Food-based dietary guidelines beyond language: context, food security, nutrition, and sustainability: authors' reply

We thank Rios-Castillo and Kanter¹ for their thoughtful Correspondence regarding our analysis of dairy-related messages in food-based dietary guidelines (FBDGs). We welcome the opportunity to clarify the scope and interpretation of our findings. Importantly, several of the cautions they raise are consistent with limitations explicitly acknowledged in our Manuscript.²

Our study was designed to characterize descriptive lexical and semantic differences in dairy-related messaging across national FBDGs, not to rank guidelines or assess their quality, appropriateness, implementation, or effectiveness. Throughout the analysis, natural language processing (NLP)-derived metrics were treated as descriptive indicators of policy language. We explicitly stated that the observed semantic differences should be interpreted as variation in dietary messaging rather than as direct measures of policy quality, implementation, or effectiveness. We therefore agree that simpler language may be appropriate, and potentially preferable, when the objective is to communicate clear, accessible, and actionable recommendations. Lexical complexity and communicative effectiveness are distinct constructs, and our study did not equate them.

Rios-Castillo and Kanter also highlight heterogeneity in guideline availability, length, structure, and comprehensiveness. We explicitly acknowledged these limitations. Although frequencies were normalized per 1000 tokens and a standardized preprocessing pipeline was applied, differences in document structure, level of detail, and country representation could influence the observed patterns. We also noted the more limited availability and comprehensiveness of FBDGs from low-income countries and cautioned that semantic differences may partly reflect broader inequalities in policy-development capacity. These structural differences are therefore not unrecognized sources of bias but part of the context in which our findings should be interpreted. Previous cross-country evidence has similarly documented substantial heterogeneity in FBDG implementation and emphasized the need for broader food-systems approaches, explicit implementation strategies, and adequate resources.³

We similarly agree that NLP cannot reconstruct the full cultural, agricultural, economic, and political context in which dietary guidelines are developed. Recommendations are embedded within food systems that differ substantially in affordability, food availability, infrastructure, environmental pressures, and equity, all of which may shape policy priorities and the feasibility of translating recommendations into practice.⁴ Our analysis did not attempt to capture these dimensions. We recognized that geography, cultural context, publication year, guideline–development processes, and food-system characteristics were not formally controlled for, and that income classification may act as a proxy for broader structural differences. NLP should therefore complement, rather than replace, contextual, qualitative, implementation, and policy analyses. Its contribution lies in providing a reproducible and scalable approach for identifying textual patterns across 98 national guidelines that would otherwise be difficult to examine systematically. Similar NLP and topic-modelling approaches have demonstrated their utility for systematically analysing large volumes of dietary-guideline text.⁵

The concern regarding our focus on dairy also warrants clarification. A focused analysis of one food group should not be interpreted as an attempt to define the totality of a healthy or sustainable diet. Dairy was selected a priori as the analytical domain in which to investigate whether computational linguistic methods could identify cross-country differences in dietary-policy messaging. During extraction, references to other food groups—including meat, fish, and eggs—were retained when they occurred within dairy-related sections to preserve the broader semantic context. Focused analysis of individual guideline components and whole-diet analyses therefore address different, complementary research questions.

We also welcome the opportunity to clarify the institutional collaboration reported in our Manuscript. The study received academic research support from Universidad San Sebastián, and the collaboration of the Scientific Committee of Dairy Products of the Chilean Dairy Consortium was transparently disclosed. This collaboration had no role in country or guideline selection, extraction of recommendations, development of the NLP pipeline, statistical or computational analyses, interpretation of results, or manuscript preparation. All analytical decisions and conclusions remained the responsibility of the authors. We agree that explicit disclosure of such roles is essential, particularly in nutrition research.

Finally, sustainability, food security, and the multiple burden of malnutrition are not external to our framework. Our Manuscript discusses food security, affordability, food-system infrastructure, environmental sustainability, and the persistent triple burden of malnutrition, particularly in the Americas. However, these dimensions were not measured as outcomes, and we therefore deliberately avoided conclusions beyond the textual evidence analysed.

The central distinction is between what NLP can describe and what it cannot establish. NLP methods can systematically extract and compare textual information, including thematic structures and patterns across policy-related corpora,⁶ but cannot by themselves determine whether one guideline is better, more culturally appropriate, more sustainable, or more effective than another. Our study addressed the former, not the latter.

We therefore share Rios-Castillo and Kanter's view that FBDGs must ultimately be evaluated within the food systems and populations for which they are designed. Rather than challenging our findings, this perspective reinforces their appropriate interpretation: computational analysis can reveal patterns in nutrition-policy discourse, whereas understanding why those patterns exist—and whether they translate into better diets and population health—requires complementary contextual and empirical research.

Contributors

AB: conceptualization, data curation, formal analysis, methodology, visualization, writing—original draft, and writing—review & editing.; HC-O: data curation, formal analysis conceptualization, visualization, writing—original draft, and writing—review & editing.; SD-A: Conceptualization, investigation, methodology, and supervision. RV: Investigation and visualization; RV investigation, visualization, writing—original draft, and writing—review & editing. All authors had full access to the data, read and approved the final version of the manuscript.

Declaration of interests

The authors declare no competing interests.