Processing modifier–head agreement in L1 and L2 Finnish: An eye-tracking study

Abstract

This study investigated the effect of first language (L1) on the reading of modifier–head case agreement in second language (L2) Finnish by native Russian and Chinese speakers. Russian is similar to Finnish in that both languages use case endings to mark grammatical roles, whereas such markings are absent in Chinese. The critical nouns were embedded in sentences, where the head noun was either preceded by an agreeing modifier or the modifier was absent. Readers’ eye fixation patterns were used as indices of online processing. Both natives and non-natives showed a facilitatory effect of agreement; reading head nouns was easier when they were preceded by an agreeing modifier. Typological distance in terms of the structural complexity of words between L1 and L2 did not influence the processing.

Keywords

eye movements modifier–head agreement reading second language acquisition transfer

I Introduction

In their review of the differences between first language (L1) and second language (L2) processing, Clahsen and Felser (2006c) single out the following question as important for future research: ‘How does the typological distance between the L1 and L2 influence L2 processing?’ (p. 569). This question is important because it is linked to how speakers of structurally different native languages (henceforth L1) can learn and process a non-native language (henceforth L2). It may or may not be that language learners have challenges related to structural differences between L1 and L2.

Although typological distance is always present in studies that deal with L2 learning and processing, the new aspect of the present study is to experimentally test adult language processing of modifier–head case agreement – e.g. suure en talo on = ‘the big (illative) house (illative)’ – in a situation where L1 and L2 differ maximally in terms of the structural complexity of the words. The structural complexity refers to the number of morphemes per word. The present study addresses the question of how intermediate-level language learners, who are L1 speakers of an analytic language (Chinese) versus a synthetic language (Russian), process a synthetic, structurally complex L2 (Finnish). The distinction between analytic and synthetic languages is denoted by the fact that in synthetic languages grammatical meanings are marked with affixes resulting in the words being morphologically complex, whereas in analytic languages’ grammatical meanings are encoded with particles or are left to be inferred from the context. The study was carried out using eye-tracking as the experimental technique combined with a non-violation agreement paradigm (for more details, see below).

II Processing modifier–head agreement in L1 and L2

The effects of modifier–head agreement in L1 processing have been studied, for instance, in Finnish (Hyönä and Hujanen, 1997; Hyönä and Lindeman, 1994; Vainio, Bertram, Pajunen et al., 2011; Vainio, Hyönä and Pajunen, 2003, 2008), Russian (Akhutina et al., 1999; Taraban and Kempe, 1999), Serbo-Croatian (Gurjanov et al., 1985) and Spanish (Barber and Carreiras, 2005). These studies have indicated that the processing of the head noun is faster when it is preceded by an agreeing modifier rather than a modifier–head combination containing an agreement violation (Gurjanov et al., 1985; Hyönä and Lindeman, 1994). Analogous effects were found for Spanish head–modifier combinations (modifier typically follows its head in Spanish; Barber and Carreiras, 2005). Moreover, the results of Vainio et al. (2003, 2008, 2011) suggest that L1 speakers make use of case agreement between the modifier and the head, as the presence of an agreeing modifier facilitates the processing compared to a situation where an agreeing modifier is absent. Thus, generally speaking, the prior L1 studies indicate that modifier–head agreement facilitates the processing of the head of the noun phrase. This is due to the fact that native speakers use morphological information provided by the modifier in assigning a grammatical status to the head noun.

Apparently, there are no previous L2 studies focusing on syntactically based modifier–head case agreement, which is the focus of the present study and typical in the Finnish language. The function of this kind of agreement is to bind the phrasal constituents together by case repetition. It should be noted that syntactic agreement is triggered positionally, whereas semantically or lexically motivated agreement is triggered by the lexical–semantic properties of the head noun (e.g. gender agreement in Russian). Examples (1) and (2) illustrate this point. In Finnish, the lexical–semantic properties of the verb define the object case:

(1) lyödä miestä

hit man-partitive

‘hit a man’

(2) rakentaa talo

build house-nom

‘build a house’

When a modifier is added to the noun phrase, the same case inflection is added both in the modifier and the head:

lyödä isoa miestä hit big-partitive man-partitive

rakentaa iso talo build big-nominative house-nominative

Despite the absence of L2 studies on syntactically based agreement, there are L2 studies that deal with gender and number agreement motivated either semantically or lexically. The studies on gender agreement in modifier–head phrases (e.g. Gillon-Dowens et al., 2009, 2011; Keating, 2009; Sagarra and Herschensohn, 2011) have shown that non-native speakers are able to detect violations in grammatical gender, although their processing is not necessarily similar to that of natives. However, L2 speakers of Spanish were not able to make predictive use of gender-marked articles when processing article–noun pairs in a looking-while-listening procedure; i.e. they were not faster at looking at the picture that had the same gender as the article in the spoken input (Lew-Williams and Fernald, 2010).

Most of the aforementioned studies have used relatively advanced L2 speakers as participants. The findings of Kempe and MacWhinney (1998) are most important for the present study. These findings have shown that in the acquisition of case marking in L2 German and L2 Russian, adult native English speakers learn to use case marking earlier in Russian than in German, although the Russian system is more complex. They suggest that the reason for this is that case marking in Russian has higher cue validity than case marking in German. Cue validity has two components: availability and reliability. As Finnish has a complex but highly reliable case marking system (Laine et al., 1999), we considered it justified and interesting to study the processing of modifier–head agreement among intermediate-level L2 speakers of Finnish.

It should be noted that case endings have two functional roles in Finnish. On the one hand, they are used to mark grammatical roles (subject, object, adverb, etc.) and, on the other hand, they indicate that consecutive words belong to the same phrase, e.g. lyödä isoa (partitive) miestä (partitive = ‘hit big (partitive) man (partitive)’.

III The non-violation paradigm for studying agreement effects

Presumably, it can be agreed that violating the grammar of a test language is a strong experimental manipulation. Therefore, it is easy to understand that different forms of the syntactic violation paradigm have been used productively in psycholinguistic and language learning studies. The main difference between the violation and the non-violation paradigm employed in the present study is elegantly expressed by Trenkic et al. (2014): ‘Being able to detect violations in ungrammatical sentences, however, is not the same as being able to facilitatively utilize grammatical information in the processing of well-formed sentences’ (p. 3). In other words, the non-violation paradigm allows its user to examine how linguistic structures, such as modifier–head agreement being either absent or present, are utilized during online language processing in the absence of grammatical violations (for those interested in electrophysiological correlates of phrasal morphosyntactic integration in German, see Davidson et al., 2012).

Taraban and Kempe (1999) were the first to inspect the processing of native and non-native modifier–head agreement without including any agreement violation. In their study, participants read sentences one at the time on a computer screen presented by a non-cumulative moving window paradigm; i.e. the prior text segment disappeared when a new text segment appeared. The head noun was either preceded by a modifier or the modifier was absent. Moreover, the head noun was either transparent or opaque in its gender; therefore, the modifier disambiguated the gender of an opaque head noun. Taraban and Kempe used masculine and feminine nouns, an examples of which are presented in (3)–(5); these examples are of a feminine transparent and an opaque noun with a filler masculine condition.

