.03	0.05	0.40	.529	96	80	27.94	41.38	35.94	NA
	Articles	0.03	0.03	0.93	.334
	Clitics	0.09	0.03	9.00	.003
7	Age	–0.11	0.09	1.56	.212	96	80	22.22	47.1	32.22	.248
	Verbs	0.09	0.04	4.16	.041
	Articles	0.01	0.05	0.10	.757
	Clitics	0.09	0.04	5.24	.022
8	Age	–0.14	0.12	1.44	.230	96	80	22.02	47.3	34.02	.139
	Verbs	0.08	0.04	4.10	.043
	Articles	0.01	0.05	0.09	.759
	Clitics	0.09	0.04	4.80	.028
	Subjunctive	0.01	0.03	0.21	.651
9	Age	–0.18	0.14	1.63	.201	96	80	21.67	47.65	35.67	.115
	Verbs	0.09	0.04	4.12	.042
	Articles	0.00	0.05	0.00	.969
	Clitics	0.08	0.04	4.39	.036
	Subjunctive	0.02	0.03	0.23	.634
	Adjective	0.04	0.06	0.33	.565
10	Age	–0.22	0.16	1.96	.161	96	76	20.85	48.46	36.85	.168
	Verbs	0.10	0.05	4.43	.035
	Articles	0.00	0.06	0.00	.988
	Clitics	0.09	0.04	4.29	.038
	Subjunctive	0.02	0.03	0.24	.624
	Adjective	0.05	0.08	0.42	.515
	Plurals	–0.03	0.03	0.72	.395

Note. χ² are for all models versus Model 4. Degrees of freedom for all models = 1. 2LL = log-likelihood ratio; LL = likelihood ratio; AIC = Akaike information criterion; NA = not applicable.

We also considered the clinical significance of the various models examined. Models 4, 6, 7, 8, and 9 all had the same diagnostic accuracy (sensitivity = 96% and specificity = 80%). However, Model 4 had the fewest predictors (i.e., would have the lowest testing burden for children); therefore, we determined that Model 4 was the best model following our clinical criteria. Thus, taken together, both statistical and clinical criteria for selecting the best fitting model pointed to Model 4, which included clitics and verbs as the best set of grammatical markers to predict DLD in Spanish monolingual children.

Model 4 had a positive likelihood ratio of 4.8 and a negative likelihood ratio of 0.05. Dollaghan's (2007) rule of thumb to interpret likelihood ratios is that a diagnostic measure with a positive likelihood ratio over 10 is clinically informative to identify a child with a language disorder and a measure with a negative likelihood ratio of 1 or below is clinically informative to rule out the presence of a language disorder. Therefore, the likelihood ratios found in this study suggest that a combination of clitics and verbs is clinically informative to rule out the presence of DLD and suggestive to identify a Spanish monolingual child with DLD.

As Model 4 was deemed the best predictive model, we employed the results of Model 4 to identify cutoff scores for clitics and verbs. Number of correct combined items (clitics + verbs) were used for these cutoff scores. The maximum total number of items was 20 (10 clitic items and 10 verb items). Children between the ages of 48 and 54 months were classified as having DLD if they had at nine items correct or less (i.e., cutoff score is 9 items), whereas children in the same age range were classified as TD if they had 10 or more items correct. Cutoff scores for the remaining ages were 10 items at 55–60 months, 11 items at 61–72 months, and 12 items at 78–82 months.

Discussion

The goal of this study was to comprehensively examine the various morphological structures that have been previously identified to be problematic for Spanish-speaking children in a group of monolingual children with and without DLD to find the grammatical markers with the best diagnostic accuracy. We found group differences with large effect sizes for articles, clitics, adjective agreement, verb agreement, and the subjunctive mood with monolingual Spanish-speaking children without DLD demonstrating higher accuracy rates than children with DLD. Although most of the grammatical structures revealed group differences, a combination of clitics and verbs uniquely predicted group membership with acceptable sensitivity and specificity. These findings will be discussed in terms of group differences, diagnostic accuracy, and the relevance of these findings for the identification of morphological markers of DLD for Spanish-speaking children.

Group Differences

Article and direct object pronouns are the most studied morphological structures for Spanish-speaking children with DLD (see review in Table 1). As predicted, this study found statistically significant group differences for both articles and direct object pronouns. These results are in agreement with previous studies investigating articles (Anderson & Souto, 2005; Bedore & Leonard, 2001, 2005; Castilla-Earls et al., 2016; Morgan et al., 2013; Restrepo & Gutierrez-Clellen, 2001; Simon-Cereijido & Gutierrez-Clellen, 2007) and direct object pronouns (Bedore & Leonard, 2001, 2005; Castilla-Earls et al., 2016; Morgan et al., 2013; Simon-Cereijido & Gutierrez-Clellen, 2007). In terms of the magnitude of the group difference, both articles and direct object pronouns had effect sizes within the range of previously reported studies (articles: 1.45, range: 1.2–2.3; clitics: 2.41, range: 0.88–2.96), which suggests that the group differences in articles and clitics observed across the various studies are relatively stable in terms of size. These results confirm the results of previous studies suggesting that both articles and direct object pronouns show large group differences between Spanish-speaking children with and without DLD.

