Stress in Context: Morpho-Syntactic Properties Affect Lexical Stress Assignment in Reading Aloud

Spinelli, Giacomo; Sulpizio, Simone; Primativo, Silvia; Burani, Cristina

doi:10.3389/fpsyg.2016.00942

ORIGINAL RESEARCH article

Front. Psychol., 22 June 2016

Sec. Psychology of Language

Volume 7 - 2016 | https://doi.org/10.3389/fpsyg.2016.00942

This article is part of the Research Topic Prosody in written language – acquisition, processing, emergence of norms View all 14 articles

Stress in Context: Morpho-Syntactic Properties Affect Lexical Stress Assignment in Reading Aloud

$\r\nGiacomo Spinelli*$ Giacomo Spinelli^1*

Simone Sulpizio^2,3

Silvia Primativo⁴

Cristina Burani^5,6

¹Department of Psychology, University of Western Ontario, London, ON, Canada
²Department of Psychology and Cognitive Science, University of Trento, Trento, Italy
³Fondazione Marica De Vincenzi ONLUS, Trento, Italy
⁴Dementia Research Center, Department of Neurodegenerative Disease, University College London, London, UK
⁵Institute of Cognitive Sciences and Technologies (ISTC-CNR), Rome, Italy
⁶Department of Life Sciences, University of Trieste, Trieste, Italy

Recent findings from English and Russian have shown that grammatical category plays a key role in stress assignment. In these languages, some grammatical categories have a typical stress pattern and this information is used by readers. However, whether readers are sensitive to smaller distributional differences and other morpho-syntactic properties (e.g., gender, number, person) remains unclear. We addressed this issue in word and non-word reading in Italian, a language in which: (1) nouns and verbs differ in the proportion of words with a dominant stress pattern; (2) information specified by words sharing morpho-syntactic properties may contrast with other sources of information, such as stress neighborhood. Both aspects were addressed in two experiments in which context words were used to induce the desired morpho-syntactic properties. Experiment 1 showed that the relatively different proportions of stress patterns between grammatical categories do not affect stress processing in word reading. In contrast, Experiment 2 showed that information specified by words sharing morpho-syntactic properties outweighs stress neighborhood in non-word reading. Thus, while general information specified by grammatical categories may not be used by Italian readers, stress neighbors with morpho-syntactic properties congruent with those of the target stimulus have a primary role in stress assignment. These results underscore the importance of expanding investigations of stress assignment beyond single words, as current models of single-word reading seem unable to account for our results.

Introduction

Recent years have seen a growing interest in how readers assign lexical stress to both known and novel words. This is especially true for free-stress languages such as English or Italian, where stress is neither orthographically marked nor predictable based on rules. Among the factors found to affect stress assignment, three received special attention: lexical retrieval (i.e., the stress pattern stored within each word’s entry in the mental lexicon: Colombo, 1992; Rastle and Coltheart, 2000), stress dominance (i.e., the most frequent stress pattern in a language: Colombo, 1992; Rastle and Coltheart, 2000) and stress neighborhood, that is the proportion of words sharing the stress pattern and orthographic final sequence (e.g., most English disyllables ending in -et bear first-syllable stress: Burani and Arduino, 2004; Arciuli and Cupples, 2006; Ševa et al., 2009; Arciuli et al., 2010; Sulpizio et al., 2013; Burani et al., 2014; Jouravlev and Lupker, 2014). Recently, a few studies investigating reading in English and Russian (Arciuli and Cupples, 2006; Jouravlev and Lupker, 2014) also reported an effect of stress typicality (i.e., the dominant pattern within a grammatical category). They found that grammatical categories not only play a role in stress assignment, but also interact with stress neighborhood. For instance, in Russian no dominant stress pattern is distinguishable overall, however, adjectives show a clear preference toward trochaic stress. Using both a reading and a lexical decision task, Jouravlev and Lupker (2014) found facilitation for typically stressed Russian adjectives (i.e., trochaic). Conversely, atypically stressed adjectives (i.e., iambic) were disadvantaged, especially when stress neighborhood was inconsistent.