(3) Dashe (obychnaja) muka teper’ izhezla iz magazinov.

Even (ordinary, feminine) flour (transparent feminine) now vanished (feminine) from the stores.

(4) Dashe (obychnaja) sol’ teper’ izhezla iz magazinov.

Even (ordinary, feminine) salt (opaque feminine) now vanished (feminine) from the stores.

(5) Dashe (obychnij) ris teper’ izhez iz magazinov.

Even (ordinary, masculine) rice (transparent masculine) now vanished (masculine) from the stores.

In the experiment, the participants saw both verb forms in capital letters (IZHEZLA IZHEZ iz magazinov), and they had to choose which gender-marked past tense form was the correct one in the sentence. The results indicated that both natives and non-natives were able to use an adjective inflection present in the modifier to disambiguate the gender of the head noun. However, this had no effect on the reading time of the critical noun. Instead, the effect was delayed and showed itself as shorter reaction times to the verb in the presence of a gender-marked modifier. Taraban and Kempe (1999) suggest that less able language speakers presumably resort to implicit inferencing to compensate for incomplete learning.

Vainio et al. (2003, 2008, 2011) combined the non-violation paradigm with the eye-tracking method to study agreement processing during sentence reading among native Finnish speakers. A key advantage of the eye-tracking method is that it allows readers to freely process the linguistic materials as they wish, thus closely mimicking the natural reading process. Another attractive feature of the method is that it exploits the process of reading and understanding as it proceeds. Measures of the immediate effects of the first-pass reading index obtained during the initial encounter with the target word. In contrast, the measures of the second-pass reading reflect delayed effects, that is, the processing was made after the initial processing. The standard first-pass reading measures are a first fixation duration and a gaze duration on the word, whereas the typical second-pass reading measures are the probability of making a regression back to the target word, and fixation time on the subsequent word (Word N+1).

The results of the two most relevant experiments for the present study on the effects of modifier–head agreement in L1 Finnish are summarized in Table 1. In both studies there were sentences with an agreeing modifier (e.g. Voidaan kysyä, missä määrin isoksi rotaksi haukuttu julkisuuden henkilö on kuvauksensa kaltainen = ‘It may be questioned, to what extent the celebrity, disparaged as the big rat, resembles his description’) and matched sentences where the modifier was absent (e.g. Voidaan kysyä, missä määrin ankaksi haukuttu julkisuuden henkilö kävelee haukkumanimensä mukaisesti = ‘It may be questioned, to what extent the celebrity, disparaged as the duck, walks as implied by his nickname’). It should be noted that in the example the word in italics refers to the modifier, the word in bold signals the head noun, and the underlined word is the word following the head noun. In both studies, the reading patterns concerning the head noun and the noun following the target were used as the dependent variables, as is evident in Table 1.

Table 1.

Processing of modifier–head agreement by native Finnish speakers.

	Agreement	Case	Agreement × Case
Vainio et al. (2008, Exp.1): Average word length 7.7 characters; inessive and translative cases: Measure for head noun:
First fixation duration	no*	no	no
Gaze duration	no	no	no
Probability of rereading	✓	no	no
Vainio et al. (2008, Exp.1): Average word length 7.7 characters; inessive and translative cases: Measure for word N+1:
First fixation duration	✓	no	no
Gaze duration	✓	no	no
Vainio et al. (2011); Average word length 14.8 characters; inessive and translative cases: Measure for head noun:
First fixation duration	no*	no	no
Gaze duration	✓	trend	no
Vainio et al. (2011); Average word length 14.8 characters; inessive and translative cases: Measure for word N+1:
First fixation duration	✓	no	no
Gaze duration	✓	no	no

Note. * the frequency of Word N–1 was used as a covariate.

Table 1 shows that when the target word has fewer than eight characters, facilitation in the modifier–head agreement in L1 processing is seen as a delayed effect, i.e. in the rereading probability of the target and in the first-pass reading of Word N+1. However, when the target word is about 15 characters long, an effect of modifier–head agreement can already be seen in the first-pass reading of the target, albeit the effect is also seen as a spill-over effect in the first-pass reading of Word N+1. These studies indicate that the agreement effect requires some time to develop. With short head nouns, the effect manifests in the second-pass reading measures, whereas with long head nouns, which take longer to be identified, the effect already starts to emerge during the first-pass reading. As reading in L2 is generally slow, it is possible that agreement processing may catch up with word identification, in which case one might already see an agreement effect on the target word itself.

Among eye-tracking studies of L2 language processing, the non-violation paradigm has been most extensively employed. A subset of this paradigm is briefly mentioned next to illustrate that non-violation paradigms can successfully be employed to study L2 language processing. However, this subset is not reviewed in detail because the questions they investigated are not directly relevant to the present study. Bax (2013) studied how non-native (Malesian) English speakers read grammatically correct English passages during the language test; the results showed that successful test takers read faster and focused more efficiently on important aspects of the texts than unsuccessful test takers. Godfroid et al. (2013) presented to the participants grammatically correct sentences to inspect L2 vocabulary acquisition. Their results indicate that the longer participants looked at the unknown pseudoword in the text, the better they were able to recognize that word in the vocabulary post-test. Siyanova-Chanturia et al. (2011) examined the processing of idioms in L1 and L2 English. Their results imply that non-natives processed idioms as if they were novel phrases, and that figurative text is processed more slowly than literal text. Finally, Miwa et al. (2013) investigated how Japanese speakers processed L2 English words. The results suggest that the reaction times and eye fixation durations were not determined by the cognate status, but instead by L1 Japanese word frequency and cross-language semantic and phonological similarities. Readers interested in the topic of using eye-tracking to inspect L2 processing and L2 acquisition should consult a recent review article of Roberts and Siyanova-Chanturia (2013).

In a language such as Finnish, the use of the non-violation paradigm is preferable, because even if the case marking is transparent and the modifier–head marking quite regular, there are exceptions. Modifiers in the determiner position (see Table 2) in a noun phrase do not always agree with the head as the case ending might be idiosyncratic; in which case, there is a subset of adjectives that do not agree in the modifier position. These exceptions are difficult for L2 Finnish speakers and would make the use of the violation paradigm challenging.

Table 2.

Typological features of Finnish, Russian, and Chinese.

