When is Wall a Pared and when a Muro?: Extracting Rules Governing Lexical Selection

Abstract

Learning fine-grained distinctions between vocabulary items is a key challenge in learning a new language. For example, the noun "wall" has different lexical manifestations in Spanish -"pared" refers to an indoor wall while "muro" refers to an outside wall. However, this variety of lexical distinction may not be obvious to non-native learners unless the distinction is explained in such a way. In this work, we present a method for automatically identifying fine-grained lexical distinctions, and extracting concise descriptions explaining these distinctions in a human-and machine-readable format. We confirm the quality of these extracted descriptions in a language learning setup for two languages, Spanish and Greek, where we use them to teach non-native speakers when to translate a given ambiguous word into its different possible translations. Code and data are publicly released here. 1

Teacher Perception of Automatically Extracted Grammar Concepts for L2 Language Learning

“…'bhaat' and 'tandul' both refer to 'rice' in Marathi, but the latter refers to raw rice and the former refers to cooked rice). Chaudhary et al (2021) propose a method for iden-tifying such word pairs, along with explanations on their usage, using parallel sentences between English and the L2. Each pair of sentence translations is first run through an automatic word aligner (Dou and Neubig, 2021), which extracts word-byword translations, producing a list of English words with their corresponding L2 translations.…”

“…Vocabulary We follow the accuracy computation from Chaudhary et al (2021) to evaluate the model used for extracting semantic subdivisionsfor each sentence the model prediction is compared with the gold label which is the observed L2 word for the L1 word. The baseline uses the most frequently observed L2 word translation for the given L1 word and is compared with the gold label to compute the baseline accuracy.…”

“…Therefore, we train our own parser as outlined in subsection A.1. To learn models for extracting descriptions, we follow the same modeling setup as Chaudhary et al (2021) and Chaudhary et al (2022) and use decision trees (Quinlan, 1986) to extract the patterns, explanations and accompanying examples (subsection A.2). For suffix usage, we additionally train a morphology decomposition model (Ruzsics et al, 2021) which breaks a word into its lemma and suffixes, over which we learn a classification model.…”

“…To extract descriptions in human-readable format, we follow Chaudhary et al (2022) and Chaudhary et al (2021) and learn interpretable models such as decision tree (Quinlan, 1986) and SVM as they are conducive to interpretation.…”

“…For extracting rules for suffix usage and word usage, we follow Chaudhary et al (2021) and use a SVM classifier. The respective tasks are formulated as follows:…”