Graduate Student and Postdoctoral Fellow Research Day

In this presentation, I examine the properties of comparison constructions in Secwepemctsín. We regularly compare objects, places and people to each other, e.g. ‘yesterday the weather was nicer than it is today’, so this is a crucial part of human language.
It has become evident, in light of further cross-linguistic research, that languages vary with respect to how they express these notions (e.g. Beck et al. 2009). However, little work has been carried out on this topic in Salish languages (notably Davis & Mellesmoen 2019). The first question asked is (i) What is the syntactic form of a comparative? Is comparison expressed by morphology (e.g. like 'more' in English) or is there an alternative (pragmatic) strategy? Secwepemctsín, like English, uses morphology. More specifically, the language uses the predicate p’7e7cw ‘more’. Another question is (ii) Is there a restriction on what we can compare? In English, we freely compare to entities (e.g. ‘I am taller than John’) and actions (i.e. I bought more apples than John bought bananas’), though this is not the case across languages. We find that Secwepemctsín expresses comparison to both entities and actions. These results tell us that although Secwepemctsín and English superficially express comparison quite differently, they ultimately express core notions in the same way.
The Secwepemctsín data was collected via original fieldwork on Zoom with language consultants (Daniel & Leona Calhoun and Ron Ignace) during the Field Methods 2020-2021 class in the Department of Linguistics.

Both auditory and non-auditory effects of noise have been linked to deficits in language acquisition related to speech perception and reading comprehension. We conducted a literature review to prepare for a proposed study that will examine the association between e exposure to noise and language acquisition among 27,760 children attending Kindergarten in Metro Vancouver. To review the evidence of effects of noise exposure on language acquisition, we used two literature databases - Medline and PubMed - including keywords related to "Language development" AND "noise" (MESH term including, for example, language development, language acquisition, noise exposure, Signal-To-Noise Ratio). Keywords were searched in abstract and title. We identified grey literature by searching "exposure to noise on language acquisition" in Google Scholar. We identified additional literature by using a snowball method in included articles' bibliographies. A total of 74 studies were reviewed based on their relevance to the research question. The majority of studies in laboratory settings found an adverse association between noise and speech perception. However, due to lack of longitudinal studies, it is difficult to establish causality and effects of longer-term exposure to noise. Although a few empirical studies suggested detrimental effects of chronic noise exposure on reading comprehension, poor adjustment for potential confounding factors or risk of residual confounding, such as adverse birth outcomes, impairs the level of evidence. Through the planned study, we aim to address some of the limitations in current literature and improve the understanding of the effects of noise exposure on childhood language acquisition.

While distributional semantic models (DSMs) can successfully capture the similarity structure within a semantic domain, less is known about their ability to represent abstract semantic properties that hold across domains. Such properties can form the basis for abstract semantic classes that are a crucial aspect of human semantic knowledge. For example, the abstract class of extreme adjectives (such as brilliant and freezing) spans a wide range of domains (here, INTELLIGENCE and TEMPERATURE). Using a model that compares query items to an aggregate DSM representation of a set of extreme adjectives, we show that novel adjectives can be classified accurately, supporting the insight that a cross-domain property like extremeness can be captured in a word’s DSM representation. We then use the extremeness classifier to model the emergence of intensifier meaning in adverbs, demonstrating, in a separate task, the effectiveness of detecting this abstract semantic property.