	Finnish	Russian	Chinese
Property word	Adjective	Adjective	Verb
Noun phrase order	Fixed	Fixed	Fixed
Noun phrase structure	Determiner + Adjective + Noun	Determiner + Adjective + Noun	Determiner + Numeral Classifier + Adjective + Noun
Modifier–head agreement	Case + Number + Possession	Case + Number	–
Classifier system	–	Gender	Numeral Classifiers
Semantic roles	Case	Prepositions + Case	Coverbs + Prepositions + Pragmatics + Word Order
Locative phrase	Noun + Case	Preposition + Noun + Case	Preposition + Noun + Locative Particle
Category per word	1 + 5	1 + 4	1 + 0

IV Finnish, Russian, and Chinese from a typological point of view

Being a synthetic language with 14 cases, all of which participate in case agreement in noun phrases (and with no gender), Finnish is a highly suitable language for studying case agreement in modifier–head phrases. The phrase-level word order is rigid in Finnish, as the (adjectival) modifier always precedes its head noun without any intervening words. In addition, the modifier strictly agrees in case (and number) with its head (see Sulkala and Karjalainen, 1992). Russian is also a synthetic language with six cases and three genders. Every noun has a gender, which influences the form of its modifiers (and the finite verbs as well). The form varies depending on the declension. On a scale from fully agglutinative to fully fusional, the scale referring to the degree of transparency in morpheme boundaries, a synthetic language can be more or less fusional. In a clearly agglutinative language the boundaries are transparent (i.e. one meaning – one form principle); on the other hand, in a clearly fusional language one form might express several meanings. In this scale Finnish and Russian differ to some degree, Russian being more fusional. However, Finnish is not a pure agglutinative language:

there are many kinds of stem alternations (käsi: käde-: kät- stems of the word ‘hand’);

the boundaries between stems and affixes are not always clearcut (mäki+i+a ‘hill+PL+Part > mäki +ä);

some suffixes change when attached to stems (see mäkiä but kala+PL+PART > kaloja).

For the effects of morphophonological transparency on L2 word processing, see Vainio et al. (2014).

Chinese, in contrast, is an analytic language that mostly uses sentence-level word order and particles to express case roles. The phrase-level word order is rigid in all three languages. It is noteworthy that Chinese uses a unit (de) which converts verbs into modifiers and functions as a link between the modifier and the head (for a description of the three languages, see Table 2).

The complexity of words in a given language is measured by the maximum number of formatives words that it can contain, producing a category-per-word (CPW) value. CPW expresses the number of affixes with differing grammatical meanings in nouns or verbs. For example, in the Finnish word talo-i-ssa-ni-ko-han (CPW 1+5), there is one lexical unit and five formatives (expressing the corresponding meanings: ‘house - number (plural) - case (inessive) - possession (1st singular possessive) - question (ko), emphasis (clitic = ‘too’)’; Itkonen and Pajunen, 2011). On the other hand, in Chinese words are structurally very simple (Li and Thompson, 1981). Thus, CPW is much higher in Finnish (4–5) and Russian (4–5) than in Chinese (1). In this sense, Finnish and Russian are typologically similar to each other and different from Chinese; phylogenetically all three languages are unrelated.

Russian and Chinese resemble each other as regards noun classification. In both languages, there is a classifier system that determines the relationship between the head noun and some other element in the phrase. In Russian, the classifier system is a highly grammaticalized gender system that ascribes a feminine, masculine or neuter form to the modifier (Corbett, 1991). In Chinese, countable, definite nouns are provided with a measure element that classifies the noun according to its inherent properties. The classifying particle precedes the modifier–head combination (Li and Thompson, 1981: 117). Example (6) clarifies the point:

(6) tā shi yi ge hăo rén

3sg be one CL good person; CL is a classifier

‘s/he is a good person’

The classifier system forces the language users to pay close attention to the meaning of the head noun and also to the overall context, as every noun can have several acceptable classifiers (Aikhenvald, 2000). By contrast, Finnish does not have any kind of classifier system. In this sense, Russian and Chinese are typologically similar to each other and different from Finnish.

V Alternative hypotheses concerning agreement processing in L2 Finnish

In principle, there is a continuum of possible outcomes of agreement processing in L2 versus L1. At one end of the spectrum, there is the possibility for different type of processing. Thus, a facilitatory agreement effect will be seen in L1, but L2 learners would show no agreement effect. This could indicate that intermediate-level L2 proficiency is not sufficient for utilizing modifier–head case agreement, despite the fact that the Finnish agreement system is simple and based on general principles (i.e. not lexeme-based). The problem with this sort of hypothesis is that the null effect for L2 learners might also be due to some intervening uncontrolled factors.

At the other end of the continuum, there is the possibility that L2 learners would display processing highly similar in nature to that of L1 speakers (although being generally slower in their processing, L2 learners may reveal agreement effects with longer delay than L1 learners). This type of result would have two implications. First, the result would suggest that structural transfer from L1 to L2 is not crucial in learning modifier–head agreement and, second, Chinese speakers can learn morphologically based modifier–head agreement despite the lack of case morphology in their L1. As mentioned above, both Russian and Chinese have a noun classification system. In Chinese, the classifying particle precedes the modifier–head combination (Li and Thompson, 1981). Both classifier systems encourage readers to pay attention both to the meaning of a head noun and the modifier as well as the classifier that precedes them. Thus, linking of classifiers to their head nouns and processing morphological agreement between a modifier and its head entails integration of word-internal and word-external information. A similar idea has been put forward by Koda (2000), who proposes a general transfer system, in which a language learner makes use of a general language ability learnt in L1. Most importantly in the present context, Koda suggests that Chinese speakers are adept at integrating word-internal and contextual (word-external) information because of their experience in using radical (Chinese written words are formed by radicals) and contextual information in word identification in Chinese. It can also be noted that the meaning and pronunciation of Chinese characters are context dependent in that a character may mean different things depending on the context. The hypothesis that both non-native groups are able to utilize modifier–head agreement is also compatible with the shallow structure hypothesis (SSH), which suggests that typological distance between the L1 and L2 does not affect L2 sentence processing, as L2 speakers of very different L1s have shown similar processing patterns in L2 (Clahsen and Felser, 2006b). This is also mentioned by Clahsen and Felser (2006a), who stated: ‘The SSH claims that during L2 processing, learners compute grammatical representations that lack complex hierarchical structure and abstract, configurationally determined elements such as movement traces, and that native-like grammatical processing is restricted to “local” domains such as word segmentation or morphosyntactic agreement between closely adjacent constituents’ (p. 111). Thus, according to SSH, L2 speakers are capable of native-like processing of morphosyntactic agreement between adjacent constituents, which is the case in modifier–head agreement in Finnish.

Finally, hypotheses taking an intermediate position may also be considered. First, the typological distance between L2 and L1 may play an important role. Thus, L1 Russian speakers may make use of modifier–head agreement in Finnish, whereas L1 Chinese speakers will not. This hypothesis rests on the fact that Russians have in their L1 morphologically overt modifier–head agreement, which is similar to Finnish in that the agreement is presented by suffixes in both languages. Although both of the non-native groups in the study have prior experience in inspecting cohesion within phrases, the Chinese speakers did not have experience from their native language in combining case information and grammatical roles, because they do not have cases in their L1. In general, the predicted results suggest that transfer is based on similar grammatical representations in L1 and L2.

The above hypothesis is also supported by our previous finding (Vainio et al., 2014) obtained for the same Chinese and Russian (partially) participants enrolled in the present study. In the previous study, we compared the recognition of three kinds of words presented in isolation: morphologically simple nouns in nominative singular (e.g. koulu = ‘school’), transparently inflected nouns in partitive singular (e.g. tuoli+a = ‘chair+PART’), and semi-transparently inflected nouns in the genitive singular (e.g. ken k ä → ken g ä+n = shoe+GEN). The pattern of results was taken to suggest that native Finnish and Russian speakers made use of suffixal information contained in inflected Finnish words, but native Chinese speakers did not. Based on this finding, it can be predicted that the intermediate-level Chinese participants cannot utilize modifier–head agreement in processing Finnish, as modifier–head agreement is marked by case inflections.

