The Use of Tense and Agreement by Hungarian-Speaking Children with Language Impairment

Abstract

Purpose

Hungarian is a null-subject language with both agglutinating and fusional elements in its verb inflection system, and agreement between the verb and object as well as between the verb and subject. These characteristics make this language a good test case for alternative accounts of the grammatical deficits of children with language impairment (LI).

Method

Twenty-five children with LI and 25 younger children serving as vocabulary controls (VC) repeated sentences whose verb inflections were masked by a cough. The verb inflections marked distinctions according to tense, person, number, and definiteness of the object.

Results

The children with LI were significantly less accurate than the VC children, but generally showed the same performance profile across the inflection types. The types of errors were also similar in the two groups.

Conclusions

Accounts that assume problems specific to agreement do not provide an explanation for the observed pattern of findings. Although the findings are generally compatible with accounts that assume processing limitations in children with LI, one such account, the morphological richness account, was not accurate in all of its predictions. One non-morphosyntactic factor -- the retention of sequences of sounds – appeared to be functionally related to inflection accuracy and may prove to be important in a language with numerous inflections such as Hungarian.

Children with language impairment (LI) show significant deficits in language ability without accompanying deficits such as hearing impairment, neurological damage, or mental retardation. Although children with LI represent a heterogeneous population, common profiles can be identified. In English, for example, a very common profile is a mild to moderate deficit in lexical skills and a more serious deficit in morphosyntax. Within the area of morphosyntax, the use of tense and agreement morphemes seems to be especially problematic.

One complicating factor in the study of LI is that a common profile in one language is uncommon or even absent in another language. For example, word order errors are common in Swedish and German, but not in English. In Italian, verb inflections that express agreement with the subject are not among the areas of special difficulty, unlike the case for English.

Proposals for these cross-linguistic differences are beginning to emerge in the literature. Following a brief review of these proposals, we will describe a study employing Hungarian, a language that represents an excellent test case for the suitability of these alternative proposals. Hungarian differs from other languages studied by LI researchers in key respects. One characteristic is the agglutinating morphology with respect to tense and agreement, where an inflection marking tense is followed by an inflection marking agreement, both attached to the verb stem. A second important characteristic of Hungarian is the fact that verb inflections agree with both the subject (in person and number) and the object (in definiteness). As will be seen below, these characteristics have implications for current accounts of the morphosyntactic difficulties seen in LI.

Recent Accounts of Morphosyntactic Deficits in LI

Morphological richness

The “morphological richness” account has evolved from the findings of Leonard and his colleagues (Leonard, Sabbadini, Leonard, & Volterra, 1987; Leonard, 1998, pp. 255–257; Dromi, Leonard, Adam, & Zadunaisky-Ehrlich, 1999). According to this account, extraordinary difficulties with tense and agreement morphemes are the result of an interaction between a more general limitation in language ability and the properties of the particular system of grammar that must be learned. Key details of the morphological richness account were inspired by the Competition Model (e.g., MacWhinney, 1987; Bates & MacWhinney, 1989), such as the views that languages differ in the details of grammar that have the greatest cue validity, that the discovery and use of these cues are probabilistic in nature, and that some cues have greater processing cost than others.

An important assumption of the morphological richness account is that children with LI have a limited processing capacity. For languages such as English, this limitation can be problematic for the learning of grammatical morphology. Inflections are sparse in English and bare stems are frequent. Faced with a limited processing capacity, then, children with LI might devote their limited resources to the more prevalent information conveyed by word order. Fewer resources would remain for the learning of grammatical morphology, requiring more encounters with grammatical morphemes before they can be learned. In contrast, children with LI acquiring languages with a rich inflectional morphology are expected to devote their limited resources to this area of the grammar. Thus, differences in the use of grammatical morphology between these children and their typically developing peers will be smaller than in a language such as English. It is for this reason that the account gets its name of “morphological richness.”

However, if the inflections themselves reflect a complex combination of grammatical dimensions (e.g., tense, number, person, gender), problems can arise even in the area of inflections in a language with a rich morphology. The more dimensions children must consider simultaneously, the greater the demands on their limited processing capacity. These demands can result in incomplete processing, requiring more encounters with the inflection before it can become a stable part of the children’s grammar. Based on findings from Italian and Hebrew, Leonard (1998) proposed that children with LI may approach their processing limitations when four dimensions must be considered simultaneously. According to Leonard, incompletely processed inflections are the functional equivalent of inflections with low frequency of occurrence because they are not registered consistently, and therefore do not achieve sufficient strength in the child’s grammar to be retrieved as reliably as can be accomplished by typically developing children. Given that children with LI must have a greater number of encounters with each inflection before it is sufficiently established to be retrieved for production with facility, the frequency of occurrence of the inflection in the input is an important factor in the morphological richness account. It is predicted that accuracy will be greater for inflections that are encountered more frequently in the input.

The morphological richness account’s focus on the number of dimensions in an inflection system differs from an approach such as the Competition Model in that the latter places an emphasis on cue validity. Thus, an inflection that reflects a complex combination of four dimensions would be expected to be challenging for children with LI according to the morphological richness account, but if that inflection has high cue validity, the number of dimensions would play a much smaller role according to the Competition Model.

Another assumption of the morphological richness account is that if errors occur, the substitute inflection is expected to share features with the inflection that it replaces. That is, the chief competitors of an appropriate inflection that is only weakly represented in the child’s grammar will be inflections that are similar in their features but with greater strength. In many instances, this may be a “near-miss” error – an inflection that possesses most but not all features reflected in the correct form (e.g., Dromi et al., 1999; Bedore & Leonard, 2001). For example, a third person plural form in the past might be replaced by a third person plural form in the present or a third person singular form in the past. Children with LI are not expected to resort to a default form. For example, it is not expected that a third person singular form in the present will be used as a general substitute without regard to the features reflected in the correct form.

Agreement deficit

Clahsen and his colleagues (Clahsen, Bartke, & Göllner, 1997; Clahsen & Hansen, 1997; Clahsen & Dalalakis, 1999; Eisenbeiss, Bartke, & Clahsen, 2005) have proposed that children with LI have a selective syntactic deficit that affects agreement in particular. These investigators adopted Chomsky’s (1995) distinction between interpretable and non-interpretable features, and posited that in LI, the verb’s non-interpretable features are not properly acquired. Even in null-subject languages, subject-verb agreement is posited to be problematic (Clahsen & Dalalakis, 1999). Errors are expected to be productions of default forms, such as the production of a present third person singular inflection in contexts that obligate a different inflection. The agreement deficit account does not predict difficulties with tense.

