69th Language at Edinburgh lunch

The Language@Edinburgh Lunch is a bi-monthly opportunity to present your work to an interdisciplinary audience in an intimate and feedback-rich setting, all while enjoying a buffet lunch.

Posters by both postgraduate students and academic staff are welcome on any area of human language research - including all sub-fields of linguistics, philosophy of language, natural language processing, psycholinguistics, and any other language related discipline. Reporting on work in progress is equally welcome.

Posters

Regional Variation in Scottish T-glottaling - Lauren Hall-Lew, Nina Markl, Brandon Papineau and Matthew Sung

T-glottaling is a well-studied variable in UK English and a well-known feature of Scottish varieties. Speaker age, gender, social class, and formality of talk are typical social predictors (e.g., Stuart-Smith 1999; Schleef 2013). Some have posited that the glottal variant may have been innovated in Scotland separately from in Southern England (Schleef 2013; Smith & Holmes-Elliott 2018). To explore the constraints on glottal realization, as well as this polygenetic hypothesis, we consider variation within Scotland. While Stuart-Smith (1999) identified differences within Glasgow according to social class, we consider differences within Scotland according to region and formality.

A database of 17 famous Scottish women, selected based on the public availability of their speech and with the aim of maximizing regional and socioeconomic diversity, was compiled by a class of 36 university students. Born between 1955 and 1994, the speakers were classified according to five broad regional groups and three broad social class groups. Groups of coders obtained roughly 10 minutes of continuous speech per speaker. 3,469 tokens were fully coded twice, auditorily. Tokens were coded according to a binary distinction: ‘glottal stop’ and ‘alveolar stop’. Word-final and word-medial contexts were modeled separately. Internal constraints included number of syllables and following phonological environment. External constraints included speaker factors and contextual factors. Results from a best-fit mixed-effects logistic regression model for word-medial /t/ found only phonological context and number of syllables to be significant predictors. The model for word-final /t/ found these same effects, but also effects of region and formality. To unpack the interspeaker results we conduct an intraspeaker analysis to demonstrating how the ‘regional’ results may be a consequence of differences in social personae.

Challenging Functional Explanations of Variable TD Deletion in American English - Annie Holtz

This study examines the implications of pursuing a functional explanation for variable phonological rules, focusing on TD Deletion in American English. The phonological process whereby a final /t/ or /d/ is deleted in a consonant cluster has been a topic of intense scrutiny for decades, as it appears to be a prime example of a variable rule with non-phonological conditioning factors (e.g. Twaddell, 1935; Bybee, 1997). One such factor is morphological class, whereby monomorphemes, e.g. mist, exhibit higher rates of TD Deletion than regular past tense forms, e.g. missed, (Labov,1989). One attempt to explain this variability comes from the functionalist framework and involves positing a general strategy for ambiguity avoidance (Kiparsky, 1972). Consider the following two sentences:

a. The city was covered in mist.

b. I missed my grandmother.

According to this functionalist approach a speaker would be less likely to delete the final stop in a than in b since doing so eliminates crucial grammatical information and modifies the meaning, whereas this does not happen if the stop is deleted in a. By this same reasoning a general strategy for ambiguity avoidance could affect levels of TD Deletion within the class of monomorphemes. For example, if a monomorpheme undergoes deletion and the resulting form is a homophone of another word then this also introduces ambiguity. Thus this functional theory would predict speakers to have lower rates of TD Deletion for monomorphemes where the truncated form has homophones. Using data from the Buckeye corpus I investigate this prediction by comparing rates of TD Deletion between monomorphemes with and without homophones. The results indicate that the presence of homophones do not affect rates of TD Deletion in monomorphemes and thus calls into question the explanatory power of ambiguity avoidance.

Plan, Write, and Revise: an Interactive System for Open-Domain Story Generation - Seraphina Goldfarb-Tarrant

We present a neural narrative generation system that interacts with humans to generate stories. Our system has different levels of human interaction, which enables us to understand at what stage of story-writing human collaboration is most productive, both to improving story quality and human engagement in the writing process. We compare different varieties of interaction in story-writing, story-planning, and diversity controls under time constraints, and show that increased types of human collaboration at both planning and writing stages results in a 10-50% improvement in story quality and user engagement. Finally, we find that humans tasked with collaboratively improving a particular characteristic of a story are in fact able to do so, which has implications for future uses of human-in-the-loop systems.