VI Present study

The question studied was whether typological differences in the structural complexity of words between the native language and the language to be learned influence L2 processing; this was done by testing how native Chinese and Russian adults process modifier–head agreement in Finnish. A group of native participants was also included, which enabled an additional comparison between native and non-native processing in order to discover to what extent L2 processing is native-like or fundamentally different from native Finnish speakers.

Three important points have to be made concerning the present study. First, Finnish is a suitable language for testing case agreement effects in L2 processing because Finnish is a highly inflecting language, especially regarding noun inflections. The Finnish locative cases (i.e. inessive, elative, illative, adessive, ablative, allative) constitute a group with a rather unequivocal semantic function. To avoid concerns that the potential agreement effects would be too case specific, two of these six locative cases – inessive and illative – were used. Both cases express spatial meaning in the same sense as the English prepositions in (= inessive) and into (= illative), but they also sometimes have a more abstract meaning (e.g. ‘for somebody’). These two inflections were also chosen because they differ in morphological transparency. Agreement containing the inessive case is orthographically and phonologically transparent, as the inflection takes only one of two forms, either -ssa or -ssä, depending on the vowels of the stem (e.g. pienessä talossa = ‘small-in house-in’). The illative, on the other hand, is formed by lengthening the last vowel of the stem and adding -n to the end. Thus, the illative takes several forms, which typically are not shared by the modifier and the head (e.g. suureen taloon = ‘big-into house-into’).

Second, processing of modifier–head agreement was studied using readers’ eye fixation patterns as indices of online processing. Over the years, the eye-tracking method has established itself as the benchmark method for studying online processing of written text (for a review, see Rayner, 1998). This is because eye movements are a natural part of reading; thus, no extra task other than reading is needed to measure ongoing language comprehension. Moreover, the method provides measures sensitive enough to disentangle the comprehension process as it evolves over time. As reviewed above, Vainio et al. (2003, 2008, 2011) have successfully applied the method to the study of modifier–head agreement among native Finnish speakers. They observed an agreement effect either in the head noun or in the word following the head noun; therefore, eye fixation measures in these two regions were also examined in the current study.

Finally, all three languages have a different orthography: The Latin alphabet in Finnish, the Cyrillic alphabet in Russian, and a logographic script in Chinese. All the Chinese participants had had prior exposure to a Latin alphabet via their studies of an alphabetic Pinyin script in elementary school, and they had also studied English prior to Finnish. In addition to Finnish, most of the Russian participants had studied either German or English.

VII Method

1 Participants

Sixteen native-Chinese learners of L2 Finnish (13 female and 3 male students at a Finnish university, age range 20–22 years, median 21 years), 16 native-Russian learners of Finnish (14 female and 2 male students of Finnish universities, age range 19–48 years, median 22 years), and 22 native speakers of Finnish (20 female and 2 male students at a Finnish university, age range 19–26 years, median 20 years) participated in the experiment. All participants had normal or corrected to normal vision. The Chinese participants had all passed an intermediate level (B1) proficiency test (Common European Framework of Reference for Languages: Learning, Teaching, Assessment¹) and had studied Finnish in intensive courses for 23 months. Five Russian participants were involved in a preparation course for the B1-level proficiency test. At the time of testing, they had acquired pre-intermediate level (A2.1) competence and had studied Finnish for 24–72 months (median 36 months). Seven Russian participants were short-term exchange students at Finnish universities and had studied Finnish for 12–36 months and 36 (median 35 months), respectively; the remaining six Russian participants were taking part in Finnish-for-foreigners courses at a Finnish university and had studied Finnish for 18–36 months (median 35 months). All the Chinese participants had also studied English, 13 Russians had studied English, and 4 had studied German or other languages.

2 Apparatus

Eye movements were collected by an EyeLink II eye-tracker manufactured by SR Research Ltd (Canada). The eye-tracker is an infrared video-based tracking system combined with hyperacuity image processing. There are two cameras mounted on a headband (one for each eye) including two infrared LEDs for illuminating each eye. The headband weighs 450 g in total. The cameras sample pupil location and pupil size at the rate of 500 Hz. Registration can be done either monocularly or binocularly. The data were collected from the dominant eye, which was usually the right eye, by placing the camera and the two infrared lights 40–60 mm away from the eye. The resolution of the eye position is 15 seconds of arc and the spatial accuracy above 0.5 degrees. The head position with respect to the computer screen is tracked via a head-tracking camera. Four LEDs are attached to the corners of the computer screen, which are viewed by the head-tracking camera, once the participant sits directly facing the screen. Possible head motion is detected as movements of the four LEDs and is compensated for on-line from the eye position records.

3 Materials

Twenty Finnish nouns in the illative singular case and 20 in the inessive singular case were used as the targets in the experiment; all nouns were 7–8 characters long. Two case endings were used for two reasons: (1) to make sure that the possible effects of agreement were not case specific; (2) to make sure that the agreement effects were not based on morphophonological repetition of the same case ending (as is the case in inessive) in the modifier and head (see below, for examples). The inessive is orthographically and phonologically more transparent than the illative. The inessive takes only one of two forms, -ssa or -ssä, depending on the vowels of the stem. On the other hand, the illative is less transparent. It is typically formed by lengthening the last vowel of the stem and adding -n to the end. With eight vowels existing in Finnish, it can take one of eight forms. Moreover, the modifier and the head typically contain different forms (e.g. suureen taloon = ‘big-illative house-illative), which makes modifier–head agreement orthographically and phonologically semi-transparent.

Modifier–head agreement was manipulated using two conditions: The modifier present vs. the modifier absent. Thus, the experimental setting is similar to that of Taraban and Kempe (1999). All the sentences were presented in Courier New font with a font size of 13 in one text line; the target word appeared approximately in the middle of the sentence. All the target nouns were frequent and were taken from the learners’ course materials. As the same target nouns were used in the two agreement conditions, they were perfectly matched. The lexical statistical properties of the target nouns were as follows: The logarithmic lemma frequency (i.e. cumulative base frequency) was 3.4 (SD = 0.5) for the illative case and 3.4 (SD = 0.4) for the inessive case; logarithmic surface frequency (i.e. the frequency of the surface form) was 2.2 (SD = 0.3) for the illative case and 2.3 (SD = 0.4) for the inessive case; length in characters was 7.4 (SD = 0.5) for the illative case and 7.5 (SD = 0.5) for the inessive case. The frequency measures were based on Turun Sanomat newspaper corpus of 22.7 million word tokens (Laine and Virtanen, 1999; Virtanen and Pajunen, 2000).