Non-Morphosyntactic Language Processing Factors

The morphological richness account is concerned with processing limitations within the scope of morphosyntactic learning and use. This emphasis is well placed, of course, given the striking limitations that children with LI exhibit in this area of language. However, other important areas are important in LI, and these may have at least an indirect, negative impact on morphosyntactic ability. Bishop, Adams, and Norbury (2006) have identified two fundamental impairments in children with LI that are both heritable yet show minimal etiological overlap (see also Conti-Ramsden, 2003). Not surprisingly, one of these is a reduced ability to carry out grammatical computations. The behavioral measure most frequently used to identify this limitation is a test of morphosyntactic ability, including the use of tense and agreement morphemes (e.g., Rice & Wexler, 2001). The other fundamental impairment is a deficit in the ability to retain sequences of speech sounds for brief periods of time. Nonword repetition tasks constitute the most frequent measures for this type of problem (e.g., Gathercole, Willis, Baddeley, & Emslie, 1994).

Although an ability to retain sound sequences of sounds is often associated with word learning (e.g., Gathercole & Baddeley, 1993), it should be clear how limitations in the ability to retain sound sequences could also play havoc with the learning of inflections. If a child cannot retain a sequence that represents an inflection that marks tense and agreement, it is likely that the acquisition of this inflection will be delayed. To the degree to which the inflection system of a language contains many different sequences, the detrimental effect of this retention problem could be considerable. This influence could occur even though retention of sound sequences and grammatical computation are genetically and etiologically distinct. First, as noted byBishop et al. (2006), many children with LI have a double deficit – a deficit in both of these areas. Second, although poor retention of sound sequences appears to be a deficit distinct from a deficit in grammatical computation, if the inflection system of a language involves many different sequences, each of which must be detected and retained by the child, the functional relationship between these two areas may be stronger than in a language such as English.

The Contribution of Hungarian

Hungarian possesses characteristics that make it extremely useful for evaluating the morphological richness and agreement deficit accounts. Research on LI in this language, then, might not only contribute to the development of clinical assessment and treatment methods for Hungarian-speaking children with LI, but also contribute to theory development or refinement. We provide a more detailed description of the structure of Hungarian tense and agreement morphology in the next section. However, some of the highlights of Hungarian and its relevance to these accounts of LI can be stated here. Hungarian is a null-subject language with inflections for tense and inflections that simultaneously mark agreement with the subject in person and number and agreement with the object (if any) in definiteness.

The agreement deficit account assumes that the difficulty with agreement resides in the agreement features of the verb. Therefore, even in a null-subject language such as Hungarian, agreement inflections will be difficult for children with LI. This may be especially so given that agreement is of two different types – agreement between the subject and verb, and agreement between the verb and the object. Errors of agreement are expected to be default forms such as present third person singular. However, tense features are not affected; for this reason, errors on the tense marking of inflections are not predicted.

According to the morphological richness account, children with LI acquiring a language such as Hungarian, in which inflectional morphology plays a central role, will differ from typical peers to a lesser extent than in a language such as English. However, this account explicitly predicts that the processing capacity of children with SLI will begin to reach its limits when four dimensions must be considered simultaneously as in Hungarian, in which tense, person, number, and definiteness play a role in the verb inflection system. Errors should not be default forms; rather inflections that differ from the correct inflection by only a single feature (e.g., present first person singular indefinite in place of present first person plural indefinite) should be the most likely. Accuracy will be greater for inflections with higher frequency of occurrence in the language.

Hungarian is also a highly suitable language to evaluate the role that limitations in the ability to retain sound sequences might play in the use of tense and agreement inflections by children with LI. Although problems in nonword repetition are notorious in this population, their effects on tense and agreement inflection use has not yet been put to a stringent test as the languages studied have relatively sparse inflection systems. In contrast, the verb inflections of Hungarian make 24 different distinctions, with all but one of these involving two or more different allomorphs. Problems in the retention of sound sequences might well slow the development of inflections in this language. If problems of this type are playing a role, the children’s accuracy with inflections should be related to factors such as inflection length and nonword repetition ability.

A Sketch of Hungarian Tense and Agreement Morphology

In Hungarian, verb inflections mark tense and mode, agreement with the subject in person and number, and agreement with the object in definiteness. (Of these dimensions, distinctions according to mode are not examined in the present study; all inflections assessed are in the indicative.) Although Hungarian is often referred to as an agglutinating language, the dimensions of person and number are clearly fusional, and there is a complex relationship between agglutinating and fusional elements. We will return to this issue after introducing the verb inflections under investigation.

Table 1 provides the tense and agreement inflections with their allomorphs. Table 2 shows the tense and agreement inflections applied to the verb tol “push”.¹ Inflections appear in bold for ease of illustration. In these tables, we divide the inflections into four “paradigms”. However, this division is primarily for illustrative purposes, as the inflections for tense, person, number, and definiteness can be viewed as a single paradigm.

Table 1.

Inflections and their allomorphs for the four paradigms tested in the study

		Definite (e.g., Én tolom a biciklit “I am pushing the bycicle”)		Indefinite (e.g., Én tolok egy dobozt “I am pushing a bycicle”)
		Singular	Plural	Singular	Plural
Present	1st	-om/em/öm	-juk/jük	-ok/ek/ök	-unk/ünk
	2nd	-od/ed/öd	-játok/itek	-sz/ol/el/öl	-tok/tek/tök
	3rd	-ja/i	-ják/ik	0	-nak/nek
Past	1st	-tam/tem	-tuk/tük	-tam/tem	-tunk/tünk
	2nd	-tad/ted	-tátok/tétek	-tál/tél	-tatok/tetek
	3rd	-ta/te	-ták/ték	-t/ott/ett/ött	-tak/tek

Table 2.