From an early age, infants can integrate audiovisual speech (1), however, the development of the underlying cortical specialisation remains understudied. In adults, integration results in specific cortical responses: higher (super-additive) or lower (sub-additive) than responses to unisensory speech (2). Our study aimed to show whether such responses emerge in infancy, by comparing responses to synchronous audiovisual and asynchronous auditory/visual syllables. In the asynchronous condition visible articulation lagged 660ms behind audible articulation, creating perception of alternating syllables (3). In a cross-sectional study, we used 34-channel fNIRS to measure 5- and 10-month-olds’ (N=42) fronto-temporal responses to synchronous and asynchronous audiovisual syllables. We used multivariate pattern analyses (MVPA) to decode patterns of cortical responses to synchronous and asynchronous speech. MVPAs showed that the patterns of oxygenated haemoglobin responses over the right hemisphere could be classified at a level significantly greater than chance at five months of age (p=.009). Channel-wise RM ANOVAs showed that the successful classification was driven by sub-additive responses over right inferior frontal channels (p≤.001, corrected). In the older age group, classifications were unsuccessful (p>.2) even though two bilateral channels showed sub-additive responses (p<.05, uncorrected). In an adult validation fMRI study (N=12), we found sub-additive responses over the left temporal region, and super-additive responses over the bilateral temporal and right frontal regions. Together these results show the possible developmental timeline of emergence of specific cortical responses to integration of congruent audiovisual speech.

We investigate transfer learning based on pre-trained neural machine translation models to translate between (low-resource) similar languages. This work is part of our contribution to the WMT 2021 Similar Languages Translation Shared Task where we submitted models for different language pairs, including French-Bambara, Spanish-Catalan, and Spanish-Portuguese in both directions. Our models for Catalan-Spanish (82.79 BLEU) and Portuguese-Spanish (87.11 BLEU) rank top 1 in the official shared task evaluation, and we are the only team to submit models for the French-Bambara pairs.

The present study is a computational stylometric approach to the hitherto elusive writings of Louis Zukofsky (1904-1978). While modernist contemporaries of Zukofsky have been largely examined in a qualitative framework (Stanley 1990, Golston 2006, Parker 2009, Monacell 2011, Jennison 2012, Scroggins 2015, Rowe 2019), this study approaches Zukofsky’s work from a lexicogrammatical perspective and seeks to illuminate the often inscrutable poetry of the thoroughly experimental Modernist by evaluating various qualitative claims regarding their “disjunctive poetics” (Quartermain 1992), supplementing previous discourses and close readings with quantitative analysis.
Castiglione asserts that the joint usage of participle clusters and grammatical incompleteness clearly manifest throughout Zukofsky’s active years, establishing a linguistic phenomenon that Yaron names Defective Representation (DR) (2013, 2002). In order to ascertain both the prevalence and uniqueness of Zukofsky’s deployment of DR in relation to his contemporaries, I compile and scan the collected works of Zukofsky’s early modernist predecessors, notably his mentors Ezra Pound and William Carlos Williams, as well as his late modernist peers, rendering them into digitally analyzable text. Methodologically, my approach uses Stylo and Antconc softwares to inspect the large amount of text, calculate participle cluster and dash frequencies, and subsequently, record the examples of DR.
The results of this study suggest that there is visible lexical similarity between Zukofsky, fellow late Modernist Lorine Niedecker, and Pound, however, Pound’s usage of participle clusters is minimal, indicating that Zukofsky deviated from his teacher. Additionally, Zukofsky, Niedecker, and Williams use a commensurate amount of dashes, achieving a stylistic effect that could be considered akin to DR, a finding that demonstrates DR could arguably be constituted of multiple poetic techniques.

The articulation of a vowel is a dynamic process, and the measurements that are used in linguistics to capture the quality of a vowel (namely, F1 and F2), change over the course of its duration. Often, this change is non-linear, as in the case of diphthongs such as /oʊ/. For this reason, statistical analysis of formant values at only the midpoint of a vowel may not accurately capture the dimensions of difference in their articulation, when comparing between speakers. In this talk, I will explain how to use Generalized Additive Mixed Modeling (GAMM) to analyze and compare formant contours. The data come from tokens of English /oʊ/ and Korean /o/ in various phonological environments, recorded in naturally-occurring bilingual speech. The analysis used for this illustration looks at the effect of coarticulation-inducing adjacent segments (such as a preceding coronal consonant or a following lateral) on the F1 and F2 contours of both vowels. The GAMM analysis shows that the Korean /o/ stays more level in F2 over the course of the vowel, while English /oʊ/ is more affected by neighboring segments not only at its midpoint, but throughout the entire vowel trajectory. This talk will briefly touch on the theoretical implications on bilingual phonology, but its primary focus is on the usefulness and methodology of using GAMM in phonetics research.