However, while reading, morpho-syntactic properties – such as gender, number, and person – other than grammatical category are active, and whether they play any role in the process of stress assignment is currently unknown. We used Italian to investigate such issue.

In Italian, the position of lexical stress is predictable from either orthography or phonology only in a small number of cases: when the last syllable is marked with a diacritic, it bears stress (orthographic rule, which occurs in ∼2% of words, e.g., poverTÀ ‘poverty’)¹; when there is no diacritic on the last syllable and the penultimate syllable ends with a consonant, the latter bears stress (phonological rule, e.g., conCERto ‘concert’)². For the most part of the Italian lexicon, however, the latter rule does not apply. Thus, readers mainly use lexical retrieval and distributional information such as stress dominance and stress neighborhood in order to assign stress. Note that unlike Russian, Italian shows an asymmetrical distribution of stress patterns overall, with about 80% of words bearing penultimate stress (e.g., teSOro ‘treasure’) and 18% bearing antepenultimate stress (e.g., Anima ‘soul’: Thornton et al., 1997). Accordingly, Colombo (1992) argued for a primary role of stress dominance in Italian, with penultimate stress working as a default. More recent studies, however, succeeded in finding an effect of stress dominance only in populations with little or reduced ability to detect more specific distributional information in the lexicon. Examples of such populations are second graders, children with dyslexia and adults with acquired language disturbances (for a review, see Sulpizio et al., 2015). On the other hand, stress neighborhood, a more specific source of distributional information, has been reported as the strongest factor driving stress processing and stress assignment in both low-frequency words and non-words for unimpaired adults and older children (Sulpizio et al., 2013; Burani et al., 2014), at least in reading aloud (Burani and Arduino, 2004; Colombo and Sulpizio, 2015).

While the literature focused on the relative role of stress dominance and stress neighborhood in Italian readers, no attention was paid to the role of morpho-syntactic properties such as grammatical category, number, gender, and person, in stress processing. These properties are known to affect word processing in different tasks and languages (Bates et al., 1996; Federmeier et al., 2000; Bentrovato et al., 2003; Monaghan et al., 2005; Vigliocco et al., 2008; Arciuli et al., 2012; see also Lalović, 2010).

The distribution of stress patterns with respect to morpho-syntactic properties raises two noteworthy questions. First, do Italian readers, like English speaking and Russian ones, use distributional information specified by grammatical category when assigning stress? While all grammatical categories in Italian show a prevalence of penultimate stress words, there are some outstanding differences between them, with adverbs and nouns showing higher proportions of dominant stress words than adjectives and verbs (95.3; 84.9; 76.2, and 70.3%, respectively). It may be the case that readers are sensitive to those differences. Second, how specific is stress neighborhood information? Note that stress neighborhood is defined as the number of words sharing the same final sequence (i.e., the last three graphemes in Italian) and stress pattern. For instance, stress neighborhood for the final sequence -era is prevalently penultimate, as most words ending in -era have penultimate stress (e.g., panTEra ‘panther’). However, words with the same final sequence can have different stress patterns depending on their morpho-syntactic properties. In fact, third person verbs ending in -era mostly bear antepenultimate, rather than penultimate, stress (e.g., MOdera ‘he/she moderates’). Thus, in reading, morpho-syntactic properties may interact with stress neighborhood in specifying a stress pattern. Here, we address those issues by presenting data coming from both word and non-word reading.

Experiment 1 investigated the role played by grammatical category in reading words aloud. Studies addressing stress processing in reading aloud showed that there is no advantage for penultimate stress words (Burani and Arduino, 2004; Burani et al., 2014), although most Italian words bear penultimate stress. However, since nouns were generally used and verbs show a lower proportion of penultimate stress words than nouns, we might expect this difference to be reflected in readers’ performance, especially when morpho-syntactic processing is enhanced. We were also interested in the role of morpho-syntactic properties in stress assignment to novel words. Thus, Experiment 2 turned to non-word reading aloud. Non-words were constructed in such a way that information coming from stress neighborhood and information specified by morpho-syntactic properties (i.e., grammatical category, gender, number, or person) contrasted with each other. Our question was whether morpho-syntactic properties could have a stronger influence than stress neighborhood in driving participants’ responses.

