Date created: | Last Updated:

Category: Communication

Description: Nouns phrases constitute is a major source of language transfer phenomenon, both positive and negative (interference), in the second language acquisition of languages with gender. For speakers of languages without gender, the grammatical gender agreement places an extra difficulty during the course of acquisition, especially for languages with nominal system, such as Spanish, which require a lot of inflections. However, for words with incongruent genders in the native language and the target language, the case could be worse for learners of languages with gender as the first language. As showed in [PCM+10], such incongruence posed an extra processing load to the learners and resulted in a longer response time. They concluded that there should be a direct nonsemantic lexical route in the mental lexicon connecting the two languages at some specific grammatical level, such as the grammatical gender in this case. It is also suggested that the processing or retrieval of the gender information precedes the phonological information (e.g. [CLP+05]). While in the field of psycholinguistics, most of the studies focus on the difference in response (or translation) time in formal experimental setting, in this work, we employ a publicly available dataset of second language learners [SBG+18]. The goal of the present work is to analyse the common errors that a second language learner make in a given language. We would like to check empirically whether a language learner would be influenced by his/her native language in the errors they make, particularly on gender agreement. We would like to address the following research question: - Given a language learner that speaks a specific language (L1) who is learning another lan- guage (L2), with both languages having two genders. - Given also a set of words which are incongruent in terms of gender, will the learners make significantly more mistakes? We try to answer these questions by analysing the dataset presented for the Duolingo shared task [SBG+18]. The dataset contains information about some simple exercises encoded in a format similar to the CONLL-U format. This dataset contains inputs from language learners in their first 30 days of learning and it includes data about the country in which the student is doing the exercises, the number of days that has been spent learning the language, type of exercise and the amount of time spent for each trial, and other information about the specific user. Together with this, they include automatically generated linguistic information, such as POS-tags, morphological features and also syntactic information. In the experiment, we will assume that a person that does the exercises from only one country, has mastered the language in question. In order to check whether a specific community of L2 learners have more difficulties in learning a language due to lexical incongruence in the languages, we will proceed as follows. We will check if the students of our interest, namely Italian learners of Spanish, make significantly more mistakes than the general trend in the case of gender-incongruent words.