Also, at the noun phrase level, our results supported our hypothesis for group differences for adjectives but not for plurals. These two morphological structures had been studied considerably less than article and direct object pronouns, so replication of findings was of importance. These results are in agreement with the results for adjectives but in disagreement with the results for plurals in the study of Bedore and Leonard (2001), who studied bilingual children. The current results about plurals are in agreement with Grinstead et al.'s (2008) data on monolingual children, in that we did not find statistically significant differences. It is important to note that Grinstead et al. (2008) found an effect size of 1.56 for plurals, but their sample size was small (n = 9), which probably affected the statistical significance of their results. It is also possible that plurals are not as problematic for Spanish-speaking monolingual children in comparison to bilingual children.

At the verb phrase level, this study found significant differences for both the subjunctive and verb agreement. These results are in agreement with previous studies reporting group differences between children with and without DLD for the subjunctive mood (Castilla-Earls et al., 2016; Morgan et al., 2013) and verb agreement (Bedore & Leonard, 2001; Grinstead et al., 2009, 2013; Jacobson & Schwartz, 2002). The effect sizes found in this study (subjunctive: 1.23; verbs: 1.56) were slightly higher than the effect sizes reported for the subjunctive in previous studies and within the range of previously reported studies for verb agreement. The difference in the subjunctive might be due to the type of subjunctive used in the various studies. In general, the results for the verb phrase replicated previous studies. These results are of particular importance because the errors seen in DLD in Spanish have been traditionally described more in terms of noun phrase errors than verb phrase errors. This is the first study investigating various elicited morphological structures, including verbs, in monolingual Spanish-speaking children. Therefore, the differences in verbs at the group level seen in this study highlight the importance of examining verb production in Spanish-speaking children who are suspected to have DLD.

Although error patterns were not the emphasis on this investigation, the results in Table 3 provide some interesting insights about potential differences between the groups. In general, children with DLD produce a high percentage of unscorable responses (no responses, unintelligible responses, and unrelated responses). In addition, the type of error varied by structure. For example, clitic omission seem to be the most prevalent error for children with DLD (about 20%), while it is almost not existent in TD children (1%). These results warrant further investigation to explore the possibility that the type of error might add information about the clinical status of a child. Current efforts in our research laboratories are examining this possibility in detail using a complex error coding that is distinct to each grammatical structure to truly tap into its morphological complexity.

Diagnostic Accuracy of Morphological Structures

Our results suggest that, in isolation, articles were the only morphological structure with acceptable diagnostic accuracy (sensitivity and specificity of 80%). Clitics in isolation showed excellent sensitivity but poor specificity. Similarly, verbs showed good sensitivity but poor specificity. These results indicate that performance on both articles and verbs works better to identify children with a language disorder than to exclude those with typical language skills at this age when children are still acquiring verbal and nominal morphology. These results are similar to the results of Simon-Cereijido and Gutierrez-Clellen (2007) in that articles, clitics, and verbs showed lower diagnostic accuracy when tested separately. However, Simon-Cereijido and Gutierrez-Clellen found that when articles, verbs, and clitics were tested separately, they had higher specificity but poor sensitivity. The difference in these results might be due to the fact that Simon-Cereijido and Gutierrez-Clellen examined spontaneous language while we examined elicited language. However, our results differed from the results of Grinstead et al. (2013) in that verb accuracy in isolation did not have good diagnostic accuracy. The examination of morphological structures separately shows that no single morphological structure is an ideal marker of DLD in monolingual children, and therefore, pooling morphological structures might be a better approach for assessment and identification of DLD.

Pooling morphological structures also reveals important traits about the morphological structures. In this study, when articles and clitics were entered together in Model 6, articles were no longer statistically significant in their prediction of group membership. In fact, in all models where articles and clitics were entered together, articles were not significant. This was due to the fact that articles and clitics shared a significant amount of variance (r = .645), and when entered together, clitics explain variability above and beyond the variability explained by articles. This is consistent with previous work showing the similarities between articles and clitics in terms of developmental timing and phonological characteristics, but also the differences in terms of their developmental trajectories (Pérez-Leroux et al., 2012).