Experiment 1

In this experiment, we aimed to assess the extent to which Italian readers use grammatical category as a source of distributional information in order to assign stress. In Italian, stress patterns are not distributed equally within grammatical categories. This fact might have been underestimated so far because previous analyses (Thornton et al., 1997) were based on lemmas (i.e., the canonical or citation form of a word) rather than word forms (i.e., any inflected form of a word). However, in Italian, counts made on lemmas may differ dramatically from word forms. This is especially true for verbs, which show different inflectional markers for each of the 12 simple tenses, each of the three conjugations, and each person. Most importantly, they show alternating stress patterns and number of syllables within inflectional paradigms. Thus, we analyzed word forms in phonItalia (Goslin et al., 2014), a lexical database providing orthographic and phonological information for 120,000 word forms extracted from CoLFIS (Bertinetto et al., 2005). The results showed that, although penultimate stress is the dominant pattern for each grammatical category, verbs have a lower proportion of penultimate stress word forms as compared to nouns (70.3 vs. 84.9%).

That proportion for verbs may be even lower depending on what one considers to be the relevant source of distributional information. In languages such as Italian and English, morphology is known to affect the position of stress, as suffixes are either stressed or unstressed (Rastle and Coltheart, 2000). In Italian, however, the first, second, third, and sixth persons in present indicative and subjunctive may have either ultimate, penultimate, antepenultimate, or even pre-antepenultimate stress (sixth person only, e.g., GIUdicano ‘[they] judge’). In those verb forms, the position of stress is unpredictable, provided that the penultimate syllable does not end with a consonant. Thus, one may want to consider only those forms where stress is not fixed, i.e., governed by neither inflectional suffixes nor stress rules. We selected those verb forms from phonItalia and found that penultimate stress patterns dropped to 60.7%, with antepenultimate stress patterns accounting for 39.3% of the set. Furthermore, four-syllable and five-syllable verb forms turned out to prefer antepenultimate rather than penultimate stress.

In sum, there is evidence that the proportion of penultimate and antepenultimate stress patterns differs between nouns and verbs, with the latter showing a relatively larger proportion of antepenultimate stress words. We wanted to test whether those distributional differences are sufficient to replicate, in Italian, the findings reported by Arciuli and Cupples (2006) for English and Jouravlev and Lupker (2014) for Russian. Note that, although penultimate stress is the dominant pattern for both nouns and verbs, we might expect a different pattern of results for the two grammatical categories if such distributional information is relevant to Italian readers when assigning stress. Besides, differences, if any, might be larger when morpho-syntactic processing is enhanced, i.e., when information concerning the grammatical category of the word is available to readers beforehand.

We therefore designed a reading aloud task where nouns and verbs were presented either isolated or preceded by a congruent context word (i.e., an article in the case of nouns, and a pronoun in the case of verbs). In line with previous studies where stress neighborhood was controlled for, we did not expect processing times differences between penultimate and antepenultimate stress nouns. In contrast, we hypothesized that if Italian readers are sensitive to distributional differences at the level of grammatical categories, then the larger proportion of antepenultimate stress words within verbs may lead to some processing advantage for antepenultimate over penultimate stress for verbs. A larger effect was hypothesized for the with-context as compared to the without-context condition.

Participants

Thirty-two undergraduate students (11 males, mean age: 24.2, SD: 1.9) from Sapienza University of Rome took part in the experiment. They were all native Italian speakers with normal or corrected-to-normal vision. This study was carried out in accordance with the recommendations of the Research Ethics Committee of the Institute of Cognitive Sciences and Technologies in Rome, with written informed consent from all participants in accordance with the Declaration of Helsinki.

