Edited by:
Reviewed by:
*Correspondence:
This article was submitted to Language Sciences, a section of the journal Frontiers in Psychology
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
The current study used event-related brain potentials (ERPs) and behavioral measures to examine effects of genre awareness on sentence processing and evaluation. We hypothesized that genre awareness modulates effects of genre-typical manipulations. We manipulated instructions between participants, either specifying a genre (poetry) or not (neutral). Sentences contained genre-typical variations of semantic congruency (congruent/incongruent) and morpho-phonological features (archaic/contemporary inflections). Offline ratings of meaningfulness (
We use language to express ourselves, to refer to the external world, and to appeal to others (
Poetry is particularly interesting in this respect: as the writing process is not subject to the same practical constraints as daily communication, poets are free to create texts that satisfy a range of additional self-imposed formal constraints on the selection and combination of linguistic elements (e.g.,
Semantics in poetry enjoys similar licenses. Poets assert that the language of poetry is “vitally metaphorical” (
As a result, poetic discourse differs significantly from everyday verbal behavior (e.g.,
But what is specific about poetry?
To address this question, the present study contrasted the processing of prosodic variation in identical sentence contexts with and without explicit genre attribution. We manipulated sentence prosody by means of a slightly archaic case suffix whose optional presence or absence constitutes one of German poetry’s routine licenses for balancing metrical structure. In addition, we manipulated the semantic congruency of the sentences to test whether the poetry mode of reading immediately affects the computation of meaning or whether semantic effects are to be found only at later stages of comprehension.
As previous studies on genre-specific reading have typically used global measures such as reading rate (
Note that while we are testing for genre-dependent adaptations that are
A secondary goal of Experiment 2 was to investigate potential effects of genre awareness on memory encoding by providing behavioral evidence from a probe recognition task: at the end of each trial participants were presented with a word, and indicated per button press whether or not it had occurred in the preceding sentence. Not only did this task ensure attentive reading, it also yielded information about the surface structure representation of the sentence after sentence wrap-up.
In previous research, enhanced memory of a text’s surface structure has been reported in delayed recognition and recall tasks for literary vs. non-literary texts (
These questions were addressed with a mixed design: instruction was manipulated between participants, but sentence prosody and semantic congruency were crossed within participants. Participants were assigned to one of two instruction groups: Instruction was either
Example materials.
Instruction | Condition |
Example sentence |
---|---|---|
Neutral ( |
SchwaDAT/CongruentSEM | [In |
SchwaDAT/IncongruentSEM | [In |
|
ZeroDAT/CongruentSEM | [In |
|
ZeroDAT/IncongruentSEM | [In |
We constructed 48 sets of sentences exemplified in
This configuration allowed us to use the initial PP as a prosodic prime that established an alternating rhythm pattern (trochaic or iambic)
Consistent with the practice of poetic license in lyrical poetry, we used historical vs. contemporary case suffixes in sentence-initial constituents to manipulate word prosody without altering syntax or semantics. This morpho-phonological manipulation exploited a peculiarity of German grammar: the so-called
While functionally equivalent regarding grammar, these suffixes differ in terms of their effect on word prosodic structure: ZeroDAT marking does not affect the monosyllabic nouns we used, but SchwaDAT marking adds a reduced syllable nucleus and results in trochaic word forms. This difference in word prosody motivates the seemingly free variation of these forms in German lyrical poetry, and it has been identified as one of the factors that co-determine their variation even in contemporary standard German (
Comparison of the two dative suffixes.
Example | Stylistically marked | Number of syllables | Syllable structure | Stress on final syllable | |
---|---|---|---|---|---|
ZeroDAT | Tod (deathDAT) | No | 1 | CVC | Yes |
SchwaDAT | Tode (deathDAT) | Yes | 2 | CV.CV | No |
We used a variety of prepositions and modifiers to vary the form of sentence-initial PPs while keeping their basic constituent structure identical: [preposition – [modifierDAT – nounDAT]
Despite their length, we decided to present entire PPs at a time so they could serve as efficient prosodic primes. This presentation mode allows readers to immediately assign phrasal accent to the modified nouns in the PP, including monosyllables in the ZeroDAT conditions. Thus, we ensured that modified nouns were correctly stressed during silent articulation. We refrained from making predictions for the PP as we suspected that inference and interpretation might be complicated by two factors: (a) by the unusual three-word presentation mode that only allows extracting ERPs time-locked to the onset of the whole phrase (and not to the dative-marked noun alone), and (b) by potential temporal jitter, component overlap and component smearing that might result from inter-item differences in terms of prosodic complexity (3–6 syllables), as well as word classes (determiners vs. adjectives) and resulting conceptual complexity (one vs. two content words).