Inflected forms for tol “push” in the four paradigms tested in the study

		Definite (e.g., Én tolom a dobozt “I am pushing the box”)		Indefinite (e.g., Én tolok egy dobozt “I am pushing a box”)
		Singular	Plural	Singular	Plural
Present	1st	Tolom	Toljuk	tolok	Tolunk
	2nd	Tolod	Toljátok	tolsz	Toltok
	3rd	Tolja	Tolják	tol	Tolnak
Past	1st	Toltam	Toltuk	toltam	Toltunk
	2nd	Toltad	Toltátok	toltál	Toltatok
	3rd	Tolta	Tolták	tolt	Toltak

Several details can be noted from an inspection of the tables. First, Hungarian’s use of agreement between the verb and the object (in definiteness) as well as between the subject and the verb (in person and number) effectively doubles the size of the paradigm. The number of inflections that must be learned by Hungarian-speaking children, then, is quite large indeed. Verb-object agreement is typologically much less common than subject-verb agreement. In fact, many languages show subject-verb agreement without verb-object agreement, but the reverse does not seem to occur. Note from the tables that any difficulty that is specific to verb-object agreement should be detectable. For example, in contexts requiring a present first person singular form, a child might produce tolok instead of tolom (or vice-versa).

The indefinite conjugation is regarded as unmarked. It is used with intransitive verbs as well as with transitive verbs with indefinite objects. It is also employed when the object is a first or second person pronoun.² The definite conjugation is chosen when the object Noun Phrase (NP) is clearly marked with a definite article (a or az “the”) and when the object is a possessively modified noun. Proper names as object NPs also take the definite conjugation. There are additional factors that are associated with the choice of a definite or indefinite inflection that go beyond the scope of the present study. For a more detailed description, see Bartos (1997) and MacWhinney and Pléh (1997).

A second notable detail that is evident in Tables 1 and 2 is the relatively large number of allomorphs. Most of the variation in the form of the inflection is a function of the vowel harmony rules of Hungarian. These rules seem to be acquired at a rather young age by Hungarian-speaking children (e.g., MacWhinney, 1985), even if they render the relationship between agreement inflections in present and past tense less clear. Other allomorphs are a product of phonological conditioning. Chief among these is the present indefinite second person singular allomorph, -sz, whose form is determined by the particular consonant appearing at the end of the verb stem.

Many languages with rich inflectional paradigms do not permit bare verb stems. Hungarian is an exception, in that the present indefinite third person singular inflection is a “zero” form, as in tol. The existence of a finite bare stem form in Hungarian means that, in principle, a child could employ such a form as a default whenever the appropriate inflected form is not known or is difficult to retrieve in the moment. Finally, it can be seen in Tables 1 and 2 that there is minimal syncretism (MacWhinney & Pléh, 1997); the only neutralization occurs in the past first person singular forms where the same inflection is used for both definite and indefinite objects (thus, toltam is used for both “I was pushing the box” and “I was pushing a box”)³. Hereafter, we employ the following abbreviations: “1”, “2”, and “3” for first, second, and third person, respectively; “Sg” for singular and “Pl” for plural; “Pres” for present and “Past” for past; and “Indef” for indefinite and “Def” for definite.

The subject-verb agreement (for person and number) reflected in Tables 1 and 2 corresponds to that seen in many other languages (apart from its fusion with definiteness marking). However, Hungarian subject-verb agreement operates somewhat differently because quantified nouns do not formally agree in number with their quantifiers. For example, ten bottles is expressed with a singular noun tíz üveg “ten bottle” rather than a plural noun *tíz üvegek “ten bottles”. The same is true for nouns preceded by terms corresponding to “many”, “some” and “all”. This characteristic has implications for subject-verb agreement because agreement is based on formal marking and not conceptual plurality. Thus a subject such as “ten bottle” would require a verb inflected for singular.

The relationship between agglutinating and fusional elements of the inflection system is very complex. When (past) tense is overtly marked, this element precedes elements reflecting person and number. Thus, in Table 2 it can be seen that in the indefinite past third person plural, past tense -t- precedes third person plural –unk; the present tense counterpart has no overt tense element preceding –unk. However, for inflections marked for definite, position is less transparent. For example, whereas definite past third person plural has the sequence –t-uk, definite present third person plural has the sequence –j-uk, with –j- representing an element marking definiteness, not tense. This complexity has led to proposals (e.g., Rebrus, 2005) that the same position can serve more than one grammatical function, depending on the particular tense, definiteness, and person and number features involved. Phonologically conditioned allomorphy in Hungarian can also reduce the transparency of the agglutinating elements of the inflections. For example, whereas tolom is the form for definite present first person singular “I am pushing”, the form toltam is used for definite past first person singular “I was pushing”, not *toltom, due to lowering of mid-vowels after past tense –t-.

Hungarian-Speaking Children: Previous Findings

Although no systematic experimental examination have been done thus far on the development of agreement marking by typically developing Hungarian-speaking children, two case studies (Lengyel, 1981, data from a boy between 1 and 3 years; Meggyes, 1971, data from a girl between 1;8 and 2;2) and a more extensive analysis of data from 3 Hungarian children between 1;8 and 2;9 from the CHILDES database (Babarczy, 2005) report errors in agreement or other inflection details. According to these studies, the very first verb forms are usually either imperative forms or third person singular declarative forms that are sometimes applied to non-third person referents. In early verb usage, Hungarian children generally use all three singular forms together with Pl1 to express Sg1 meanings. For example, in contexts requiring tolok “I am pushing [indefinite]” a child might produce tolok, tolsz, tol, or tolunk (see Table 2). Because these utterances usually lack a subject, there is no overt error of subject-verb agreement in such utterances. Based on these three studies, there seems to be individual variation in the extent children use Sg2 as a substitute for Sg1, but for some children such errors are more common in the beginning than Sg3 substitutions, which frequently occur with all children and for a longer period. Pl2 first only appears in imperative forms, and even when it does appear in declarative form, it is fairly uncommon. There are very few errors in marking Pl3 from the beginning, but these forms are also not frequent. Past tense forms also appear towards the end of the second year, and at first they are generally used to express completed actions.

