62nd 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.

If you would like to present your language-related research, please email us with a title and an abstract by TUESDAY 3 April, and a digital version (pdf) of your poster by THURSDAY 5 April, which we will print for you and you will get to take home at the end. 

Please send the abstracts, posters and other enquiries to:

languagelunch@ed.ac.uk

We would ask you to mind these few guidelines when preparing your abstract:

• aim at about 100-300 words in length (including any example sentences and references)
• if linguistic examples are glossed, they should follow the common Leipzig Glossing Rules
• the abstract may be sent within an email message or as an attachment (doc/docx or txt are fine)
• if not using plain text, kindly refrain from using formatting that is prone to causing errors or getting lost between conversions (e.g., tabs, small caps; automatic indentation, bullets and numbering).

Posters

Narrative abilities of 4-to-6 year old Welsh-English bilingual children with and without SLI - Diana Lopez-Lugo, Dr Vicky Chondrogianni

Background: Narrative tasks are less biased methods that assess the language abilities of bilingual populations. They provide valuable information about the children’s linguistic development and can distinguish between mono-or bilingual children with and without SLI (Specific Language Impairment).

Aims: 1) To analyse the macro and microstructural narrative abilities of 4-to-6 year old Welsh-English bi-SLI and bi-TD children. 2) To evaluate the diagnostic validity of the LITUMS-MAIN [1] as a clinical tool that can identify the absence or presence of SLI when bilingual children are tested in their L2.

Methods: Narrative abilities were examined on macro and microstructure levels (N = 18 bi-TD) (N = 10 bi-SLI). Multiple regression analyses were performed to measure the children’s narratives. Binary logistic regressions and operating characteristic (ROC) curve analyses were run to check the diagnostic validity of the test battery.

Outcomes & Results: SLI negatively affects the narrative performance of bilingual children. The LITMUS-MAIN can identify TD with 94% of accuracy and SLI with 50% of accuracy when children’s narratives are tested in their L2.

Conclusion: SLI negatively affects the children’s macro and microstructural narrative abilities. The LITMUS-MAIN can successfully rule out the presence of SLI when bilingual children are tested in their L2.

Abstract Meaning Representation for Paraphrase Detection - Marco Damonte 

Abstract Meaning Representation (AMR) parsing aims at abstracting away from the syntactic realization of a sentence, and denoting only its meaning in a canonical form. As such, it is ideal for paraphrase detection, a problem in which one is required to specify whether two sentences have the same meaning. We show that naive use of AMR in paraphrase detection is not necessarily useful, and turn to describe a technique based on latent semantic analysis in combination with AMR parsing that significantly advances state-of-the-art results in paraphrase detection for the Microsoft Research Paraphrase Corpus. Our best results in the transductive setting are 86.6% for accuracy and 90.0% for F-1 measure.

Onset of disambiguation as word learning strategy delayed in multilingual infants – In-sights from eye-tracking research - Kate Repnik

To break down the complexity of word learning children might use certain learning constraints, such as ‘disambiguation’ (e.g. Markman, 1991; Halberda, 2003), which has been regarded as typical for monolingual children. Disambiguation describes the assumption that new words tend to refer to new referents precluding the application of a novel label to a familiar object. This means that children assume that each concept or object only has one label. However, newer associative network models (Roder, et al. 2000) describe lexical development with domain-general learning processes that contribute to ‘in-the-moment’ referent selection, whilst also con-sidering ecological and environmental factors. According to this view, disambiguation is the product of prior episodes of learning, which are the consequence rather than the cause of a pruned associative network.

Interestingly, multilingual children (aged 18 months) make less use of disambiguation than their monolingual peers (Byers-Heinlein & Werker, 2009). However, these studies did not control for within- vs. across language disambiguation. This was because (a) they included translation equivalents which can lead to interference from the untested to tested language, and (b) they did not include different language testing modes for multilinguals (i.e. no bilingual (switch) mode).

The central question addressed in this research is: do multilingual children differentiate between different kinds of disambiguation when mapping novel labels to an object? If they do, it would imply that multilinguals do indeed have well-structured networks, but that these networks are not as strongly linked and more independent of each other. In the present study, we focused on across languages disambiguation. To avoid interference of translation equivalents and test the replicability of previous findings, we familiarised children with a new item. The research questions we addressed were (a) whether children disambiguate and retain the trained word-object mapping, and (b) how age and language background modulate disambiguation and retention.

Eye-tracking data from 18, 24, and 30-month-old monolingual and multilingual infants were collected. In total, 57 children (21 female) from monolingual English (n=28) and multilin-gual (English plus any language, n=29) backgrounds were tested. They were exposed to a looking-while-listening paradigm with two objects on screen. Trials included two familiar items (ball, car), a novel item (dax), and a trained item (nil). Results were analyzed using mixed-effects regression models. In significant interactions on disambiguation trials with target items, language background, and age, we found that age modulated the use of disambiguation for multilingual infants indicating that they start relying more on disambiguation as they get older.

The youngest monolingual infants (aged 18m) perform better on disambiguation trials than their multilingual peers, however, already by the age of 24 months this difference is levelled out.

Results also indicate that only 24- and 30-month-olds could retain the trained word-object pairing with an interaction of language background for the 30-month-olds. The learning curve for multilingual infants on retention trials from 18 to 30 months old is significantly steeper than the one of their monolingual peers. Performing better on retention trials as 30-month-olds and making more use of disambiguation at that age, one first conclusion to draw from these results is that multilinguals start using disambiguation at a later stage, and furthermore, is a suitable strategy also for them to retain words.

Contact

LANGUAGE Lunch

Language at Edinburgh Lunch committee

Joachim Fainberg, Andres Karjus, Madeleine Long, Joana Ribeiro, Eva-Maria Schnelten

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.