Verbs were identical across conditions. Critical verbs were monosyllabic and unambiguously marked for agreement with a third person singular subject. In line with the Principle of Rhythmic Alternation, dative case marking in the preceding constituent resulted in distinct prosodic preferences for the verb position. In ZeroDAT conditions, an accented monosyllabic noun precedes the verb and an unstressed syllable is preferred at the verb position. In SchwaDAT conditions, a trochaic noun precedes the verb and results in a preference for a prominent syllable. Thus, monosyllabic verbs occurred in either a metrically prominent position (SchwaDAT) or in a non-prominent one (ZeroDAT). As the polysyllabic word forms of these verbs (e.g., infinitives) are initially stressed, we hypothesized that lexical access would be facilitated if the monosyllabic forms occurred in a position expected to be prosodically prominent, i.e., in SchwaDAT trials.
Subject NPs consisted of a determiner and a noun; determiners were marked for nominative case and indicated subjecthood. The event participant roles assigned by the verbs (e.g.,
The critical materials were pretested in a pen-and-paper rating study. Our aims were (1) to ascertain that our sentence variants indeed differed in terms of semantic congruency, and to test for potential effects of (2a) sentence prosody (=Dative) and (2b) genre awareness (=Instruction) on offline evaluation of meaningfulness. We tested the full set of critical stimuli. Note, however, that the sentences did not contain the postcritical adverbial we inserted for the EEG study based on methodological considerations. This ensured that observed differences in meaningfulness were most likely due to the semantic fit between verb and subject. Items were distributed over four lists so that each participant was presented with 12 sentences per condition and saw only one variant from each item set; four constrained randomizations were prepared for each of these lists. In addition to the resulting 48 critical sentences, each list contained 10 control items (5 highly implausible, 5 perfectly plausible) intended to represent the extreme ends of the scale, and enabling us to assess whether participants had understood the task.
Concerning aim (1), we expected a clear effect of semantic congruency with lower average meaningfulness ratings for IncongruentSEM vs. CongruentSEM sentences. Regarding aim (2a), we expected an effect of Dative on semantic evaluation only if phonological processing fluency is misattributed to perceived meaningfulness (
A total of 128 members of the University of Frankfurt community were recruited on campus, all of them adult native speakers of German according to self-report (acquisition started no later than age 3). They were given a campus cafeteria coffee voucher as compensation for participation in the study, which took about 10–15 min. No further personal information was collected. Participants received one of two instructions, asking them to rate either “sentences” (NeutralINST) or “verses” (PoetryINST) on a 6-point rating scale ranging from 1 =
English translations of the instruction texts for the Neutral and the Poetry group in the rating study.
Neutral instruction | Poetry instruction |
---|---|
Thank you very much for your participation. | Thank you very much for your participation. |
This study is part of a research project on |
This study is part of a research project on |
In this questionnaire you are going to read a number of single |
In this questionnaire you are going to read a number of single |
Each of these expresses a situation or state of affairs. | Each of these expresses a situation or state of affairs. |
Please read these |
Please read these |
Simply mark the box that, in your opinion, best locates the |
Simply mark the box that, in your opinion, best locates the |
Please make sure to rate all of the |
Please make sure to rate all of the |
We tested for main effects and interactions of Instruction, Dative, and Semantic congruency on ratings using linear mixed effects regression with crossed random effects for participants and items (
Results of the meaningfulness rating are summarized in
Results of the meaningfulness rating.