Babarczy (2005, 2007), in her analysis of CHILDES data from 6 Hungarian children between 1;8 and 2;10, found many errors in definiteness agreement, revealing the children’s preference for using the default indefinite form with a definite object (she was focusing on imperative forms) and fewer errors in subject-verb agreement. Based on a comparative analysis of early verb forms, she found that subject-verb agreement is delayed in English relative to Hungarian. Interestingly, she also observes that there is no sentence length effect on the agreement errors that young Hungarian-speaking children make. Lengyel points out that while mixing up first and third person is common in the indefinite conjugation, it is very rare in the definite conjugation. In summary, typically developing children first mainly use singular forms, most often to refer to first person, and they make many errors of using Sg3, Sg2 and Pl1 forms for Sg1 meanings. Indefinite verb forms are sometimes used in place of definite forms.

Systematic studies of Hungarian-speaking children with LI have also been few in number. Vinkler and Pléh (1995) reported on a child with LI who had difficulty with noun as well as verb morphology. This child often resorted to a more frequently occurring inflection as a substitute for the required form. Marton, Schwartz, Farkas, and Katsnelson (2006) compared the working memory performance of Hungarian-speaking and English-speaking children with specific language impairment. They found that, for the Hungarian-speaking children, morphological complexity played a larger role than sentence length, whereas syntactic complexity was the most influential factor for the English-speaking children.

Hypotheses

Given the details of tense and agreement inflections in Hungarian, several hypotheses can be advanced. First, according to the agreement deficit account, children with LI should be significantly less accurate than their typically developing peers in the agreement details of the inflections. Errors are likely to be default forms such as third person singular forms. Tense should be correctly marked. According to the morphological richness account, the rich inflectional morphology and null-subject character of Hungarian will lead children with LI to make much more use of tense and agreement inflections than is the case for children with LI in English. However, the four dimensions of tense, definiteness, person, and number that are required in Hungarian inflections (rather than the more commonly encountered three dimensions seen in other languages studied) will place demands on these children’s limited processing capacity, leading to small but statistically reliable differences between children with LI and typically developing children. When errors are observed, many should constitute near-misses; use of a default form is not expected. If non-morphosyntactic language processing factors such as poor retention of sound sequences are involved, errors not clearly attributable to the number of dimensions involved in the inflections should be found, and the children’s use of inflections should prove to be related to factors such as the length of the inflection and the children’s ability in nonword repetition.

Method

Participants

Fifty children participated in the study. Twenty-five children were selected for the LI group from two special schools for children with language impairments. All of these children met the criteria for LI. Each child scored above 85 on the Raven Coloured Progressive Matrices (Raven, Court, & Raven, 1987), a measure of nonverbal intelligence. All children passed a hearing screening, and no child had a history of neurological impairment. Each child scored at least 1.5 SDs below age norms on at least two of four language tests administered. These four tests included two receptive tests and two expressive tests. The receptive tests were the Hungarian standardizations of the Peabody Picture Vocabulary Test (PPVT) and the Test for Reception of Grammar (TROG). The expressive tests were the Hungarian Sentence Repetition Test, and a nonword repetition test. The rationale for including a nonword repetition test (described below) in the assessment battery is that the ability to repeat nonwords has proven to be one of the most accurate means of identifying children with LI (e.g., Dollaghan & Campbell, 1998; Tager-Flusberg & Cooper, 1999), demonstrating excellent sensitivity and specificity, and seems to be one of the fundamental and heritable weaknesses seen in this type of disorder (Bishop et al., 2006).

Although the PPVT (Dunn & Dunn, 1981; Csányi, 1974) was used as one of the language tests in our selection battery, it was also used as the basis for matching participant groups, as will be seen below. The Hungarian adaptation of the original TROG (Bishop, 1983) is being standardized on children from 4 to 12 years of age⁴. Items assess the children’s comprehension of increasingly more difficult grammatical structures. The test consists of 20 blocks, each with 4 sentences of the same construction (such as sentences with comparatives, postmodified subjects and embedded clauses). The test has a booklet containing 80 pages, each with 4 pictures, and on each page the child must point to the picture that matches the sentence spoken by the experimenter. A block is considered completed if the child responds correctly to all 4 pictures in the block. Performance is measured in terms of number of blocks correctly completed.

The Hungarian Sentence Repetition Test (Magyar Mondatutánmondási Teszt, MAMUT, Kas & Lukács, in preparation) manipulates length and structural complexity independently. Its 40 sentences are distributed evenly across 5 types of grammatical constructions: (1) simple Subject-Verb-Object (SVO) and (2) simple OVS sentences; complex sentences containing (3) SO subject relative clauses; (4) SO object relative clauses; and (5) OS object relative clauses. Sentence length varies between 8 and 15 syllables within each type. The task of the participant is to immediately and accurately repeat the sentences presented by the experimenter. Performance is measured in terms of the number of correctly repeated sentences, which can be evaluated based on grouping by syllable number and by grammatical construction as well.

The nonword repetition test (Racsmány et al., 2005) requires the repetition of meaningless but phonotactically licit strings of Hungarian phonemes. The test contains 36 nonwords between 1 and 9 syllables in length. Each length is represented by 4 nonwords. The phonological structure of the nonwords does not reflect frequency distributions of Hungarian phoneme sequences, but the test avoids sequences that would be articulatorily difficult for speakers. The span of the participant is the highest syllable number for which s/he could correctly repeat at least 2 out of the 4 nonwords.

The remaining 25 children were typically developing. These children scored above -1SD on each of the four language tests that were administered to the children with LI. These children were matched with the LI group on the basis of their raw scores on the PPVT. Because the children with LI scored below age level on the PPVT, the typically developing children matched on this measure were younger. A typically developing child was considered a match if his or her PPVT score was within 3 points of the PPVT score of a child in the LI group. Hereafter, this group will be referred to as the vocabulary control (VC) group. The use of younger typically developing children matched on a non-grammatical language measure was designed to detect whether the difficulties of the children with LI on tense and agreement morphology exceeded their more general limitations in language. If so, group differences favoring the VC group should be seen. Of course, differences in the two groups’ pattern of use across the different tense and agreement morphemes was also of interest. Means for age (in years; months) and raw scores on each of the tests together with ranges for both groups are given in Table 3.

Table 3.

Means and ranges (in parentheses) for the LI and VC groups for age in years; months and in raw scores on the PPVT, the TROG, the Nonword Repetition, and the Sentence Repetition tests.