Two examples of the experimental sentences are presented below. The target noun appears in bold, the agreeing modifier is in brackets, and the agreeing morphemes are underlined. Each participant read only one version of each target sentence, i.e. with or without the modifier. Two sentence lists were created that included 20 modifier-present sentences and 20 modifier-absent sentences. Only one version of each sentence pair appeared in each list. The two lists were counterbalanced across participants for all language groups.

4 Targets

Set 1: Targets in illative case, shown in (7):

(7) On epäselvää, milloin (pitkään) lakkoon voidaan löytää sopuratkaisu.

Is unclear when (long-illative) strike-illative can (passive) find accommodation.

‘It is unclear when a reconciliation can be found for the (long) strike.’

Set 2: Targets in inessive case, shown in (8):

(8) Toimittajan mukaan (vanhassa) junassa havaittiin häiriöitä liian myöhään.

Journalist according to (old-inessive) train-inessive discovered (passive) malfunctions too late.

‘According to the journalist the malfunctions in the (old) train were discovered too late.’

In addition, there were 36 filler sentences that appeared in different syntactic structures; hence, participants read altogether 76 sentences.

After the experiment proper, the L2 participants provided written translations of the target nouns in their L1, but they were also allowed to translate words into English; the words were given in the nominative singular, i.e. without bound morphemes. Depending on their knowledge of English, approximately half answered in their L1 (Russian or Chinese) and half in English. If the participant did not know the critical noun, his or her data for the sentence containing that word was excluded from the analysis. Three items were entirely excluded from the L2 data because they were poorly known by the non-native participants. Note that both the Finnish case system and modifier–head agreement are taught in the first elementary course of Finnish (see, e.g. Lauranto 1996), as learning Finnish is based on learning morphology (White, 1997).

Skeptics may argue that possible facilitation seen in the processing of the head noun does not necessarily reflect modifier–head agreement, but is instead based on the increased predictability of the head. Three types of action were taken in order to prevent this from occurring. First, the sentence materials were constructed in such a way that the neutral sentence frame leading to the head noun did not semantically constrain it (see the examples above). Second, each modifier was chosen from among the most frequent and common adjectives (e.g. old, new, small, big, good, bad, long) that combine with a great number of nouns and thus do not predict any particular noun. Third, 10 native Finnish speakers (university students; one male, nine females; average age 21.5 years, age range 20–24 years) carried out a sentence continuation task where they had to add a word that would be a suitable continuation for a sentence fragment. Only the modifier present versions of the target sentences combined with a set of filler sentences were tested. The order of the sentences was individually randomized. The results showed that the target noun was never provided as the sentence continuation; on the other hand, the case of the provided word nearly always (96.0%) agreed with the modifier. Hence, based on the above evidence it seems highly implausible that either the sentence beginning or the adjective modifier would help predict the identity of the head noun in the current material.

5 Procedure

Prior to the experiment, the eye-tracker was calibrated using a 3-point horizontal calibration grid. It should be noted that all the sentences appeared in one line of text. Prior to each sentence, the calibration was checked by presenting a fixation point on the left corner of the screen coinciding with the location of the first letter of the sentence about to be presented. If necessary, the calibration was automatically corrected. Participants were instructed to read sentences silently for comprehension at their own pace. They were further told that periodically they would be asked to paraphrase the last sentence they have just read to make sure that they were paying attention to what they read. However, it was emphasized that the task was to comprehend, not to memorize, the sentences. Participants were asked to paraphrase the sentence approximately after every 10th sentence. The experimental session was divided into two sections, and it lasted a maximum of 45 minutes.

6 Eye fixation measures

The analyses were confined to the head noun and the word following the head (Word N+1). This was done for two reasons. First, prior research on native language processing has identified these regions as the relevant ones (see Table 1). Second, we exerted tight experimental control over these regions but not across the entire sentence. Eye fixation measures of word processing in reading are broadly categorized to first-pass and second-pass measures. Gaze duration (GD) was used as an index of first-pass reading (first fixation duration did not show any significant effects). GD is the sum of all the fixations made on the word until the reader either goes forward in the sentence or looks back to earlier parts of the sentence. This measure reflects the difficulty of processing words in sentence context as well as immediate syntactic effects in reading. Total fixation time (TOT), which is the sum of all fixations made on the target word during the first-pass and second-pass reading, reflects later integration of the word into the sentence context. Probability of regression (RegProb) into the target word indicates the probability of returning back to the target word from later parts of the sentence.

VIII Results

As mentioned earlier, L2 participants provided written translations of the target nouns in their L1 after the experiment. If the participant did not know the critical noun, his or her data for the sentence containing that word were excluded from the analysis. In addition, in all the trials where gaze duration was less than 100 ms, when the critical noun was initially skipped, or when there was a track loss they were excluded. Overall, the missing data amounted to 3.9 % for the Finnish L1 group, 24.1% for the Russian L1 group, and 30.8 % for the Chinese L1 group, respectively. It should be noted that that the missing data was mostly a result of L2 speakers not knowing the target noun (14.7% for the Russian L1 group, and 19.3% for the Chinese L1 group). This reflects our interest in examining L2 processing among intermediate level learners. However, it should be noted that the missing data introduces more variability in the mean estimates and thus weakens the possibility of finding significant agreement effects among L2 learners.

Due to the considerable variation between and within L2 participants, the fixation time data were log-transformed. However, to allow direct comparisons to earlier studies, the data presented in Table 3 are reported without log-transformation.

Table 3.

Mean gaze duration (ms), Probability of regression (%), and Total fixation time (ms), and Their standard deviations (in parentheses), for the Target nouns; and Gaze duration (ms) for the Word following the target (Word N+1).

Fixation time measure	Finnish		Russian		Chinese
Fixation time measure	Modifier present	Modifier absent	Modifier present	Modifier absent	Modifier present	Modifier absent
Gaze duration	296 (53)	297 (67)	552 (104)	592 (97)	616 (166)	650 (171)
Probability of regression	16.7 (14.5)	26.8 (16.1)	46.0 (24.0)	56.0 (28.4)	31.1 (15.6)	38.3 (20.9)
Total fixation time	455 (180)	487 (229)	1191 (483)	1386 (538)	1183 (612)	1359 (655)
Gaze duration on N+1	300 (51)	324 (64)	608 (145)	629 (152)	703 (191)	745 (205)

Three sets of analyses are reported. In the first set, two groups of non-native Finnish speakers (Russian and Chinese) were compared to each other. These analyses address the question of whether the typological distance between L1 and L2 has an effect on processing modifier–head agreement in L2. In the second set, the non-native speakers were compared to native Finnish speakers. These analyses examine possible differences in agreement processing between non-native and native speakers of Finnish. In both sets of analyses, a 2 × [native language: Russian vs. Chinese; native Finnish speakers vs. non-native Finnish speakers) × 2 (agreement: agreeing modifier present vs. absent) × 2 (case: illative vs. inessive)] mixed design ANOVA was performed on the data with agreement as a within-participant and within-item variable, case as a within-participant and a between-item variable, and L1 as a between-participant and a within-item variable. Only the results that were at least marginally significant are reported, except those concerning L1 × Agreement interactions and their follow-up analyses. The analyses were performed for the target word and the word following the target (Word N+1). Immediate agreement effects were measured by gaze duration on the target noun. Lagged or spillover effects of agreement were examined by total fixation time on the target noun, the probability of regression to the target noun, and gaze duration on the word following the head noun (Word N+1). For the last analyses, only those trials were included where the eyes went directly from the target noun to Word N+1. Fortunately, the percentage for such cases was relative high for all groups (92.2% for L1 Finnish speakers, 86.5% for L1 Chinese speakers, and 86.7% for L1 Russian speakers). The above percentages were calculated for cases where the target noun was included in the analysis.