ZeroDAT |
SchwaDAT |
|||
---|---|---|---|---|
CongruentSEM | IncongruentSEM | CongruentSEM | IncongruentSEM | |
NeutralINST | 5.25 (0.65) | 3.20 (0.74) | 5.23 (0.73) | 3.21 (0.81) |
PoetryINST | 5.41 (0.51) | 3.62 (0.70) | 5.46 (0.49) | 3.54 (0.75) |
These simple effects were dissimilar within factor levels, as indicated by the significant interaction of Instruction and Semantic congruency [
Mean ratings of meaningfulness for IncongruentSEM and CongruentSEM sentence variants by Instruction (NeutralINST vs. PoetryINST;
Supplementing the findings of
We expected to find modulations of the N400, an ERP component that has traditionally been linked to semantic processing (see
Crucially, amplitude and topography of the N400 – and to a lesser degree also its temporal characteristics – are sensitive to phonological processes that facilitate (or delay) lexical access, such as onset priming, rhyme priming, and prosodic priming. This has been demonstrated in auditory word, sentence, and text processing (e.g.,
Thus, we expected that strict rhythmical alternation (SchwaDAT) would allow predicting upcoming stress locations and thus facilitate lexico-semantic processing and result in reduced N400 responses (compared to ZeroDAT) at the verb following the morpho-phonological manipulation.
Importantly, the N400 is also sensitive to task demands and varies as a function of the depth of processing (e.g.,
Our predictions concerning the probe task were as follows: if the processing system adapts attention allocation and memory encoding to the input category, i.e., the genre, then a memory effect should already be detectable in immediate recognition, leading to enhanced probe task performance for the poetry group (PoetryINST > NeutralINST). If, on the other hand, we find no differences between the groups in terms of probe task performance (PoetryINST = NeutralINST), this would support (but could not dissociate) the explanations that the memory effects observed in the delayed tasks of previous studies reflect slower reading and thus longer encoding (
Forty-eight monolingually raised native speakers of German (acquisition started no later than age 3) were recruited from the University of Mainz community and were paid 7 EUR per hour for participation in the experiment. All participants reported normal or corrected-to-normal vision and no known neurological or reading disorders, and all were right-handed as assessed with an abridged German version of the Edinburgh handedness inventory (
The 192 critical sentences resulting from the 48 item sets were distributed over two lists, each containing 96 critical sentences (2 chosen from each item set to total 24 sentences per condition) and 48 filler sentences
Participants received written instructions on a sheet of paper prior to the experiment; the experimenter ensured that instructions had been read and understood. Translations of the instruction texts are provided in
English translations of the instruction texts for the Neutral and the Poetry group in the ERP study.
Neutral instruction | Poetry instruction |
---|---|
Thank you very much for your participation in our |
Thank you very much for your participation in our |
You are going to read a number of single |
You are going to read a number of single |
Please read these |
Please read these |
Before each |
Before each |
After reading a |
After reading a |
Your task is to decide whether this word had occurred in the |
Your task is to decide whether this word had occurred in the |
Experimental sessions were conducted in a dimly lit and sound-attenuated room. Participants were seated comfortably at a distance of about 100 cm from a 17″ computer screen. Following a short practice session, the experiment consisted of four blocks of 36 sentences/verses each (∼8 min per block). Participants were encouraged to take short breaks (∼1–2 min) between blocks to relax and rest their eyes, as we had asked them to avoid movements and eye blinks during sentence presentation.
Sentences were presented constituent-wise in a 29-point font in the center of the display. Each trial began with the presentation of a fixation asterisk for 1000 ms, followed by a 200 ms inter-stimulus interval (ISI). Presentation rates were adapted to constituent size (e.g.,
Schematic trial structure (
Pre-stimulus |
Sentence presentation |
Probe task | |||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
∗ | PP | V | NP | ADV | ? | probe | |||||||
1000 | 200 | 450 | 200 | 300 | 200 | 400 | 200 | 300–450 | 500 | 300 | 200 | <5000 | 1500 |
The EEG was recorded from 26 Ag/AgCl electrodes (ground: AFZ) fixed at the scalp by means of an elastic cap (Easycap GmbH, Herrsching, Germany). Four additional electrodes monitored the electro-oculogram (EOG) at the outer canthus of each eye (horizontal EOG) and above and below the participant’s right eye (vertical EOG). EEG and EOG channels were amplified by means of a BrainAmp amplifier (Brain Products, Gilching, Germany) and digitized with a sampling rate of 500 Hz. Recordings were referenced to the right mastoid but rereferenced to linked mastoids offline. Electrode impedance was kept below 5 kΩ during the experiment. We applied a 0.3–20 Hz band-pass filter to the raw EEG data offline in order to eliminate slow signal drifts and high frequency noise. The filter setting avoids potential stimulus-independent differences between compared conditions without performing a baseline correction which is susceptible to carrying over transient differences in the baseline period into the critical region (for a motivation of this approach, cf.