Materials

We used low-frequency words in line with the assumption that distributional effects are more likely to emerge with low- rather than high-frequency words (Colombo, 1992). One-hundred and sixty low-frequency words, half nouns and half verbs, were selected from phonItalia (Goslin et al., 2014). All words had a final sequence with a VCV (vowel-consonant-vowel) orthographic structure. For each grammatical category, half of the words had a penultimate stress neighborhood, and half had an antepenultimate stress neighborhood. For each stress neighborhood, half of the words had a penultimate stress pattern, and half had an antepenultimate stress pattern. As a result, half words were consistent and half were inconsistent with their stress neighborhood. In sum, there were eight sets of 20 words each varying in grammatical category, stress neighborhood, and stress pattern. Words were chosen in such a way to cover as many morpho-syntactic combinations as possible. Nouns were either masculine or feminine, and either singular or plural. Verbs included first, second, and third person present indicative. Morpho-syntactic properties were matched across sets of the same grammatical category. Each set included 6–8 different final sequences. Of these, 2–4 final sequences were set-exclusive, and the other 4–5 final sequences appeared in another set. Each shared final sequence appeared half of the time in stress-consistent and half of the time in stress-inconsistent words. For each set, two final sequences were shared within grammatical categories (e.g., buFEre ‘storms,’ MASchere ‘masks’), and 2–3 between grammatical categories (e.g., FORfora ‘dandruff,’ imPLOra ‘[he/she] implores’). All words had three syllables, except for eight four-syllable words which were equally distributed across sets. Since the aim of the study was to compare responses to nouns and verbs with penultimate and antepenultimate stress, only sets differing in stress pattern were matched on initial phoneme and other variables like frequency, length in letters, bigram frequency and orthographic neighborhood (see Table 1). Due to the low-frequency of stimuli, familiarity ratings were collected from an independent group of participants who did not take part in the experiment. Twenty-one university students (12 females, mean age: 24.89, SD: 3.89) were asked to estimate the familiarity of all nouns and verbs on a seven-point scale, following the procedure described by Bates et al. (2001). Familiarity resulted not to be fully matched across the experimental sets of stimuli; thus, such measure was introduced as covariate in the analyses.

TABLE 1

TABLE 1. Mean characteristics of stimuli in Experiment 1.

Eighty filler words were also included. Half were nouns and half verbs, and for each grammatical category, half bore stress on the penultimate and half on the antepenultimate syllable. Filler words differed from experimental items in the following: (1) their stress neighborhoods were ambivalent (see Colombo et al., 2014), and thus, did not provide clear cues for stress; (2) they included a high proportion of four-syllable words (34 out of 80, 42.5%).

Procedure

The list of 240 items was presented in four blocks, each composed of the same number of nouns and verbs, penultimate and antepenultimate stress neighborhoods, and consistent and inconsistent words. Effort was made to limit repetitions of the same final sequences within each block. The presentation order of stimuli within each block was randomized, and block order counterbalanced across participants.

Sentence context was also manipulated. Morpho-syntactic properties may be stored in the lexical entry of each word in the lexicon (e.g., Levelt et al., 1999). Nevertheless, words have been shown to benefit from a morpho-syntactically congruent context in lexical decision (Goodman et al., 1981; Seidenberg et al., 1984; Wright and Garrett, 1984; see also Carello et al., 1988 and works cited) and naming tasks (Bowey, 1996; Bentrovato et al., 2003; Lalović, 2010). Thus, in order to enhance the processing of the morpho-syntactic properties of words, we used a sentence context: words were read in two conditions, namely with context and without context. In the with-context condition, a fixation cross was displayed for 400 ms at the beginning of each trial. Then, the fixation cross disappeared and a context word appeared, with the first letter in the same spatial position of the fixation cross. After 425 ms, the stimulus was displayed next to the context word. Participants were asked to read aloud the stimulus (but not the context word) as quickly and as accurately as possible. Both the context word and the stimulus disappeared at pronunciation or after 1,000 ms since the stimulus onset. There was an interstimulus interval of 2,500 ms. In this condition, both the fixation cross and the context word were displayed slightly to the left of the screen, in order for the stimulus to appear in roughly the same position as in the without-context condition. The context word consisted of either an inflected article (e.g., il ‘the, masculine singular’) or a personal pronoun (e.g., egli ‘he’). All context words were morpho-syntactically congruent with their associated stimuli. The without-context condition differed from the with-context condition in that there was no context word, and both the fixation cross and the stimulus were centered on the screen.