	LI	VC
Age	9;10 (7;6–11;10)	7;1 (5;2–8;5)
PPVT	91.3 (61–114)	92.1 (62–115)
TROG (blocks correct)	12.3 (8–18)	13.76 (6–20)
Nonword repetition	3.5 (1–5)	5.8 (3–8)
Sentence repetition	22.0 (0– 39)	33.6 (18–40)

Method

Given the large number of tense and agreement inflections in Hungarian, we devised a structured method of eliciting responses that ensured multiple opportunities for the child to produce each inflection of interest. The children were asked to repeat sentences; however, the target inflections in each sentence were actually masked by a carefully inserted cough that prevented the child from hearing the inflection but not the stem or the remaining portions of the sentence. This method was adapted from Warren’s (1970) phoneme restoration procedure. The restoration effect has been demonstrated at the morpheme level as well, such as for affixes in Hungarian (Dankovics & Pléh, 2001), but the effect has not yet been exploited in developmental studies as an elicited production method. Importantly, in our study the fully audible portions of the sentence (notably, the temporal adverbial, the person and number of the subject, and the definiteness of the object) made it clear (to a mature speaker of Hungarian) which verb inflection was the appropriate one to use. The child was only asked to repeat the sentences and was not told that information was missing.

Specifically, children were instructed to repeat sentences they heard through a loudspeaker. The sentences were recorded by a female speaker, and digitized, with coughs inserted to replace the inflections only (see below). All sentences were normalized for a length between 8 and 14 syllables. Although the target inflections in the middle of the sentence were replaced by a cough, the remainder of the sentence contained all the source features for unequivocal identification of the missing inflection. Children occasionally commented that the speaker was coughing a lot; in these cases we told them that she had a cold, and that they should just disregard the coughs.

Six verbs were used, in both present and past tense, in both the definite and indefinite conjugations, in both singular and plural, and in first, second, and third person. Thus, 144 sentences (6 × 2 × 2 × 2 × 3 = 144) were created. The sentences were blocked according to tense and definiteness paradigm. That is, all 36 sentences marked for present definite were presented together, as were the 36 sentences marked for present indefinite, past definite, and past indefinite. Children were tested in at least two different sessions, with the order of the four blocks counterbalanced across children.

Given the vowel harmony involved in the allomorph used for the inflection, we selected three verbs whose stems had front vowels and three that had stems with back vowels. The six verb stems selected for the task were: tol “push”, olvas “read”, simogat “stroke” (= pet), kerget “chase”, épít “build”, and fésül “comb”.

All sentences were simple SVO sentences. Past tense sentences were systematically longer than present tense sentences because they contained the temporal adverbial tegnap “yesterday”, used to make the past time of the described event clear. (Hungarian does not possess a temporal adverbial that is unique to present tense.) The examples in (1) illustrate the types of sentences used for each tense and definiteness combination. The location of the inflection masked by a cough is indicated by “XXX”.

• 1(a)Mi olvasXXX egy mesét

  Target: olvasunk [“read” 1PlPresIndef]

  “We are reading a story”

• (b)A gyerekek simogatXXX a malacot

  Target: simogatják [“stroke” 3PlPresDef]

  “The children are petting the pig”

• (c)Tegnap én építXXX egy tornyot

  Target: építettem [“build” 1SgPastIndef]

  “Yesterday I built a tower”

• (d)Tegnap te tolXXX a biciklit

  Target: toltad [“push” 2SgPastDef]

  “Yesterday you pushed the bike”

It was important to ensure that the inserted coughs were sufficient to obscure the inflection, and that there were no anticipatory coarticulatory cues in the verb stem that might have provided the children with an indication of the inflection that was masked. Accordingly, we extracted the verb stem plus cough from each recorded sentence and presented them to 15 adult listeners. The listeners were asked to guess which inflection was used with the stem in each case (for all 144 verb forms). For every item, they had to select from 24 possible forms, and they guessed correctly on 5.6% of the items, which, as will be seen, is significantly below the performance level for either group of children, (LI = 62%, χ2-test, p<0.001, VC=83%, χ2-test, p<0.001). These findings indicated that our stimuli probably did not contain unintended cues that could lead to correct performance without knowing the appropriate inflection. In fact, the adult listeners’ guessing behavior suggested that other factors were influencing their choices. The log frequency of allomorphs in Hungarian based on the Hungarian Webcorpus (Halácsy et al., 2004; Kornai et al., 2006) was a significant predictor of the frequency of the listeners’ specific choices (R² = 0.132, beta = 0.363, p<0.001). Not surprisingly, the items whose inflections happened to correspond to the listeners’ most frequent choices were most likely to be guessed correctly. However, even the inflection type that was most frequently guessed correctly was associated with only 14% accuracy.

Scoring

Our scoring method emphasized accuracy of tense and agreement marking rather than accuracy of the sentence as a whole. That is, we allowed for differences between the child’s response and the stimulus sentence provided that the child’s response showed internally accurate agreement as well as tense marking. This scoring method was selected to reduce the effects of recall errors and to provide as clear a view of inflection use as possible to evaluate the agreement deficit and morphological richness accounts – two accounts expressly developed to explain the tense and agreement inflection problems of children with LI.

According to this scoring method, if the children used a non-target verb with correct inflection or if the child used a different subject or object but the verb inflections were appropriate for this change, the response was scored as correct. In addition, if a child produced a past tense form when the stimulus sentence was in present tense (without any other change), the child was credited with a correct response. Although in such cases it is more customary to assume such sentences are in present tense, recall that there is no adverbial that is unique to present tense. (To use the closest English equivalent, whereas we must use past tense with “yesterday”, either past or present tense might be appropriate with “today”). As Hungarian has somewhat flexible word order, variations in word order were also permitted, provided that all of the above details were included. Using this method, the following errors could occur: (1) person error; (2) number error; (3) tense error; (4) definiteness error; or (5) other error, such as a sentence that bore no resemblance to the stimulus sentence. If errors (1)-(5) or any of their combinations occurred, the answer was scored 0. The children’s use of the wrong allomorph in otherwise-correct responses was also noted, but not scored as an error.⁵ Examples of error types and deviations from the stimulus sentence that were counted as correct are shown in Table 4.

Table 4.