Dyslexia is a neurodevelopmental reading disorder that is also often associated with motion perception deficits[1,2]. Although the reason for motion deficits in dyslexia and other developmental disorders is not fully understood, these deficits have been attributed to vulnerability in the dorsal processing stream that projects from occipital to parietal cortex[3]. However, the ventral stream that projects from occipital to temporal lobe has also been implicated in dyslexia[4]. To evaluate dorsal stream vulnerability in dyslexia, we used resting-state functional magnetic resonance imaging data collected before and after a 3-month reading intervention for children in grade 3[5]. We determined the effect of the intervention on functional connectivity between the motion-processing cortical area V5/MT and regions of the dorsal and ventral visual streams that have been implicated in reading. Reading ability was also assessed before and after the intervention using the Kaufman Test of Educational Achievement (KTEA)[6]. Following the intervention, we found improvements in reading fluency, increased functional connectivity between V5/MT and dorsal stream areas (inferior parietal cortex), and decreased functional connectivity between V5/MT and ventral stream areas (lateral occipital cortex). The MT-dorsal stream connectivity changes were positively correlated with KTEA reading fluency scores, whereas the MT-ventral stream connectivity changes were negatively associated with the reading fluency scores. Our results suggest that improved reading fluency after the intervention is associated with increased MT-dorsal processing stream connectivity, but with decreased MT-ventral processing stream connectivity, providing evidence for the dorsal stream vulnerability hypothesis in dyslexia.

We describe models focused at the understudied problem of translating between monolingual and code-mixed language pairs. More specifically, we offer a wide range of models that convert monolingual English text into Hinglish (code-mixed Hindi and English). Given the recent success of pretrained language models, we also test the utility of two recent Transformer-based encoder-decoder models (i.e., mT5 and mBART) on the task finding both to work well. Given the paucity of training data for code-mixing, we also propose a dependency-free method for generating code-mixed texts from bilingual distributed representations that we exploit for improving language model performance. In particular, armed with this additional data, we adopt a curriculum learning approach where we first finetune the language models on synthetic data then on gold code-mixed data. We find that, although simple, our synthetic code-mixing method is competitive with (and in some cases is even superior to) several standard methods (backtranslation, method based on equivalence constraint theory) under a diverse set of conditions. Our work shows that the mT5 model, finetuned following the curriculum learning procedure, achieves best translation performance (12.67 BLEU). Our models place first in the overall ranking of the English-Hinglish official shared task.

Velopharyngeal Opening (VPO) is associated with nasality in speech. Earlier studies on French suggested that there is a continuum of VPO: nasal vowels, nasal consonants, contextually nasalized vowels and, finally, oral segments. However, most of these studies used isolated words/syllables, very small numbers of participants, and/or indirect measures of the VPO. In this study, we tested whether the predictions from the previous studies extend to sentence-level speech with a substantially larger speech sample drawn from more participants using a more direct measurement of the VPO. Using the Université Laval X-ray videofluorography database containing speech from nine Quebecois French speakers, the VPO was measured for speech segments and rest positions. Results indicate that VPO was greatest during between-utterance intervals, followed by (in descending order) phonemically nasal segments, contextually nasalized segments, and oral segments. Contrary to previous suggestions of differences of VPO in nasal segments, no difference was found between nasal vowels and nasal consonants suggesting that such classification is not necessary (and the traditional binary [±nasal] feature still holds). Furthermore, in contrast to previous reports, anticipatory nasalization was found to have greater VPO than carryover nasalization.

This paper considers the narrative style of the popular series of novels for young readers, A Series of Unfortunate Events by Lemony Snicket. Snicket frequently features “narrative moments” (Williams 1998) which call attention to the act of narration and storytelling. Metafiction has long been seen in children’s fiction (e.g., Moss 1985; Philpot, 2005; Sanders, 2009), but the use in this series is comparatively unique to most previous examples. Here Snicket is constructed as simultaneously the author of the novels, the narrator of the story, and a character existing within the story world. 