Each participant read each stimulus once in either the with- or without-context condition. Of the four blocks, one pair of consecutive blocks was read in the with-context condition and the other pair was read in the without-context condition. Condition order was counterbalanced across participants. A practice block of 10 trials preceded each pair of blocks. The experiment was run in a quiet room using DMDX software (Forster and Forster, 2003).

Results

Reaction times (RTs) were measured by means of a voice-key, and defined as the time difference between the appearance of the target stimulus on the screen and the onset of participants’ response. Response waveforms were then manually inspected with CheckVocal (Protopapas, 2007) in order to determine the accuracy and placement of timing marks. Invalid trials due to technical failures, as well as responses that exceeded the time limit, accounted for 1.4% of the data points and were discarded from the analyses. Phonemic and stress errors were also discarded, as they accounted for only 3.5% of the data points. Mean RTs are shown in Table 2.

TABLE 2

TABLE 2. Mean latencies (and standard deviations) in ms for correct responses in Experiment 1.

Reaction times were analyzed using linear mixed effects modeling (Baayen, 2008) with crossed random effects for participants and items. The models were fitted with lme4 1.0-5 (Bates et al., 2013), and lmerTest (Kuznetsova et al., 2013) packages implemented in R 3.0.2 (2013–09–25, R Core Team, 2013). Models included near-maximal random structure to the extent allowed by convergence constraints (Barr et al., 2013). Once the models were fitted, outliers were identified and removed (employing 2.5 SD of the residual errors as criterion; Baayen, 2008). Statistics of the refitted models are reported. As materials were designed to compare penultimate and antepenultimate stress words within each grammatical category, verbs and nouns were analyzed separately; for each category, the same analyses with morpho-syntactic context (present vs. absent), word stress (penultimate vs. antepenultimate), and stress neighborhood consistency (consistent vs. inconsistent) as predictors were performed; over and above the variables of interest, familiarity was included as a covariate.

Nouns

Main effects of consistency (β = 23.27, SE = 11.06, t = 2.10, p = 0.03) and familiarity (β = -22.34, SE = 6.20, t = -3.60, p < 0.001) were found. The interaction between morpho-syntactic context and stress neighborhood consistency was significant (β = -21.08, SE = 7.75, t = -2.72, p = 0.006). Moreover, the three-way interaction was also significant (β = 24.84, SE = 11.00, t = 2.25, p = 0.02). No further effect reached significance (for all other effects: |1.15| < ts < |1.66|, ps > 0.1).

The three-way interaction was further investigated by splitting the data for word stress. When considering the antepenultimate-stress words, other than the effect of familiarity (β = -20.39, SE = 6.58, t = -3.09, p = 0.003), only the main effect of morpho-syntactic context was apparent (β = -23.43, SE = 7.66, t = -3.05, p = 0.004). Neither the effect of stress neighborhood consistency nor the stress neighborhood consistency × morpho-syntactic context interaction were significant (both t_s < 1). When considering penultimate-stress words, other than the effect of familiarity (β = -24.04, SE = 12.25, t = -1.96, p = 0.05), no effect reached significance (consistency: t = 1.83, p > 0.05; morpho-syntactic context: t = -1.29, p > 0.2). However, the stress neighborhood consistency × morpho-syntactic context was significant (β = -20.90, SE = 7.66, t = -2.72, p = 0.006), showing that the effect of context was larger for words with inconsistent (β = -31.83, SE = 5.8, t = -5.58, p < 0.001) than with consistent stress neighborhood (β = -10.98, SE = 5.32, t = -2.06, p = 0.03).

Verbs

The interaction between morpho-syntactic context and stress neighborhood consistency was significant (β = -15.54, SE = 8.06, t = -1.92, p = 0.05), with the morpho-syntactic context exerting a larger effect on words with inconsistent (β = -22.44, SE = 7.58, t = -2.95, p = 0.005) than with consistent stress neighborhood (β = -13.12, SE = 6.08, t = -2.15, p = 0.03). Also familiarity reached significance (β = -33.25, SE = 5.31, t = -6.25, p < 0.001). No further effect reached significance (|0.50| < ts < |1.07|, ps > 0.2).