1 Russian–Chinese comparison

The ANOVA results for the comparison between native Russian and Chinese speakers are presented in Table 4. As is apparent from Table 4, a reliable main effect of agreement was observed, but no significant interactions emerged involving the native language as a factor. The agreement effect was most robust in the total fixation time, although there was already a tendency in the gaze duration and also in the probability of regression to the target noun. The results suggest that the typological distance between L1 and L2 is not a critical factor in the ability to utilize modifier–head agreement in L2 Finnish.

Table 4.

ANOVA results for the Chinese–Russian comparison.

	F₁	η_p²	p	F₂	η_p²	p
Target Noun: Gaze duration:
Agreement	4.34	.130	.046	2.97	.078	.094
Case	7.82	.212	.009	5.97	.146	.020
L1	.55	.019	.465	3.89	.100	.057
L1 × Agreement	.18	.006	.672	.01	< .001	.934
Target Noun: Total fixation time:
Agreement	18.11	.384	< .001	19.71	.360	< .001
L1	.25	.009	.621	4.89	.123	.034
L1 × Agreement	.07	.002	.793	.01	< .001	.912
Target Noun: Probability of regression:
Agreement	3.31	.103	.079	5.28	.131	.028
L1	5.02	.148	.033	36.68	.512	< .001
L1 × Agreement	< .01	< .001	.951	1.40	.038	.245
Word N+1: Gaze duration:
Agreement	.43	.014	.519	1.53	.042	.224
L1	3.11	.094	.088	13.52	.279	.001
L1 × Agreement	1.75	.055	.196	2.63	.070	.114

Note. p values for the main effect of agreement and the Agreement × L1 interaction appear in bold.

2 Native-non-native comparison

In the second set of analyses, the two non-native learner groups were contrasted with native Finnish speakers, i.e. the Russian and Chinese speakers were merged into one group of non-native Finnish speakers. The ANOVA results for this comparison are presented in Table 5. As is apparent from Table 5, a significant main effect of agreement emerged in two second-pass measures, total fixation time, and probability of regression. Both measures displayed a facilitation effect due to the presence of modifier–head agreement. Moreover, there was a reliable interaction in total fixation time and gaze duration for Word N+1 Word. As was presented earlier, the non-native group comparison demonstrated a significant main effect of agreement in total fixation time for the target noun, but no effect in gaze duration for Word N+1. Thus, the observed interactions suggest an opposite pattern for native speakers. In order to confirm this, the data for the native speakers were analysed separately.

Table 5.

ANOVA results for the native–nonnative comparison.

	F₁	η_p²	p	F₂	η_p²	p
Target Noun: Gaze duration:
Agreement	1.90	.035	.174	1.57	.043	.219
Case	8.24	.135	.006	4.94	.124	.033
L1	151.4	.741	< .001	992.8	.966	< .001
L1 × Agreement	2.90	.052	.094	2.77	.073	.105
Target Noun: Total fixation time:
Agreement	14.90	.219	< .001	9.43	.212	.004
L1	83.97	.613	< .001	1034.2	.967	< .001
L1 × Agreement	4.46	.078	.040	3.52	.091	.069
Target Noun: Probability of regression:
Agreement	11.26	.175	.001	23.13	.398	< .001
L1	15.93	.231	< .001	71.64	.672	< .001
L1 × Agreement	.27	.005	.607	.01	< .001	.916
Word N+1: Gaze duration:
Agreement	.926	.017	.340	.552	.015	.462
L1	136.8	.717	< .001	320.1	.899	< .00
L1 × Agreement	3.93	.068	.053	4.69	.115	.037

Note. p values for the main effect of agreement and the Agreement × L1 interaction appear in bold.

The ANOVA results for the native speakers are presented in Table 6. As suggested above, the native speakers, in fact, showed no agreement effect in total fixation time, but a reliable agreement effect for gaze duration on Word N+1. These results are highly similar to those obtained earlier with comparable materials (see Table 1).

Table 6.

ANOVA results for the native speakers of Finnish.

	F₁	η_p²	p	F₂	η_p²	p
Target noun: Total fixation time: Agreement	1.55	.063	.225	.91	.023	.345
Word N+1: Gaze duration: Agreement	10.05	.304	.004	10.77	.216	.002

Note. p values for the main effect of agreement appear in bold.

IX General discussion

To summarize, we observed that intermediate-level non-native speakers of Finnish made use of modifier–head agreement in processing the head nouns. It should be observed that reliable agreement effects were found despite the small size of the L2 groups and the relatively large amount of missing data, both of which reduced the possibility of obtaining an agreement effect. Moreover, the agreement effect was of a similar size for the native Chinese and native Russian learners of Finnish. A hint of a facilitative agreement effect was observed in the gaze duration for the head noun, and it was fully significant in the probability of regression and total fixation time, both of which indicate delayed effects in written language processing. When the head noun was preceded by an agreeing modifier, the gaze duration on the target noun was shorter, the probability of making a regression back to the target was smaller, and the total fixation time on the target was shorter rather than when no modifier was present. In addition, native speakers of Finnish demonstrated a facilitatory effect of agreement that manifested as a shorter gaze duration on the word following the head noun and as a smaller probability of making a regression back to the target. These results are comparable to those obtained in a similar earlier study (Vainio et al., 2008). Thus, for all reader groups the processing of the head noun was easier when it was preceded by an agreeing modifier compared to a situation where the modifier was absent. The only difference between the native and non-native speakers was the spatial location of the effect, the non-native speakers already showing some signs of facilitation by agreement on the head noun itself, whereas for native speakers the effect appeared only as a delayed effect (i.e. they proceeded with their eyes beyond the head noun before the effect showed itself). It should be noted, however, that even for non-native learners the most robust agreement effect was observed as a delayed effect (a 186 ms effect in total fixation time versus a 38 ms effect in gaze duration). Only the exact location of the delayed effect differed between non-native and native language speakers. The finding that the delayed effect was observed in the target word rereading among non-native speakers but not among the native speakers may reflect the fact that non-native speakers generally need more extensive visual input to support their written language processing.

