Session 5C

L2 knowledge is characterised as noisier representations, resulting in L2ers' reduced capacity to deploy the target knowledge[1] and their learning trajectories distinctive from L1 development.[2,3] Of various factors contributing to this nature, we attend to L2 textbook input and L1 properties under foreign language learning contexts. We ask whether/how these factors jointly affect L2ers' comprehension of Korean subject–predicate honorific agreement, manifesting cross-linguistic consistency (systematic subject–predicate dependency) and language-specific aspects (context-driven optionality; indirect honorification).[4,5] Considering that L2-Korean textbooks over-emphasise honorification,[6,7] we employ two typologically-distinctive SVO-L1s for L2ers: Czech (synthetic; highly-inflectional; active agreement) and English (analytic; little inflection; less-active agreement), with emphasis on L2ers' understanding of two mismatch conditions (Table.1): (b) infelicitous but acceptable contingent upon the speaker's relation to the addressee (honorifiable-subject + no-honorific-suffix) versus (c) ungrammatical (non-honorifiable-subject + honorific-suffix).
28 L1-Czech learners (CZH; Mage=24.1; SD=2.8), 24 L1-English learners (USA; Mage=23.3; SD=4.2), and 40 native speakers of Korean (NSK; Mage=23.6; SD=4.1) participated in an acceptability judgement task (AJT). L2ers' Korean proficiency was measured separately;[8] their scores were above 94 (out of 188) and there was no statistical by-group difference in the scores (W=399, p=.251). Thirty-two sentences (eight instances per condition), together with fillers, were split into four sub-lists and were randomly assigned to participants. We asked participants to rate the acceptability of each sentence as quickly as possible with a 6-point Likert scale (zero: very unacceptable; five: very acceptable). The duration from the first moment when a sentence was presented to the final decision for the acceptability of that sentence (RT) was also collected as a measure of the processing cost during AJT. After data-trimming (Figures.1-2), we fitted the AJT/RT data to the respective linear mixed-effects models in a pairwise (between-group) manner.
Results(Figures.1-2 & Tables.2-3). Overall: all participants rated (a) and (d) very acceptable (above 4.5/5); L2ers spent more time than NSK for all the conditions. [(b)~(c)] Global (α=.05) & post-hoc (α=.025) analyses: NSK–CZH revealed main effects of Group/Condition and interaction in AJT, the locus of which was the higher acceptance of (c) for CZH*** than for NSK. NSK–USA revealed a main effect of Group and interaction in AJT, attributed to the inverse rating trend for (b) and (c) by USA*** compared to NSK. CZH–USA revealed a main effect of Group in AJT: whilst CZH demonstrated uniform acceptability rates for both conditions, USA rated (c)*** more acceptable than (b). In RT, we found a main effect of Group; additional analyses revealed that CZH* spent more time for (b) than USA.
Together, our findings suggest L2ers' manifestation of intended L2 knowledge subject to L2 input characteristics (over-emphasising honorification -> accepting (c) illicitly) and L1 properties (agreement more active in Czech than in English -> CZH more sensitive to the agreement computation for (b), with increased processing cost, than USA). These factors, we argue, play an interactive role for constructing the aforementioned nature of L2 knowledge.

Just as for native speakers, the pronunciation of foreign or second language (L2) learners varies according to different speaking styles [1]. In this study, we present a corpus that serves as a diagnostic basis for an on-going pronunciation training program targeted at Mandarin learners of French and aim to characterize the Mandarin L2 accent in French at the segmental and pitch levels. More than 38000 vowels (excluding voiceless segments) extracted from 6h30' recording composed of isolated word (1h), text reading (4h) as well as spontaneous speech (1h30) of 20 Mandarin learners were analysed. The results will be compared to a parallel corpus of 10 French native speakers from the Parisian region which contains approximately 21000 vowels across the three different speech tasks and is in the process of being analysed. All the data were processed with the help of the LIMSI forced alignment system [2], and the acoustic vowel parameters were extracted with Praat [3]. 