Discussion

Although stress patterns do not distribute equally in grammatical categories in Italian, we found no evidence supporting the claim that readers use this information when assigning stress. In line with previous studies using nouns (Burani and Arduino, 2004; Burani et al., 2014), we reported no significant differences between penultimate and antepenultimate stress nouns when stress neighborhood was controlled, and the use of context did not change such pattern. For the first time, we showed that the same is true for verbs. In Italian, verbs have a higher proportion of antepenultimate stress words than nouns, and this is especially true for the verb forms we used (i.e., first, second, and third person present indicative). Even in this case, however, no difference between penultimate and antepenultimate stress verbs was observed, regardless of the presence of context. Therefore, these results indicate no role for grammatical information in stress processing for adult Italian readers, possibly because in Italian differences between categories are not as large as in languages such as English and Russian.

As expected, we did find an effect of experimental condition: words were read faster in the with- than in the without-context condition. This effect, however, was significant only for inconsistent words, an interesting finding that will be discussed in Section “General Discussion.”

Experiment 2

While adult Italian readers are insensitive to general distributional information (i.e., stress dominance in the language and within a given grammatical category), they rely on specific information (i.e., stress neighborhood) in order to assign stress to both words and non-words (Burani and Arduino, 2004; Sulpizio et al., 2013; Burani et al., 2014). However, previous studies do not say much about whether and how stress neighborhood is grammatically specific. In fact, most of the final sequences that have been used so far show a consistent proportion of either penultimate or antepenultimate stress words across all grammatical categories. For example, -ola has an antepenultimate stress neighborhood, and most feminine singular nouns, feminine singular adjectives, and third person present indicative verbs ending in -ola bear antepenultimate stress. Thus, it is not clear whether words and non-words are assigned stress on the basis of the overall count of their stress neighbors, or on the basis of just those neighbors with which they share morpho-syntactic properties.

Non-words are especially interesting in this respect. Unlike words, processing of morpho-syntactic information for non-words must be induced by using inflectional affixes and/or contextual information. Thus, on the one hand, morpho-syntactic processing may only be partial in reading tasks that involve isolated non-words (i.e., not primed by a context word). On the other hand, the role of morpho-syntactic properties in stress processing is more easily controlled with non-words than with words when a context word is used, because neither morpho-syntactic properties nor stress patterns can be retrieved from the lexicon. That is, all the relevant information to process morpho-syntactic properties and assign stress to a non-word must come from the written stimulus and contextual information. At the same time, it is important to note that non-word and word reading are tightly related. For example, dual-route models of reading assume that each input (word or non-word) activates both lexical and sub-lexical processing, and the outputs of the two routes interact in the phonemic buffer (Coltheart et al., 2001; Perry et al., 2010). Connectionist models show an even stronger picture, as each input the system receives is processed by the same mechanisms (e.g., Plaut, 1999; Harm and Seidenberg, 2004). Indeed, lexical effects on non-word reading have been systematically reported (see, e.g., Coltheart et al., 1977; Laxon et al., 1992; Peereman and Content, 1995), and there is evidence that such effects extend to non-word stress assignment (Arciuli and Cupples, 2006, 2007; Protopapas et al., 2007). In sum, although non-words are markedly different from real words, they can inform our view of the mechanisms involved in stress assignment and reading in general.

We thus used a non-word reading aloud task to assess the role of morpho-syntactic properties in stress assignment. Unlike previous studies, only inconsistent final sequences were used. In those final sequences, the stress pattern specified by stress neighborhood as a whole conflicts with the stress pattern specified by at least one subset of stress neighbors with shared morpho-syntactic properties, i.e., words that have both the same orthographic ending and the same morpho-syntactic properties. Note that shared morpho-syntactic properties include grammatical category as well as gender and number (for nouns), and person (for verbs). For example, -ine has a penultimate stress neighborhood, but masculine singular nouns ending in -ine bear antepenultimate stress most of the time. Our prediction was that, when no context is available, morpho-syntactic processing is reduced and readers assign stress on the basis of the non-words’ stress neighborhood. When context is available, however, morpho-syntactic processing is enhanced, and readers may assign stress basing on the non-words’ morpho-syntactically congruent stress neighbors, even though they conflict with overall stress neighborhood.