A combination of verbs and clitics was the best approach to improve diagnostic accuracy in this study with monolingual Spanish-speaking children (sensitivity 96% and specificity 80%). Although other models produced the same diagnostic accuracy levels (Models 7, 8, and 9), the model with clitics and verbs fitted the data better and was the most parsimonious. Therefore, it was deemed the best model to differentiate the groups. This study replicated previous findings in Simon-Cereijido and Gutierrez-Clellen (2007) in that articles, clitics, and verbs showed better diagnostic accuracy when pooled together than individually, although they used spontaneous language and studied bilingual children. The other morphological structures investigated in this study, such as adjectives, plurals, and subjunctive, did not further improve the diagnostic accuracy of the models examined in this study. These results are of clinical relevance because it is critical to distinguish the nuances of results at the group level versus diagnostic accuracy.

Morphological Markers of DLD in Monolingual Spanish-Speaking Children

Rice and Wexler (1996) defined a clinical marker as a morphological characteristic that is typical of DLD. These morphological markers in Spanish have been traditionally identified using group comparisons in terms of accuracy (Bedore & Leonard, 2001, 2005). An evidence-based assessment approach should rely on the use of analysis of diagnostic accuracy (Simon-Cereijido & Gutierrez-Clellen, 2007). Each of these approaches offer related but different results. Group differences show that, on average, a group of TD children perform better than children with DLD. However, variation among the profiles of the children in a group could indeed exist. Although group differences are of importance for general description of language abilities, they are not enough to identify whether performance on a specific marker is typical of a group. Examination of diagnostic accuracy reveals whether performance on a morphological structure can differentiate groups of children with and without language disorders. This approach seems to be more appropriate for the identification of morphological markers of DLD disorders in Spanish-speaking children.

We believe that both clitics and verbs should be considered markers of DLD in Spanish-speaking monolingual children for two reasons. First, these two morphological markers taken together accurately identified 96% of children with language disorders and 80% of children with typical development, which is considered acceptable in terms of diagnostic accuracy (Plante & Vance, 1994). The high accuracy found in this study is not trivial, considering that performance on two morphological structures identified 88% of the cases correctly. Second, the results of this study replicate previous finding in the literature that identify these morphological structures as markers of DLD in Spanish-speaking children.

One might argue that the results seen in this study are somewhat circular since children were included in this study because they either had or lack morphosyntactic errors using global measures of morphosyntax, and therefore, the fact that morphological errors differentiate the groups is not surprising. However, it is important to note that the goal of this study was to identify the best morphological structures to differentiate the groups since this has clinical and theoretical implications that are discussed below. Nonetheless, this points to a crucial limitation of this study because our results do not generalize to children with DLD whose deficits are not specific to the morphosyntactic domain. It is critical to consider that children with DLD are a heterogeneous group with difficulties in various language domains, not only morphosyntax (for a thorough review of current definitions and views of DLD, see Bishop et al., 2016). For example, the results of this study would not be applicable to children with semantic and word-finding difficulties. Although there is no current consensus on the possibility of subtypes of DLD (Bishop, 2017), it is essential that the results of this study are interpreted in light of a profile of DLD morphosyntactic deficit.

The results of this study can be applied to the identification of monolingual Spanish-speaking children with DLD. The data in this study could be used by speech-language pathologists for the assessment of language skills, more specifically for morphological development. Therefore, these data have the potential to advance diagnostic practices for clinicians working with monolingual Spanish-speaking children globally. However, careful considerations should be in place in regard to the application of these results to bilingual populations. Morgan et al. (2013) have shown that it is not appropriate to use monolingual norms for the assessment of bilingual children. Our current research efforts are focused on the comparisons of the diagnostic accuracy of Spanish morphological structures between monolingual and bilingual children to directly contribute to the identification of DLD in children growing up in bilingual contexts. Another limitation of the clinical application of these results is the potential for dialectal differences. There are data that suggest that other Spanish dialects show different patterns in some of the grammatical structures studied here. For example, children who use /s/-leniting dialects of Spanish might show different results for plurals (K. Miller, 2014). Similarly, children who use Andean Spanish might show a normal pattern of clitic omission (Stewart, 1999). Clinicians should be aware of dialectal differences when examining the applicability of these results to children in their caseloads.

Conclusion

Spanish monolingual children with DLD differed from Spanish monolingual children with typical language skills in their accurate elicited production of articles, clitics, adjectives, verbs, and the subjunctive mood. The accurate production of clitics and verbs discriminated Spanish monolingual children with DLD from their counterparts with typical language skills with a sensitivity of 96% and a specificity of 80%. The likelihood ratios found in this study suggest that a combination of clitics and verbs is clinically informative to rule out the presence of DLD in a Spanish monolingual child and is suggestive to identify a Spanish monolingual child with DLD.

Funding Statement

Research reported in this publication was supported by National Institute on Deafness and Other Communication Disorders Award R15DC013670 granted to Anny Castilla-Earls. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.