Automatic Translation of Spatial and Non-Spatial Expressions from English into Portuguese
A Study Involving Qualitative Spatial Reasoning
Abstract
Neural models for Machine Translation have been proving significant advances in performance in recent years, becoming the most widely employed tools for that task. Nevertheless, they still struggle with semantically complex expressions. In this paper, we investigated the performance of two of the most widely used systems available today, namely Google Translator and DeepL, in translating expressions involving representing objects in physical space, always using the same prepositions in English, divided into situations admitting spatial interpretations or not. Formalizations based on Qualitative Spatial Reasoning models are applied to the representation of the spatialized expressions, making it possible to compare the logical form of spatial information in the original texts and in automatic translations from English into Portuguese.
Results show that these two major translation systems still make many mistakes when it comes to relatively complex expressions. Among them, Google Translator had more errors globally (35.52%) whereas DeepL, with the best overall performance (11.72% of errors), made significantly more mistakes with expressions involving spatiality, which suggests that translating these expressions increases the difficulty of translation for this model. In addition, a study on the specific type of difficulties in spatialized expressions shows that the most frequent machine translation problem concerns the incorporation of manner into the predicate, which is realized by different means in English (where the verb tends to incorporate manner) and in Portuguese (where manner is realized in adjuncts to the verb). Results from the present work may provide specific criteria for evaluating and improving machine translation models between these and other languages in which the same differences in predication occur.
Copyright (c) 2025 Rafael Macário Fernandes, Rodrigo Souza, Marcos Lopes, Paulo Santos, Thomas Finbow
This work is licensed under a Creative Commons Attribution 4.0 International License.
Authors who publish with this journal agree to the following terms:
- Authors retain copyright and grant the journal right of first publication with the work simultaneously licensed under a Creative Commons Attribution License that allows others to share the work with an acknowledgement of the work's authorship and initial publication in this journal.
- Authors are able to enter into separate, additional contractual arrangements for the non-exclusive distribution of the journal's published version of the work (e.g., post it to an institutional repository or publish it in a book), with an acknowledgement of its initial publication in this journal.
- Authors are permitted and encouraged to post their work online (e.g., in institutional repositories or on their website) prior to and during the submission process, as it can lead to productive exchanges, as well as earlier and greater citation of published work (See The Effect of Open Access).