Ostensibly, the effect of the first-pass reading of the head word obtained for non-natives seems to imply that the agreement effect materialized faster for non-natives than natives. However, this is not the case, as when looking at the timing of the agreement effect, it appears that in both groups it is seen around 600 ms after the head noun is first fixated. It took about 600 ms for the non-natives to read the head noun during the first-pass reading, and it took about the same time for the natives to read the head noun and the following word (which was the region demonstrating the effect). As shown in Table 1, native Finnish speakers also already showed an agreement effect in the first pass reading of the target, when the head nouns were very long (14–16 characters). Their first-pass reading times (570 ms; Vainio et al., 2011) were comparable to those of non-natives’ reading times in the present study. As a whole, these results converge on a view that it takes a certain amount of time to make use of the case agreement between the modifier and the head. Interestingly, the required time (about 600 ms) is similar to the P600 effect observed in EEG recordings for syntactic manipulations. The above argumentation rests on the assumption that the lexical and syntactic processing stages are not strictly serial in that lexical processing would need to complete before syntactic processing is initiated. This assumes that a cascaded or parallel process takes place at the lexical and syntactic levels. Overall, it may be speculated that the agreement effect reflects syntactic integration between the modifier and the head (for an analogous interpretation, see also Vainio et al., 2003, 2008, 2011). The finding that the most robust agreement effects were observed as delayed effects also clearly supports this view.

It may be noted that the type of case inflection did not modulate the observed agreement effects; i.e. there was no significant Case × Agreement interaction. It should be recalled that, in the materials, a morphophonologically transparent (inessive) and semi-transparent (illative) case were employed. The lack of modulation by case type implies that morphophonological transparency does not play a crucial role in processing modifier–head agreement even for intermediate-level non-native speakers. The finding that all reader groups showed longer reading times for inessives than for illatives is also evidence that morphophonological transparency does not play a significant role in processing. If this were to be the case, then the more transparent inessive cases should have been easier to process overall than the less transparent illatives. This effect may reflect the fact that the inessive contains one extra syllable compared to the illative.

The current study employed the non-violation agreement paradigm, in which no grammatical violation was made to the modifier–head agreement. As argued in the Section I, the non-violation paradigm is more advantageous than the violation paradigm in that it allows its user to examine whether a grammatical feature has a truly facilitatory influence on language processing, rather than reflecting a response to a grammatical violation, which is likely to be inhibitory in nature. Because the non-violation paradigm is less robust than the violation paradigm, it should be more difficult to find effects using it. Thus, the observation of reliable modifier–head agreement effects in a non-violation paradigm lends further proof to the trustworthiness of the obtained results. Moreover, in all likelihood the results should be readily replicable using a violation paradigm.

The theoretical motivation of the present study was two fold. On the one hand, we addressed the question raised by Clahsen and Felser (2006c) whether non-native processing is affected by the typological distance between L1 and L2. Typological distance between speakers of L1 and L2 was measured by the morphological complexity of word structures. Words in Chinese are morphologically much less complex than words in Russian and Finnish. The fact that there were no interactions involving the participants’ native language suggests that typological distance – characterized as morphological complexity of words – between speakers’ L1 and L2 does not affect the processing of modifier–head agreement in L2 Finnish.

Moreover, related to the typological issue, it was hypothesized that only Russians are able to process L2 Finnish modifier–head agreement, because in Russian and in Finnish the agreement is shown by the suffixes attached both to the modifier and the head. This hypothesis proved to be wrong, as non-native processing was very similar between the Chinese and Russian speakers despite the L1 structure. In contrast, the present study implies that the structural similarity between L1 and L2 is not necessary when utilizing non-native modifier–head case agreement. This could be interpreted to indicate either that a transfer from L1 to L2 is not necessary for learning modifier–head agreement in L2, as suggested by Clahsen and Felser (2006a, 2006b); although it should be noted that as such SSH does not exclude the possibility of L1 transfer.

Alternatively, the similar agreement effect for the native Chinese and the native Russian learners of Finnish could mean that the late learners of L2 transfer processing routines from L1 to L2 rather than grammatical representations, as proposed by Sabourin and Stowe (2008). This concept is based on the fact that the Chinese participants have ample experience in combining word internal and contextual information in their native language. This is because the meaning and pronunciation of Chinese characters are very context dependent in that a character may mean different things depending on the context. Thus, native Chinese readers of Finnish are equipped with general processing routines to deal with context dependence across words. Such processing routines may then be applied to L2 Finnish to deal with morphological agreement spanning across subsequent words. On the other hand, the Chinese speakers have no native language experience in dealing with case morphology. When the current sample of Chinese participants was tested for their ability to identify isolated case-inflected words in Finnish, they showed no sensitivity to the words’ morphological structure (Vainio et al., 2014). On the whole, their sensitivity to modifier–head agreement demonstrated by the present study and their insensitivity to the morphological structure of words presented in isolation (Vainio et al., 2014), we propose that the results for Chinese participants may indeed be explained by general-purpose processing routines transferred from L1 to L2. For the Russian participants, on the other hand, both the transfer of processing routines and the transfer of grammatical representations are possible, for they have case morphology in their native language, and modifier–head agreement is expressed by case morphology.

In the literature, the notion of processing routines, alluded to above, has taken different forms. At one end of the spectrum, O’Grady (2013) claims that by adopting an emergentist perspective there is no need for grammatical rules or principles, as emerging processing routines do the work of the grammar. At the other end is the universalist approach represented, for example, by Felser et al. (2003). They posit that language processing routines enable language learners to assign structural representations to input strings in order to acquire adult grammar. Our data does not allow us to distinguish between these alternative views, although our results may be more easily accommodated by a version of the emergentist approach.

Both interpretations – no transfer or transfer of processing routines – are consistent with the present data, which cannot differentiate between them. However, it seems clear that the non-natives tested in the present study did not transfer grammatical representations from their L1 to L2 processing. Even though there are apparently no prior studies concerning L2 processing of syntactically based modifier–head case agreement, we nevertheless assume that successful use of agreement seen by L2 speakers will be a general rather than a language specific phenomenon. The notion of transfer of processing routines is generally feasible in the present context, as both Russian and Chinese have a noun classification system. In Chinese, the classifying particle precedes the modifier–head combination (Li and Thompson, 1981), whereas in Russian the classifier system is a highly grammaticalized gender system that ascribes a feminine, masculine, or neuter form to the modifier (Corbett, 1991). Both classifier systems encourage readers to pay attention both to the meaning of the head noun and the modifier as well as to the classifier that precedes them. Thus, linking classifiers to their head nouns and processing morphological agreement between a modifier and its head entails integration of word-internal and word-external information. A similar idea has been put forward by Koda (2000), who proposed that Chinese speakers are adept at integrating word-internal and contextual (word-external) information because of their experience in using radical (Chinese written words are formed by radicals) and contextual information in word identification in Chinese. In sum, the aforementioned views make the transfer of processing routines a viable option.

In conclusion, the present study has demonstrated that intermediate-level learners of Finnish are capable of utilizing modifier–head agreement when reading Finnish sentences. It also suggests that typological distance, measured as the complexity of the morphological structure of words between L1 and L2, does not have a significant influence on processing L2 modifier–head case agreement. The present results are compatible with the view that prior experience of combining word-internal and word-external information in a language is sufficient to achieve the ability to utilize L2 modifier–head case agreement. Moreover, the results are also compatible with the idea that learners may transfer processing routines acquired in L1 to the processing of L2.