For the statistical analyses of behavioral and ERP data, we tested for fixed main effects and interaction effects of Instruction, Dative and Semantic congruency using linear mixed-effects regression. For the ERP data of the verb position, where semantic information was not yet available, we only tested for fixed main effects and interaction effects of Instruction and Dative. Models contained crossed random effects for participants and items (
All analyses were carried out in
We removed outliers exceeding 2.5
We tested for main effects and interaction effects of Instruction, Dative and Semantic congruency on response accuracy (correct/incorrect) using logistic regression.
ERP data analysis was carried out by fitting separate models for midline (FZ, FCZ, CZ, CPZ, PZ) and lateral electrodes (F3, F7, FC1, FC5, CP1, CP5, P3, P7, F4, F8, FC2, FC6, CP2, CP6, P4, P8). Outlying values exceeding ±20 microvolts were excluded from the analysis (∼1% data loss). For the lateral electrodes, four regions of interest were defined via combinations of the spatial factors Hemisphere (left/right) and Region (anterior/posterior). Contrast-coded spatial factors were included in the fixed effects term of the respective models, allowing for all interactions. Note that we do not report main effects and/or two-way interactions of spatial factors.
The behavioral performance suggests that participants generally processed the sentences/verses attentively: mean accuracy was 96.05%; mean response latency was 966 ms (1042 ms before outlier removal). A summary of the behavioral results is provided in
Behavioral results.
ZeroDAT |
SchwaDAT |
|||||
---|---|---|---|---|---|---|
CongruentSEM | IncongruentSEM | CongruentSEM | IncongruentSEM | |||
NeutralINST | 1003 (192) | 976 (176) | 971 (174) | 1017 (194) | ||
( |
1.57 (2.41) | 2.78 (2.92) | 5.90 (5.20) | 5.22 (5.26) | ||
PoetryINST | 953 (224) | 937 (192) | 945 (202) | 961 (197) | ||
( |
2.78 (3.82) | 3.32 (3.26) | 3.82 (3.23) | 4.69 (5.11) |
Reaction times were on average 44 ms shorter for the PoetryINST group (
Behavioral results: Mean reaction time differences (IncongruentSEM – CongruentSEM) in ms for the dative conditions (left panel), and mean error rates for the dative conditions by instruction (right panel). Error bars indicate 95% confidence intervals; asterisks indicate statistical significance of planned comparisons: n.s., not significant; (∗)
There was a main effect of Dative on response accuracy: error rates were higher in the SchwaDAT conditions than in the ZeroDAT conditions [
Grand-average ERP waveforms at the verb. Canonical ZeroDAT conditions show an increased negativity (350–600 ms) compared to non-canonical SchwaDAT conditions.
Congruency conditions diverged around 400 ms after NP onset and showed a bilateral posterior negativity (N400) for incongruent vs. congruent NPs (see
Grand-average ERP waveforms at the subject NP. IncongruentSEM conditions show an increased bilateral posterior negativity (400–600 ms) compared to CongruentSEM conditions.
Instruction-wise comparison of semantic effects (IncongruentSEM–CongruentSEM) at the subject NP. Difference waves (left column) and difference maps illustrate semantic congruency effects for the NeutralINST (black lines; upper row of maps) and for the PoetryINST (red lines; lower row of maps) conditions. Both conditions show a posterior negativity (N400) between 400 and 600 ms. The late positive response (P600) occurs earlier in the NeutralINST condition (600–700 ms) than in the PoetryINST condition (750–900 ms).