Examples of different types of errors or deviations from the target sentence for the stimulus sentence Tegnap ti fésültétek az oroszlánt “Yesterday you (Pl) were combing (comb PastDefPl2) the lion”

Response type	Child’s response	Translation
Person Error	Tegnap ti fésültük az oroszlánt	Yesterday you (Pl) were combing (PastDefPl1) the lion
Number Error	Tegnap ti fésülted az oroszlánt	Yesterday you (Pl) were combing (PastDefSg2) the lion
Tense Error	Tegnap ti fésülitek az oroszlánt.	Yesterday you (Pl) are combing (PresDefPl2) the lion
Definiteness Error	Tegnap ti fésültetek az oroszlánt.	Yesterday you (Pl) were combing (PastIndefPl2) a lion.
Nontarget verb with correct agreement	Tegnap ti fésülködtétek az oroszlánt.	Yesterday you (Pl) were combing (reflexive, PastDefPl2) the lion.
Nontarget subject or object with correct agreement	Tegnap ti fésültetek egy oroszlánt.	Yesterday you (Pl) were combing (PastIndefPl2) a lion.

To assess interjudge reliability, the responses of five children in each group were selected at random and scored by an independent judge. Percentage agreement ranged from 97.2 to 100, with similar percentages of agreement for the LI (M = 98.75) and VC (M = 99.6) groups.

Data Analysis

The data were examined in several ways. First, we examined the children’s percentages of correct responses for each inflection type, using a general linear model analysis of variance (ANOVA) with Group as a between-subjects factor and Tense, Definiteness, Number, and Person as within-subjects factors. Second, given the predictions of the morphological richness account, we determined whether the children’s scores were related to frequency of occurrence factors. For each inflected verb form, we calculated the following: (1) inflected word frequency (the frequency of the exact inflected verb form); (2) inflection frequency (e.g., the frequency of all PresDefSg3 allomorphs combined); and (3) allomorph frequency (mostly conditioned by stem category for vowel harmony, e.g., the frequency of the –ja allomorph of PresDefSg3). The source of frequency data was the Hungarian Webcorpus (Halácsy et al., 2004; Kornai et al., 2006). Calculations employed the logarithm of frequency. Finally, we performed an analysis of the children’s errors.

Results

Accuracy According to Group and Inflection Type

The ANOVA on accuracy revealed Group as a significant main effect, F (1,48) = 10.02, η²=0.173, p < 0.01. With the exception of Definiteness, F (1,48) = 0.09, n.s., all within-subjects factors proved to be significant main effects: Tense, F (1,48) = 13.91, η²=0.225, p < 0.01, Number, F (1,48) = 8.91, η²=0.157, p < 0.01, and Person, F (1,48) = 27.19, η²=0.362, p < 0.001. The significant interactions were Tense × Definiteness × Person, F (2,96) = 7.22, η²=0.131, p < 0.01, Number × Person, F (2,96) = 10.05, η²=0.180, p < 0.001, Definiteness × Number × Person, F (2,96) = 8.85, η²=0.156, p < 0.001, and Tense × Definiteness × Number × Person, F (2,96) = 4.81, η²=0.156, p < 0.05. Pairwise comparisons (LSD tests) at the 0.05 level revealed that Past, Plural and 2nd Person were significantly more difficult than Present, Singular and 1st and 3rd Person, respectively (1st and 3rd Person did not differ). Figure 1 provides an illustration of the findings.

Figure 1.

Mean percentage correct for each inflection type for the LI and VC groups. Standard errors are also shown.

It can be seen that overall performance of the LI group was significantly lower than that of the VC group, but no interactions with Group were significant, suggesting that the two groups basically showed the same pattern of performance across the dimensions examined. The interactions involving Person and Number were due to low scores of 2nd person and, especially, of Pl2 forms. These difficulties are evident from Figure 1.

Relationship with Frequency

We examined the relationship between several frequency factors and the children’s use of the tense and agreement inflections. According to the morphological richness account, children should have greater success producing more frequently occurring inflections than less frequently occurring inflections. However, it is also true that other details, such as the frequency of the words themselves could also influence the children’s success. To determine if these factors could predict performance on the experimental task, we included them in stepwise regression analyses. We tested the effects of log inflected word frequency, log inflection frequency, and log allomorph frequency on the number of correct responses, separately for the LI and VC groups. Only variables that showed a significant correlation (p < 0.05) with the target variable were entered into the analysis.

For both groups, the factor that best contributed to predicting performance levels was log inflection frequency. As can be seen in Table 5, the LI data are somewhat better predicted by this factor, where it explains 31% of variance, as opposed to 20% explained in the VC group.

Table 5.

Log inflection frequency as a predictor of the performance of the LI and VC groups. R² shows the amount of variance in the data explained by the predictor.

		Beta	Sig	R2
VC	Log inflection frequency	0.45	<0.001	0.20
LI	Log inflection frequency	0.56	<0.001	0.31

Error analysis

Both groups of children produced many errors on the task. Out of the 3600 responses from each group, the VC group produced 371 errors (10.03%), and the LI group erred on 905 (25.13%) responses. It is notable that the number of inappropriate productions of the present third person singular indefinite – the zero-marked form – was not especially high, suggesting that this form was not used as a default. This zero-marked form constituted only 6.8% of the errors in the VC group, and 5.2% of the errors in the LI group. Inappropriate productions of these zero-marked forms were outnumbered by the inappropriate production of inflected forms. For example, the incorrect production of present third person singular definite forms represented 8.2% of the errors for each group, and inappropriate productions of present first person plural definite forms constituted 9.4% of the errors for the VC group and 13% of the errors for the LI group.