Footnotes

Acknowledgements

We are grateful to the editor and to anonymous reviewers for their constructive criticism and to Esa Itkonen for his comments on an earlier version of this article. We are most indebted to Svetlana Myhkyrä, Maija Tervola, and Heikki Kangasniemi for their invaluable work recruiting participants for the study. We also thank Tatiana Petrova, An Yang, and Li Xin for their native speaker insights into Russian and Chinese.

Declaration of Conflicting Interest

The authors declare that there is no conflict of interest.

Funding

This research was financially supported by a grant from the Academy of Finland to the first author (Grant #134498).

Notes

References

Aikhenvald

(2000) Classifiers: A typology of noun categorization devices. Oxford: Oxford University Press.

Akhutina

Kurgansky

Polinsky

. (1999) Processing of gender in a three-gender system: Experimental evidence from Russian. Journal of Psycholinguistic Research 28: 695–713.

Barber

Carreiras

(2013) Grammatical gender and number agreement in Spanish: An ERP comparison. Journal of Cognitive Neuroscience 17: 137–53.

Bax

(2013) The cognitive processing of candidates during reading tests: Evidence from eye-tracking. Language Testing 30: 441–65.

Clahsen

Felser

(2006a) Grammatical processing in language learners. Applied Psycholinguistics 27: 3–42.

Clahsen

Felser

(2006b) Continuity and shallow structures in language processing: A reply to our commentators. Applied Psycholinguistics 27: 107–26.

Clahsen

Felser

(2006c) How native-like is non-native language processing? Trends in Cognitive Sciences 10: 564–70.

Corbett

(1991) Gender. Cambridge: Cambridge University Press.

Davidson

Hanulíková

Indefrey

(2012) Electrophysiological correlates of morphosyntactic integration in German phrasal context. Language and Cognitive Processes 27: 288–311.

10.

Felser

Marinis

Clahsen

(2003) Children’s processing of ambiguous sentences: A study of relative clause attachment. Language Acquisition 11: 127–63.

11.

Gillon-Dowens

Guo

. (2011) Gender and number processing in Chinese learners of Spanish: Evidence from event related potentials. Neuropsychologia 49: 1651–59.

12.

Gillon-Dowens

Vergara

Barber

. (2009) Morphosyntactic processing in late second-language learners. Journal of Cognitive Neuroscience 22: 1870–87.

13.

Godfroid

Boers

Housen

(2013) Gauging the role of attention in incidental L2 vocabulary acquisition in late second-language learners by means of eye-tracking. Studies in Second Language Acquisition 35: 483–517.

14.

Gurjanov

Lukatela

Moscovljevic

. (1985) Grammatical priming of inflected nouns by inflected adjectives. Cognition 19: 55–71.

15.

Hyönä

Hujanen

(1997) Effects of case marking and word order on sentence parsing in Finnish: An eye fixation analysis. Quarterly Journal of Experimental Psychology 50A: 841–58.

16.

Hyönä

Lindeman

(1994) Syntactic context effects on word recognition: A developmental study. Scandinavian Journal of Psychology 35: 27–37.

17.

Itkonen

Pajunen

(2011) A few remarks on World Atlas of Language Structures. In: Itkonen

(Ed.) Papers on typological linguistics. Publications in General Linguistics 15: University of Turku, 102–39.

18.

Keating

(2009) Sensitivity to violations of gender agreement in native and non-native Spanish: An eye-movement investigation. Language Learning 59: 503–35.

19.

Kempe

MacWhinney

(1998) The acquisition of case marking by adult learners of Russian and German. Studies in Second Language Acquisition 20: 543–87.

20.

Koda

(2000) Cross-linguistic variations in L2 morphological awareness. Applied Psycholinguistics 21: 297–320.

21.

Laine

Virtanen

(1999) WordMill lexical search program [computer software]. Turku: Centre for Cognitive Neuroscience University of Turku.

22.

Laine

Vainio

Hyönä

(1999) Lexical access routes to nouns in a morphologically rich language. Journal of Memory and Language 40: 109–35.

23.

Lauranto

(1996) Elämän suolaa. Suomen kielen alkeita 1 [Basics of Finnish]. Jyväskylä: University of Jyväskylä.

24.

Lew-Williams

Fernald

(2010) Real-time processing of gender-marked articles by native and non-native Spanish speakers. Journal of Memory and Language 63: 447–64.

25.

Thompson

(1981) Mandarin Chinese: A functional reference grammar. Berkeley, CA: University of California Press.

26.

Miwa

Dijkstra

Bolger

. (in press) Reading English with Japanese in mind: Effects of frequency phonology and meaning in different-script bilinguals. Bilingualism: Language and Cognition.

27.

O’Grady

(2013) The illusion of language acquisition. Linguistic Approaches to Bilingualism 3: 253–85.

28.

Rayner

(1998) Eye movements in reading and information processing: 20 years of research. Psychological Bulletin 124: 372–422.

29.

Roberts

Siyanova-Chanturia

(2013) Using eye-tracking to investigate topics in L2 acquisition and L2 sentence and discourse processing. Studies in Second Language Acquisition 35: 213–35.

30.

Sabourin

Stowe

L A

(2008) Second language processing: When are L1 and L2 processed similarly. Second Language Research 24: 397–430.

31.

Sagarra

Herschensohn

(2011) Proficiency and animacy effects on L2 gender agreement processes during comprehension. Language Learning 61: 80–116.

32.

Siyanova-Chanturia

Conklin

Schmitt

(2011) Adding more fuel to the fire: An eye-tracking study of idiom processing by native and non-native speakers. Second Language Research 27: 251–72.

33.

Sulkala

Karjalainen

(1992) Finnish. New York: Routledge.

34.

Taraban

Kempe

(1999) Gender processing in native and non-native Russian speakers. Applied Psycholinguistics 20: 119–48.

35.

Trenkic

Mirkovic

Altmann

(2014) Real-time grammar processing by native and non-native speakers: Constructions unique to the second language Bilingualism: Language and Cognition 17: 237–57.

36.

Vainio

Bertram

Pajunen

. (2011) Processing modifier–head agreement in long Finnish words: Evidence from eye movements Acta Linguistica Hungarica 58: 134–56.

37.

Vainio

Hyönä

Pajunen

(2003) Facilitatory and inhibitory effects of grammatical agreement: Evidence from readers’ eye fixation patterns. Brain and Language 85: 97–202.

38.

Vainio

Hyönä

Pajunen

(2008) Processing modifier–head agreement in reading: Evidence for a delayed effect of agreement. Memory and Cognition 36: 329–40.

39.

Vainio

Pajunen

Hyönä

(2014) L1 and L2 word recognition in Finnish: Examining L1 effects on L2 processing of morphological complexity and morphophonological transparency. Studies in Second Language Acquisition 36: 133–62.

40.

Virtanen

Pajunen

(2000) ContextMill [computer software]. Turku: General Linguistics, University of Turku.

41.

White

(1997) Suomen kielioppia ulkomaalaisille [Finnish grammar for foreign language learners]. Helsinki: Finn Lectura.