Laterally, this effect differed between Regions [
At lateral electrodes, we found an interaction effect of Instruction and Semantic congruency [
Focusing on poetry and poetry-typical linguistic features, the present study aimed to investigate the effects of genre awareness on the processing and evaluation of isolated sentences. Differential instruction of participant groups allowed us to dissociate top–down-controlled genre effects from input-driven automatic processes.
In an offline rating study (Experiment 1), we validated our semantic manipulation and tested for effects of genre awareness on the perceived meaningfulness of the stimuli. Poetry categorization resulted in increased ratings of meaningfulness for semantically incongruent sentences.
In an ERP study (Experiment 2), we tested whether similar effects are detectable already in online processing. We expected (1) a prosodic N400-like effect at the verb due to facilitated lexico-semantic processing resulting from prosodic expectations built up during the processing of the sentence-initial constituent, and (2) a semantic N400 effect at the subject NP indexing the processing of incongruent meaning. We further hypothesized that (3a) the prosodic effect would be reinforced and that (3b) the semantic one would be attenuated by the poetry instruction. The results supported hypotheses (1) and (2) while hypotheses (3a) and (3b) were disconfirmed by the ERP data. The expected interaction of genre awareness and semantic congruency (3b) was reflected in the latency of the P600 response to incongruent vs. congruent NPs rather than in top–down modulations of N400 amplitudes. The interaction of genre awareness and morpho-phonological alternation (3a) was found in behavioral measures: rather than increasing facilitation, poetry instruction neutralized the adverse effect of stylistically marked case suffixes on response accuracy.
In the probe task, we investigated whether poetry categorization affects surface structure memory immediately after sentence wrap-up. Such an effect would provide strong evidence for the hypothesis that genre awareness modulates the level of processing and thus the resulting memory traces, which, in turn, could (partially) account for the enhanced surface structure recall and recognition reported for poetry vs. other text types. Results did not show the predicted general advantage for the poetry group and thus failed to support the level-of-processing account, at least in its strongest, aprioristic version. Regardless of the instruction, response latencies were delayed after processing sentences with combined morphological and semantic deviations. Response accuracy differed between instruction groups, reflecting that error rates increased with the presence of slightly archaic case markers when no genre had been specified, but not when sentences had been identified as lines of poetry. Most likely, this effect does not result from a modulation of the level of processing (which predicts a main effect of genre awareness), but merely reflects that stylistically marked case suffixes are more frequent and expectable in poetry than in most other registers. The results thus suggest that previously reported genre effects on surface structure memory are due to textual variables that were absent in the current study. First, the graphic layout of poetic texts seems to enhance the retrieval of surface structure representations of read poetry, possibly by providing spatial retrieval cues (
As pointed out in the introduction, historical inflections provide poets with creative options for balancing the metrical structure of a line. To contemporary readers, however, they constitute morpho-phonological oddities, as indexed by the decline in subsequent probe task performance when no register/genre information had been provided. It seems that the sentence-initial presence of these infrequent word forms interferes with the behavioral task, possibly by introducing additional uncertainty concerning the exact form of the probe word. Importantly, this penalty diminished for participants in the poetry group. To us, this suggests that the poetry readers were better equipped to deal with poetry-typical, i.e., register-dependent, grammatical deviations while performing the task than were readers without previous text type information. Most likely, the genre instruction cued and activated the grammatical register associated with 18th/19th century lyrical poetry which is arguably the most canonical and prototypical representative of the German poetic tradition.