The present contribution focuses on the production of oral vowels. Overall, the female learners showed a general "centralization" of the formants in the acoustic F1-F2 space as vowel duration decreased to around 50ms (see Figure 1.pdf). A one-way ANOVA shows that the differences of F2 values between /u, o, ɔ/ were not significant in the production of text reading and spontaneous speech, but significant in that of isolated words (see Figure 2.pdf). T he mid-vowels /oe, ø/ seemed to be difficult for Mandarin learners in general and were not differentiated in any of the first three formants. When realized in isolated words, the vowel /i/ in the glide context /ɥi/ approached /y/, especially for the segments shorter than 100ms (see Table 1.pdf). The tendency of /ɥi/ being substituted by /y/ is also seen in text reading and spontaneous speech. While the comparison to the native corpus will allow us to determine in detail to what extent the learners' pronunciation is less accurate in connected speech than in the word-isolated context, and at faster speaking rates than at lower ones, the present data suggest that different speaking styles should be included in the segmental pronunciation training program. 
Figures1-2: Mean F1 and F2 values of French oral vowels produced by the Mandarin female learners according to segment duration (Figure 1, high) and across different speaking styles (Figure 2, low)
Table 1: Mean F1, F2 and F3 values of /i/ and /y/ by the Mandarin female learners in the production of isolated words

References
[1] Gut, U. (2009). Non-native speech: A corpus-based analysis of phonological and phonetic properties of L2 English and German (Vol. 9). Peter Lang.
[2] Gauvain, J.L., Lamel, L. and Adda, G. (2002). The Limsi Broadcast News Transcription System, Speech Communication, 37(1-2):89-108. 
[3] Boersma, P. and Weenink, D. (1999). Praat, a system for doing phonetics by computer. Institute of Phonetic Sciences of the University of Amsterdam, report 132-182. 

Rhetorical questions (RQs) are syntactically interrogatives with the pragmatic function of an assertion signalling the speaker's attitude [1]. RQs can be distinguished from information-seeking questions (ISQs) through different cues, including prosodic and lexical-syntactic ones, e.g., discourse particles (DiPs). 
Research on the acquisition of RQs in general and their acquisition in heritage speakers (HSs) in particular is scarce. Research on HSs has mostly focused on their heritage language (HL) rather than their majority language (ML), for which monolingual-like competence is often assumed. Finally, previous research on the phonetic and phonological aspects of HSs has focused predominantly on segmental aspects, while suprasegmental properties, such as intonation, have only recently started to receive attention [2, 3].
This study focuses on the acquisition of RQs in German-dominant adult HSs of Italian and addresses the following questions: 1. Which prosodic and lexical cues do they use when producing RQs in German? 2. Do they use different cues compared to German monolinguals (L1ers)? 
Eighty participants completed an elicitation task that provided RQ- and ISQ-eliciting contexts, together with a recorded model sentence, as in (1).
Wer mag denn schon Bananen?! (RQ prosody)
'Who likes DiP bananas?!'
Preliminary analyses (10 HSs, 10 L1ers) suggest that, phonologically, HSs and L1ers use the same nuclear contour, i.e., nuclear pitch accent and boundary tone (L*+H L-%, low rising accent on the object followed by a low boundary tone) to mark RQs (p = .99) and this contour is seldom used in ISQs (Figure 1. Proportion of nuclear contour across illocution type (ISQ vs. RQ) and group (L1 vs. HS).pdf). This is in line with previous studies on L1ers [4]. Phonetically, the stressed syllable tends to be longer in RQs than in ISQs (p = .06) for both groups. However, HSs produce longer stressed syllables in both illocution types compared to L1ers (p < .05), suggesting that HSs differ from L1ers at the phonetic level.
With the complete data set, we will further explore whether differences can be found also at the phonological and lexical-syntactic level and discuss them in light of cross-linguistic influence from the HL to the ML. More generally, the study adds to the growing evidence that CLI is not unidirectional from ML to HL, but that it can affect the ML as well [5].
References
[1] Biezma, M., & Rawlins, K. 2017. Rhetorical questions: Severing asking from questioning. Proceedings of Semantics and Linguistic Theory, 27, 302-322. 
[2] Dehé, N. 2018. The intonation of polar questions in North American ("heritage") Icelandic. Journal of Germanic Linguistics, 30(3), 213-259. 
[3] Zuban, Y., et al. 2020. Intonation of yes-no questions by heritage speakers of Russian. 10th International Conference on Speech Prosody 2020. 
[4] Braun, B., et al. 2019. The prosody of rhetorical and information-seeking questions in German. Language and Speech, 62(4), 779-807. 
[5] Chang, C.B. 2021. Phonetics and phonology of heritage languages. In S. Montrul & M. Polinsky (Eds.), The Cambridge Handbook of Heritage Languages and Linguistics (pp. 581-612). Cambridge University Press.