68th 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.
Neural Character Models Don't Learn Rules for German Plurals, But That's Okay, Neither Do Germans - Kate McCurdy
Can artificial neural networks learn to represent inflectional morphology and generalize to new words as human speakers do? Some linguists have argued that the German number system cannot be modelled without rule-based symbolic computation, because the 'default' plural marker, /-s/, is also the least frequent; they claim that speaker preferences for /-s/ in elsewhere conditions, such as novel and phonologically atypical nouns, require representation of linguistic rules and thus cannot be learned from data alone. We present a new dataset of German speakers' production and rating of plural forms for nonce nouns, and note that the results challenge the claimed 'default' status for /-s/, reducing the potential trouble it may cause for neural models. Nonetheless, we observe that neural encoder-decoder models, while broadly successful on this 'wug' task, show distinctive failure modes suggesting they do not generalize in quite the same manner as human speakers.
Ambiguity in Explicit Discourse Connectives - Bonnie Webber
In a well-known paper, Pitler & Nenkova (2009) discussed usage ambiguity and sense ambiguity of explicit discourse connectives like 'when' and 'since'. Annotation of the PDTB-3 has revealed that explicit discourse connectives are also subject to other well-known types of ambiguity --namely, Part-of-speech ambiguity, Multi-Word Expression (MWE) ambiguity, Scope ambiguity, and Semantic role ambiguity. These observations are relevant for both cross-linguistic understanding of discourse connectives and 'shallow' discourse parsing.
Pitler, E. and A. Nenkova (2009). Using syntax to disambiguate explicit discourse connectives in text. ACL-IJCNLP '09, Singapore.
A model for temporal relations expressed with the conjunction ‘since’, and its interaction with tense and aspect in Serbian and Croatian - Hans Wilke
The temporal conjunction ‘since’ in dependent clauses can be used in two distinct ways: either to denote that one event happened prior to another event (anteriority), or to denote that two events are happening at the same time (simultaneity). In Slavic languages, these two different meanings are the result of the aspect used in the dependent clause. Consider the following two examples from Serbian:
(1) Otkad je prvi list kestena za?uteo, sunce se nije pojavilo.
‘Since the first leaf of the chestnut turned yellow, the sun has not reappeared.’
(2) Od kad je na svetu, ?ovek se predeljuje izme?u dva “ne”.
‘Since man has been on earth, he’s had to choose between the lesser of two evils.’
In example (1), the use of perfective aspect in the dependent clause results in a meaning of anteriority (i.e., the first leaf turned yellow before the sun stopped reappearing). Example (2), with imperfective aspect in the dependent clause results in a meaning of simultaneity (i.e., both events are ongoing) (Popovi? 2011). While this distinction as a result of aspect is attested across Slavic languages (e.g., Czech: van Duijkeren-Hrabová 2019; Polish: Genis 2018; Russian: Barentsen 1999), it does not take into account the role of tense and how it further impacts these meanings. For the current study, examples from Serbian and Croatian were considered to develop a more detailed description of possible temporal relations between events denoted with the conjunctions od kad, otkad(a) and otkako, ‘since’ (temporal). 246 examples were retrieved from the Amsterdam Slavic Parallel Corpus (ASPAC, Barentsen). These examples were analysed for aspect and tense in the main and dependent clause. Out of the 16 potential configurations of tense (past/present) and aspect (perfective/imperfective) between clauses, nine possible configurations were identified: three different types of anteriority and six different types of simultaneity. The different nuances in meaning will be described, and it will be explained how tense and aspect work together to denote these subcategories of meaning. This will be done using Reichenbach figures (Reichenbach 1947), which illustrate how the events in the main and dependent clauses relate to the moment of reference, the moment of speech, and to each other. Lastly, it will be explained why the seven unattested configurations – which are all of those containing perfective present tense – are a logical impossibility.
Are the effects of L1 attrition on spatial-temporal cognition in Mandarin-English bilinguals? - Winnie Yi-Jou Yeh
Does language influence the way we think? This is an essential question to linguistic relativity. Past studies have found that Mandarin and English speakers think about time differently because of the different spatial-temporal metaphors that each language uses. Both Mandarin and English speakers employ a horizontal FRONT/BACK spatial metaphors to describe time, but only Mandarin speakers employ an additional vertical UP/DOWN spatial metaphors. Therefore, Mandarin speakers are more likely than English speakers to create vertical representations of time (e.g. Boroditsky, 2001; Boroditsky, Fuhrman & McCormick, 2010). In a more recent study by Hendricks & Boroditsky (2017) examined whether learning new metaphors can create the same kinds of differences in implicit associations between space and time. They point out that exposure to new space-time associations can affect the strength of specific space-time associations (Hendricks & Boroditsky, 2017). They suggest that our representations of time are highly dynamic and is can change with new experiences. This leads to my research question of whether existing space-time mappings can be weakened due to new experiences or lack of old experiences. By using a STARC (spatial temporal association of response codes) task, I will examine Mandarin-English bilingual speakers with varying L1 and L2 experience and use while living in the L2 environment. This hopes to explore the possibility of an effect of attrition if representations of time are dynamic and vulnerable to restructuring.
When a foreign language interferes with your native language: The relationship between L2 acquisition and L1 attrition in English-Italian and Italian-English bilinguals - Mattia Zingaretti
Research on first-language (L1) attrition mainly focuses on highly-proficient second-language (L2) speakers of English who come from a variety of L1 backgrounds (i.e. Italian, Greek, Spanish, among others). Speakers undergoing L1 attrition are usually daily users of L2 English, and the changes in L1 production and comprehension that they experience occur primarily at the level of lexical access, and (following Sorace and Filiaci’s ‘Interface Hypothesis’) at the ‘interface’ of syntax and other linguistic domains, especially pragmatics. The extent to which these changes are permanent, however, is under-explored, and so is the relationship between L1 attrition and successful L2 acquisition. We thus propose to scrutinise said relationship, by carrying out a longitudinal study on English learners of Italian at Scottish Universities, and Italian learners of English at Italian Universities - half of whom will be spending one semester in the L2 environment. The proposed experimental tasks (i.e. a picture-naming task and a picture-verification task with eye-tracking) analyse domains prone to L1 attrition, such as (1) lexical retrieval and (2) the interpretation of subject pronouns, to understand whether attrition affects only students immersed in the L2, or classroom-based learners, too. We also investigate a third environment wherein languages such as Italian and English are known to differ – that is, sentence stress - as a potential domain for L1 attrition to occur in. Follow-up tests aim to uncover the extent to which possible L1 attrition effects may be ultimately reversed, and variables such as L1-L2 attitudes/ motivation/ input and use/ age of acquisition and individual cognitive profiles are also taken into account.
Language at Edinburgh Lunch committee
Lisa Gotthard, Hana Jee, Alex Lorson, Raji Satarai Madhavan, Hans Wilke
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.