Children’s literature as a genre has often served an educational function (see, e.g., Nodelman, 1996). While the use of metafiction has sometimes been used to counter the overt didacticism of the genre, efforts of anti-didacticism do not always succeed in this mission (Sanders 2009). Lemony Snicket’s narrative style, however, seems to offer a style of anti-didacticism that is engaging and possesses educational value. During narrative moments Snicket often defines words and phrases for the reader using his signature phrase, “a word which here means…” (Snicket 1999a), which I suggest introduces to young readers the idea that language is context dependent and used for communication.

The goal of this research is to provide quantitative data from corpus analysis (using the stylistic analysis tools stylo (a package for R) and AntConc) to support the argument that the anti-didactic themes present in Lemony Snicket’s A Series of Unfortunate Events have educational value.

In 2018-2019, over 54,000 (9.5%) K-12 students were enrolled in French immersion (FI) in British Columbia1. While parents of students with or without developmental disabilities (DD; e.g., autism spectrum disorder, Down syndrome, and developmental language disorder) may wish for their child to participate in FI, very little research has been conducted on the former group’s experiences in the program2, 4, 5. As a result, some children may be discouraged from participation when they have the right and potential to do so. Others may enroll only to withdraw later, finding that their needs are best fulfilled elsewhere.

This study aims to understand, from education and healthcare providers’ perspectives, what helps and hinders the participation of children with DD in FI. Data were collected as part of a larger international study3. Fourteen interview transcripts from professionals with at least five years of experience working with students with DD in the Vancouver region were coded using qualitative content analysis6.

Results indicated facilitators and barriers at many levels: child, family, provider, school, community, and system. In addition, combinations of factors were found to create supportive or hindering conditions (e.g., a child who already knows some French and can access appropriate school resources). Furthermore, some factors were presented as potential mediators (e.g., FI generally deemed not an ideal choice for children with DD unless French was spoken at home).

Results will offer providers insight into factors that may be useful to consider when counselling families of prospective students with DD interested in enrolling in FI.

Linguistic studies, especially in cognitive and usage-based approaches to language, increasingly feature multimodal and interactional data (Feyaerts et al., 2017). Nevertheless, the gestural resources that speakers use for the task of enumeration, i.e., listing and counting, have not been widely explored, perhaps due to the high degree of variation in how speakers––or signers––count on their fingers. Likewise, gestural forms used for the ad hoc listing of items or concepts are similarly varied. This study explores the co-speech gestures that accompany listing, paying particular attention to lists that end with a so-called generalized (list) extender, or GE (Overstreet & Yule 1997, 2001), such as and stuff like that, or whatever, or something, etc., as in (1) and (2) from the Red Hen multimedia archive (Steen & Turner 2013; GEs in bold). These examples exhibit the tendency for lists to contain either just one listed item and a GE, as in (1), or multiple items, as in (2).

(1) Like I was tough and cool, but my imagination would get the best of me and I would be nervous of puppets and things like that. Were you a scary movie fan?

(2) Really, I don't have the emotional bandwidth right now to give a hoot about, you know, Olive Jade wanting to have her platform back and be accepted by social media or whatever.

In earlier work, we proposed that lists with three specific items (let’s say, X, Y, and Z) are gestured differently than when two referentially specific items (X and Y) are followed by a GE, as in the screen shots below (Hinnell & Rice 2020; Bohach 2020).

Here, we focus on the gestural co-expression of lists as functions of linguistic features of the list, e.g., number and morphosyntactic complexity of the items in the list and referential vs. pragmatic function. Using videos from the Red Hen archive, we quantitatively describe a variety of co-speech embodiments which accompany three-part lists, investigating specifically which linguistic factors determine the gestural form of lists for the GE expressions and things like that, or whatever, and whatnot, and or something like that. This study furthers the discussion of the tight integration of speech and gesture while addressing the theoretical implications of the high degree of variability inherently found in gestural forms.

Image on left: Enjoy your day, enjoy your week, enjoy your year... (gestured with beat gestures)

Image on right: ...whatever (gestured with an open hand and spread fingers)