Figure 2 provides the mean number of errors, according to error type. Numbers for each error type represent errors that constituted an error only on that single dimension. Along with the responses treated as errors in the preceding analyses, we include in Figure 2 non-target responses that were scored as correct in those analyses, namely, the use of a nontarget verb with correct tense and agreement (NTV), the use of a nontarget subject or object with correct agreement (NTS/O), and the use of an incorrect allomorph (Allmor) even though agreement was correct. Figure 2 illustrates several group differences, but not all of them are confirmed by statistical analysis. The LI group made more single-dimension errors overall, F (1, 49) = 9.2, η² = 0.21, p < 0.01. ANOVAs were also performed for each error type separately. The difference reached significance for Person, F (1, 49) = 8.8, η² = 0.155, p < 0.01, and Definiteness, F (1, 49) = 4.16, η² = 0.08, p < 0.05, but not for Number, F (1, 49) = 1.6, n.s., or Tense, F (1, 49) = 2.68, n.s. More detailed comparison of dimension errors across groups shows that, among person errors, the LI group only made significantly more errors than VC children in using 3^rd person forms, F (1, 49) = 8.75, η² = 0.154, p < 0.01. In definiteness errors, the difference was only significant with using indefinite forms when the target was definite, F (1, 49) = 7.98, η² = 0.143, p < 0.01. The remaining response type treated as an error in the earlier analyses, Other, also revealed a difference between the two groups of children, F (1, 49) = 4.93, η² = 0.093, p< 0.05. None of the deviations from the target originally scored as correct showed a group difference, such as NTV, F (1, 49) = 1.97, n.s., and NTS/O, F (1, 49) = 2.34, n.s. Finally, although use of the wrong allomorph (Allmor) was not considered an error, it can be seen from Figure 2 that the two groups were highly similar in this regard, suggesting that rules of vowel harmony were well established and did not seem to be an area of particular difficulty for the LI group. An inspection of Figure 2 reveals that, although the children with LI made a greater number of errors than the VC children, the pattern of errors across error types was highly similar in the two groups.

Figure 2.

Mean number of errors on different error types in the two groups. Only errors in a single dimension are counted. Standard errors are also shown.

The morphological richness account predicts that single-dimension or “near-miss” errors will be especially frequent. To test this prediction, we compared the children’s near-miss errors to productions that constituted an error on more than one dimension (e.g., an error of tense plus number). A repeated measures ANOVA showed a significant effect of Group, F (1, 48) = 10.07, η² = 0.17, p < 0.01, but no difference according to near-miss versus multiple-dimension error, F (1, 48) = 1.02, n.s., and no interaction F (1, 48) = 0.001, n.s. For the VC group, the mean number of near-miss and multiple-diminsion errors was 10.76 (SD = 8.20) and 8.00 (SD = 12.35), respectively. For the LI group, the corresponding means were 22.28 (SD = 13.80) and 19.72 (SD = 24.63), respectively. The prediction of the morphological richness account of a predominance of near-miss errors was not borne out.

Non-Morphosyntactic Language Processing Factors

The agreement deficit account and the morphological richness account predict difficulties according to the nature of the dimension involved (e.g., agreement) or the number of dimensions involved (e.g., four) in the inflections. However, if the children’s use of inflections is also influenced by factors pertaining to the retention of sound sequences, factors other than the specific nature or number of dimensions involved should be observable. One such factor is the length of the verb plus inflection, measured in number of phonemes. Accordingly, we determined whether length in number of phonemes could serve as a significant predictor of the children’s accuracy of inflection use. This proved true for each group. For the VC group, this factor accounted for 20% of the variance in the children’s inflection accuracy scores (beta = 0.45, p < .001, R² = 0.20); for the LI group, 31% of the variance was explained by this factor (beta = 0.55, p < .001, R² = 0.31).

Recall, however, that log inflection frequency also proved to be a predictor of the children’s accuracy of inflection use. Some of inflections that were relatively low in frequency such as the second person plural inflections (e.g., játok, tatok) are also among the longest inflections. Therefore, we performed a regression analysis to determine if length in phonemes contributed to the prediction of the children’s inflection accuracy even when log inflection frequency is taken into account. The results appear in Table 6. As can be seen, for each group, length in number of phonemes proved significantly related to the children’s inflection accuracy along with log inflection frequency; together these factors explained 27% of the variance in the VC data and 41% of the variance in the LI data.

Table 6.

Length in number of phonemes and log inflection frequency as predictors of the performance of the LI and VC groups. R² shows the amount of variance in the data explained by the predictor.

		Beta	Sig	R2
VC model	Number of phonemes	−0.32	<0.001	0.27
	Log inflection frequency	0.28	<0.01
LI model	Log inflection frequency	0.38	<0.001	0.41
	Number of phonemes	−0.36	<0.001

The data in Table 6 address the degree to which length of the verbs with inflections related to the children’s inflection accuracy, but this factor cannot be divorced from the dimensions (e.g., person, number) reflected in the inflections. To gain an impression of the role of length independent of tense and agreement, we used the children’s scores on the nonword repetition test as a covariate and again compared the VC and LI groups. Although low (LI group) or age-appropriate (VC group) nonword repetition test scores constituted one of the bases on which the children were selected, the typically developing comparison group (mean age = 7;1) was on average more than two years younger than the LI group (mean age = 9;10). Nevertheless, the two groups differed on this measure (LI M = 3.5, SD = 1.5; VC M = 5.8, SD = 1.3, t (48) = 6.14, p < 0.001. When nonword repetition was entered as a covariate, the group difference in inflection accuracy disappeared, F (1,47) = 0.68, n.s.. The effect of nonword repetition was significant, F (1,47) = 4.75, η²=0.096, p < 0.05. These findings suggest that factors such as ability to retain sequences of sounds may have had a bearing on the children’s use of inflections on our experimental task.

Discussion

In this study, we found that a group of Hungarian-speaking children with LI performed significantly below the level of younger VC children in a task in which the children had to repeat sentences and supply the appropriate tense and agreement inflections. Although the two groups differed in accuracy, their patterns of performance across inflection types – both in terms of inflections with greatest and least accuracy and in terms of error types – were highly similar. Before discussing the implications of these findings, we discuss some potential limitations of the study.

One potential limitation is that we cannot be certain that our task yielded results that were representative of the children’s actual abilities. Studies of children with LI in other languages have typically employed spontaneous speech samples and/or sentence completion tasks. We believe our choice of tasks was highly appropriate given the characteristics of Hungarian. For example, the distinction between agreement inflections as a function of the definiteness of the object is not one that can be easily manipulated through sentence completion tasks. Despite the novel nature of our task, the higher scores by the younger group of TD children compared to the children with LI suggest that it was developmentally appropriate.

