Where and How Do Languages Mix? A Study of Spanish-Guaraní Code-Switching in Paraguay

Olga Kellert, Nemika TyagiNAACL 2025

TL;DR

This paper maps Spanish-Guaraní code-switching patterns in Asunción, Paraguay, using geotagged tweets. It reveals distinct geographic clusters of language use and identifies that code-switching most frequently occurs mid-sentence, often involving verbs, pronouns, and adjectives, offering new insights into the interplay of linguistic, social, and geographic factors in bilingual communication.

What Problem Does This Paper Address?

Code-switching, the phenomenon of alternating between languages within a single utterance, is a widespread linguistic behavior. While extensively studied, most research has focused on either the syntactic or geographic aspects of code-switching, or has relied on traditional methods like elicited data and interviews, which may not fully capture the natural spontaneity of everyday speech. Furthermore, there's a significant gap in understanding code-switching for less-studied language pairs and in integrating geographic and syntactic analyses. This paper addresses this by examining Spanish-Guaraní code-switching in Paraguay, a country with a unique bilingual environment, using digital communication data to provide a comprehensive, integrated analysis.

Methods

The researchers employed a two-pronged approach:

  • Geographic Localization: They used geotagged tweets from Asunción and applied a "Differential Distribution" metric. This metric calculates the relative frequency of Guaraní versus Spanish words within small, equal-sized geographic zones (100x100 bins). Positive values indicate a higher presence of Guaraní, while negative values indicate a higher presence of Spanish. This allows for fine-grained mapping of language preferences across urban areas.
  • Syntactic Localization: Sentences were segmented into three parts: initial, middle, and final. Code-switching was identified by consecutive words assigned different language labels. This segmentation allowed them to categorize the syntactic positions where switches occur and analyze their distribution.

Datasets

The study utilized a dataset of tweets collected via the Twitter API from 2017 to 2021, filtered for the Spanish language tag ("es"). Guaraní words were identified based on their frequency in the literature. The geographic mapping utilized OpenStreetMap data.

Key Insights and Ideas

  • Geographic Segregation of Languages: The study reveals a clear spatial pattern in Asunción, with Guaraní-dominant tweets concentrated in western and southwestern areas, and Spanish-dominant tweets in central and eastern regions. This suggests that local context, potentially linked to indigenous roots, informal settings, or formal/commercial districts, significantly influences language preference.
  • Mid-Sentence Switches are Common: Code-switching doesn't just happen at sentence boundaries; it frequently occurs within the middle of sentences, particularly involving verbs, pronouns, and adjectives. This highlights the fluid nature of bilingual speech and suggests that certain grammatical elements are more prone to switching.
  • Digital Platforms as a Window into Bilingualism: Social media, specifically Twitter, provides a valuable, real-time source of data for studying code-switching, overcoming the limitations of traditional research methods.
  • Guaraní as a Marker of Identity/Intimacy: The prevalence of code-switching in tweets related to emotions and social interactions suggests that Guaraní may serve as a marker of identity and intimacy in informal communication.

Key Results

  • Geographic Patterns: Red markers (Guaraní prominence) were observed in the western and southwestern parts of Asunción, while blue markers (Spanish prominence) dominated the central and eastern areas. Transitional zones showed an overlap, indicating balanced bilingual practices.
  • Syntactic Patterns: Code-switching was found to occur most frequently in the middle of sentences (Figure 2).
  • Parts of Speech Involved: Verbs were the most common part of speech involved in Spanish-Guaraní code-switching (49.0%), followed by nouns (22.4%), adjectives (18.4%), and pronouns (10.2%) (Figure 3).

Limitations and Future Work

The authors acknowledge several limitations:

  • Demographic Representation: Twitter data may not fully capture the language use of all demographic groups, particularly older or less digitally active populations.
  • Scope of Analysis: The current study focused on geographic and syntactic dimensions, leaving sentiment and discourse dynamics for future exploration.
  • Syntactic Precision: The method approximates syntactic position rather than pinpointing exact locations within complex sentence structures.

Future work could involve integrating data from other social media platforms, incorporating sentiment analysis and advanced syntactic parsing, and applying these methods to other bilingual communities to assess generalizability and further understand global bilingualism.

Summary updated

BibTeX
@inproceedings{kellert2025where,
  author = {Olga Kellert and Nemika Tyagi},
  title = {{Where and How Do Languages Mix? A Study of Spanish-Guaraní Code-Switching in Paraguay}},
  booktitle = {Conference of the North American Chapter of the Association for Computational Linguistics},
  year = {2025},
}
Figure