Classifiers in Nung, a Tai-Kadai Language in Vietnam - Esther Lam

Classifiers are elements related to the noun phrase that often co-occur with numerals and other quantifying elements (Aikenvald, 2000). Languages that use classifiers tend to lack number marking morphology and (in)definite articles. Hence, the properties of nouns in ‘classifier languages’ and the roles of classifiers in marking number and (in)definiteness have been widely debated in the literature. It has been proposed that all nouns in classifier languages are inherently mass and classifiers function to individuate (Denny, 1986; Chierchia 1998), also that ‘bare classifiers’, which is the use of classifiers without numerals, function like definite articles (Cheng & Sybesma, 1999). Nung is a ‘classifier language' that lacks number marking morphology and (in)definite articles. I will present my primary fieldwork data on Nung to shed light on the properties of nouns and the functions of classifiers in a classifier language. My data suggests that not all nouns in the language are mass; and that although classifiers have a role to play in individuation, some classifiers do not seem to perform this function in some types of sentences. My data also suggests that definiteness is not indicated by bare classifiers alone; other factors, such as the position of the bare classifier construction in a sentence, appear to affect its (in)definite interpretation.

“I’m petrified to use Arabic in an airport” – A sociolinguistic enquiry into the public use of Arabic of first-language speakers in the UK - Gerard Murphy

This project investigates the patterns in the public language use of first-language Arabic speakers in the UK, as well as the intimidation experienced by this group and how it has affected their language attitudes towards Arabic and their use of Arabic in public. Through the use of online questionnaires and personal interviews, this study looked at where participants choose to use Arabic and English, their levels of comfort speaking Arabic in various public spaces and any intimidation they have experienced for speaking Arabic in public. The findings show that participants feel particularly uncomfortable speaking Arabic in ‘transit spaces’ such as on public transport and in public transport hubs (e.g. airports, train stations) and that participants will often switch from Arabic to English in public spaces to avoid unwanted attention. Intimidation is found to have a profound effect on participants’ public use of Arabic but, surprisingly, had very little effect on language attitudes towards Arabic.

Movie Plot Analysis via Turning Point Identification - Pinelopi Papalampidi

According to screenwriting theory, turning points (e.g., change of  plans, major setback, climax) are crucial narrative moments within a screenplay: they define the plot structure, determine its progression and segment the screenplay into thematic units (e.g., setup, complications, aftermath). We propose the task of turning point identification in movies as a means of analyzing their narrative structure. We argue that turning points and the segmentation they provide can facilitate processing long, complex narratives, such as screenplays, for summarization and question answering. We introduce a dataset consisting of screenplays and plot synopses annotated with turning points and present an end-to-end neural network model that identifies turning points in plot synopses and projects them onto scenes in screenplays. Our model outperforms strong baselines based on state-of-the-art sentence representations and the expected position of turning points.

Being a dialect detective – how much can the numbers 1 to 10 tell us about where someone is from? - Matthew Sung

A recording of a Chinese dialect speaker counting from 1 to 10 was found. This recording was made in 2013 and the exact origin of the speaker is not clear. The only information we have is that this person is from the Fujian province, China. This information can save some time to find the origin of the speaker, but it is not extremely helpful, since the Fujian province is the homeland of the Min dialect. There are several sub-dialect groups within Min. They are all mutually unintelligible to one another (Yan 2006: 122) and it is not easy to find data for a variety of Min dialects. Having the Fujian province as a really linguistically diverse area, it makes it even more difficult to trace the potential origin of the dialect speaker of the recording.

I used three different methods to find the potential origin of the speaker. First I looked for salient features that are present in the recording and compared them with the representative dialects of each Min sub-dialect group. The result shows that this variety is equally similar to Northern, Eastern Min and the Puxian dialect based on 6 features only. Next, I applied the FIT-technique (Benskin 1991b). The FIT-technique eliminate dialects with features that are not present in the unknown dialect on a map. It turns out that all the possible dialects in Fujian were eliminated using this method. Lastly, I applied dialectometry (Heeringa 2004, Nerbonne & Kretzschmar Jr. 2013), using the Levenshtein distance and hierarchical clustering. Levenshtein distance is “the notion of string-changing operations”. When cognates are compared, any deletion, insertion or substitution of segments are calculated between the two. Hierarchical clustering is grouping objects together from bottom-up. In this case, the objects will be the dialects. The unknown dialect appears to be clustered with the Puxian sub-dialect group of Min.

Contact

LANGUAGE Lunch

Language at Edinburgh Lunch committee

Lisa Gotthard, Alex Lorson, Hans Wilke

Further information

The Language at Edinburgh Lunch is made possible through funding from the School of Philosophy, Psychology and Language Sciences and the Human Communication Research Centre, with the intent to facilitate interdisciplinary language research at the University of Edinburgh.