Another potential limitation is our use of younger TD children matched with the LI group according to receptive vocabulary rather than according to an expressive measure such as MLU. However, for a language with a rich morphology such as Hungarian, MLU matching would carry the risk of matching two groups on the very ability that we were wishing to compare. Nevertheless, matching on the basis of receptive vocabulary was a more stringent test of the status of tense and agreement morphology in Hungarian LI than would be the case if chronological age controls had been used. As can be seen in Table 3, the children with LI were nearly three years older than the VC children, yet they did not perform as well as these younger typically developing children.

Another potential criticism of the study is that given our use of a nonword repetition test and a sentence repetition test as two of the four tests in our diagnostic battery, it might be argued that we selected only or primarily those children with LI with limitations in working memory. However, all of the children with LI earned low scores on the PPVT – a receptive vocabulary measure that seems to place fewer working memory demands on the children than all of our other measures. In addition, the children’s enrollment in special schools for children with language impairments required a diagnosis made by professionals prior to the children’s participation in this study. Thus, although these children may have had limitations in working memory, they were not clearly different from the more general population of children with LI in having working memory limitations along with problems with language itself.

Hungarian is a language with agreement required between both the subject and the verb and between the verb and the object. According to the agreement deficit account, children with LI should have more difficulty than their TD peers in the marking of agreement. To evaluate the predictions of this account, it is important to examine the children’s accuracy with regard to tense separately from their accuracy with regard to agreement. As can be seen in Figure 2, the children with LI made a greater number of tense errors than the VC children, but this difference did not achieve statistical significance. As would be predicted by this account, agreement errors were clearly evident in the responses of the LI group. Yet, the group difference for number errors was not significant. These errors were relatively infrequent by the LI group. Furthermore, considering that 24 different inflections were required in our task, all involving agreement, the LI group’s mean percentage of correct use of 60% suggests that these children were clearly not producing inflections at random. Furthermore, these children were not relying on a default form when responding to the items. These findings suggest that if the agreement deficit account is generally correct, provisions must be made in the account to explain how children with LI can use all person, number, and definiteness forms with some degree of accuracy, and not differ from VC children in the use of number. In addition, the agreement deficit account provides no reason for the special difficulty with Pl2 forms experienced by the children with LI.

Hungarian differs from languages with a rich inflectional morphology such as Italian and Spanish in that distinctions in four dimensions – tense, person, number, and definiteness – are required rather than the distinctions in three dimensions required in these other languages. According to the morphological richness account, rich inflectional morphology is beneficial to children with LI up to a point; however, four dimensions have been proposed as the number of dimensions that begin to tax these children’s limited capacities. For this reason, Hungarian-speaking children with LI are expected to perform below the level of typically developing peers even though their levels of inflection use should be considerably higher than the levels reported for children acquiring English.

The findings were in keeping with this prediction. Furthermore, this account predicts that the inflections with the greatest likelihood of accuracy in the speech of children with LI will be those of higher frequency of occurrence. Our results were also consistent with this expectation. However, the morphological richness account also predicts that the great majority of errors will take the form of near misses. Although such errors were common, they did not exceed the number of errors on multiple dimensions. This finding runs counter to the assumptions of this account.

It seems quite possible that the morphological richness account is too simplistic in its assumption that the number of dimensions reflected in an inflection is the chief contributor to processing demands in a language with a rich inflection system such as Hungarian. For example, in an application of the Competition Model to Hungarian, MacWhinney and Pléh (1997) noted that adults’ interpretations of sentences relied less on definiteness agreement between the verb and the object than on other cues. These investigators suggested that definiteness agreement in Hungarian has relatively low “contrast availability.” That is, because in Hungarian both the subject and the object may be definite, or both may be indefinite, definiteness is often non-contrastive and as a result adults seem to depend less on this type of cue than on other types of cues. It is possible that factors such as contrast availability influence production as well, and perhaps especially so in the case of children with LI. As a case in point, we noted that children with LI produced a greater number of definiteness errors as well as number errors than the VC group but did not differ from the VC group in committing errors involving number. Both definiteness and number require agreement, both have contrasts of two features (definite versus indefinite, singular versus plural), and both were crossed with tense and person distinctions in the same way in the sentence stimuli. Therefore the fact that the LI and VC groups differed in the number of errors on one of these dimensions and not the other suggests that factors beyond the number of dimensions are probably relevant. Future research, such as an application of the Competition Model to the study of inflection use in children with LI might prove quite informative in this regard.

Along with their well documented problems in the area of morphosyntax, children with LI often have considerable difficulty retaining sequences of sounds, as measured by tasks such as nonword repetition (see Graf Estes, Evans, & Else-Quest, 2007 for a recent meta-analysis). Although these two deficits are separable (Bishop et al., 2006), many children with LI have both of these deficits. An assumption of the present study is that in a language with a multitude of inflections and allomorphic variations such as Hungarian, children’s ability to retain sequences of sounds may have a greater influence on their ability to learn the inflection system than is seen in a language such as English.

Our findings seem consistent with this assumption. The length of the verb with inflection proved related to the children’s inflection accuracy even when log inflection frequency was taken into account. More importantly, the very clear differences between the two groups in inflection accuracy were no longer evident when the children’s nonword repetition scores were used as a covariate.

Collectively, our findings lend support to the notion that processing-related factors play a role in the inflection limitations of children with LI in a language such as Hungarian. However, it is likely that we have not identified all of the factors related to processing that were at play in this study. Earlier we noted that factors considered in the Competition Model such as contrast availability may prove important. In addition, other types of processing factors might be identified. For example, the children sometimes changed the verb or a subject or object in the stimulus sentence. It is true that even when such changes were allowed (provided that the verb inflection was correct) group differences favoring the VC children were seen in inflection accuracy. Nevertheless, it seems important to determine why such substitutions of verbs, subjects, and objects were relatively frequent in the data.

In summary, the findings of this investigation indicate that models assuming processing limitations on the part of children with LI are more compatible with the pattern of verb inflection use seen in Hungarian-speaking children with LI than are accounts based on an assumption of deficits specific to agreement. One processing-related approach, the morphological richness account, seems to predict a substantial portion of the findings, but falls short in that it predicts a greater proportion of near-miss errors than we actually observed. Non-morphosyntactic language processing factors such as the retention of sequences of sounds may well account for additional details in the findings. We suspect that this factor may play a larger than usual role in a language laden with inflections such as Hungarian. Yet, it seems likely that other factors will prove important as well. Additional research is clearly warranted.