Notably, the morpho-phonological manipulation that impaired behavioral performance in one of the groups facilitated prosodic processing of the following word. In line with our predictions, we observed a reduced negativity in response to verbs following non-canonically case-marked NPs, resulting in a left-lateralized N400-like effect. It is unlikely that the observed effect is an instance of a semantic N400 that could be accounted for by semantic unification or contextual fit (e.g.,
Therefore we assume that the observed negativity indeed indexes the relative ease of lexico-prosodic processing, i.e., of accessing and integrating a lexeme whose word prosodic template matches rhythm-based predictions (
The results do not support the idea that genre expectations affect prosodic processing in reading immediately and independent of textual variables. Previously reported effects of meter on poetry processing thus most likely reflect input-driven predictions based on phonological recurrence across lines. This interpretation is still compatible with the view that poetry reading affects top–down control of attention to phonological form. For the poetry group, our out-of-context lines thus correspond to the first line of a poem, where no metrical pattern has been established yet. While readers might expect a metrical pattern to be present, it is unclear, at this stage, which one it will turn out to be. Readers most likely extract the systematic sound patterning instantiated in a text incrementally when reading or hearing the first few lines or stanzas of a poem. This interpretation is in line with
Taken together, these results suggest that the same deviant morphological feature may impair and facilitate distinct stages of stimulus processing. This constitutes initial evidence from comprehension in support of the assumed trade-off between morphological deviation and prosodic regularity in poetic language (
Offline sentence evaluation confirmed the semantic congruency difference between sentence variants, and revealed a reliable interaction of instruction and semantics: genre categorization increased ratings of meaningfulness for incongruent sentences but not for congruent ones. These findings refine and extend the intention-based meaningfulness effect for metaphors (
Confirmed in ratings of meaningfulness, the semantic congruency difference also affected online sentence processing. In the ERPs, we observed a bilateral posterior N400 in response to semantically incongruent vs. congruent NPs. While the effect showed some characteristics that contrast with the “standard N400 effect,” i.e., a broadly distributed centro-parietal negativity between 250–550 ms (
As the N400 is highly sensitive to cloze probability (
While the semantic N400 effect was unaffected by genre categorization, the effect of semantic congruency on the P600 differed between genre conditions in terms of latency and scalp distribution, suggesting both temporal and functional differentiation in controlled conflict resolution. Indeed, it seems tempting to conclude that the P600 in the neutral group is a marker of well-formedness evaluation grounded in the default expectation of semantic congruency, and that the P600 in the poetry group reflects pragmatic processing triggered by the knowledge that superficially incongruent sentences may in fact be highly significant in poetic contexts. This explanation is in line with the idea that literary genres are conventionally associated with genre-specific interpretive strategies (
Our results disfavor the notion of aprioristic genre-specific adaptation of prosodically and semantically mediated lexical access if reading time is controlled for. It remains an open question, though, whether genre affects the predictability component of the N400. Regarding the semantic manipulation, our findings indicate that literary and non-literary real-time comprehension differ in the controlled evaluation of verbal stimuli although the exact nature of this difference needs to be addressed in future research. Regarding the morpho-phonological manipulation, they support the view that the subtle metrical differences induced by morphological licenses in poetry effectively modulate readers’ implicit prosodic expectations, and that genre awareness moderates the adverse effects of these poetic licenses.
SB, WM, and MS developed the hypotheses. SB and MS developed the study design. WM provided the funding. SB developed the materials, collected the data and performed the data analysis. SB, WM, and MS interpreted the data. SB drafted the manuscript, and MS and WM provided critical revisions. SB, WM, and MS approved the final version of the manuscript.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
We thank Christine A. Knoop, Maria Kraxenberger, R. Muralikrishnan, Mathias Scharinger, and Valentin Wagner for helpful comments and discussions. We are grateful to Barbara Budai, Sanja Methner, and Brita Rietdorf for help in data collection, and to Felix Bernoully for graphics support.
The Supplementary Material for this article can be found online at:
Dative case prototypically marks recipients in transfer-of-possession events, as in: ‘
Trochaic = stressed-unstressed; iambic = unstressed-stressed.
Lexical frequency was operationalized as the number of occurrences in
We used the term ‘
Fillers were identical to the critical materials in terms of syntactic structure. They also contained implausible (50%) and metrically regular (50%) sentences. Contrary to the critical materials, fillers showed a greater variety of semantic and metrical variation. For instance, in filler sentences, implausibility sometimes arose PP-internally from semantic mismatch between adjective and noun (instead of semantic fit between verb and subject). Metrical variation included ternary meters (e.g., dactyls) and polysyllabic verbs with different word stress patterns.
To assess whether the effects differed between the first and second encounter of an item, we conducted additional analyses, fitting models that contained main and interaction effects of Block (1st half/2nd half of the experiment). Results confirmed the reported data pattern, and did not reveal additional repetition-sensitive effects involving the variables of interest. However, we found a repetition effect at midline [main effect Block: