Skip to main content

Exclusive Q&A: Walmart combines human, AI insight for Spanish search


Walmart is utilizing advanced artificial intelligence (AI) capability and employee expertise to streamline digital product discovery for Spanish-speaking customers.

Leonardo Lezcano, engineering manager – Spanish experience lead, Walmart, recently spoke with Chain Store Age about the discount giant’s new in-house-developed Spanish translation solution.

[Read more: Walmart expands online Spanish search capability]

What made Walmart first decide to offer expanded Spanish-language search? 

We started noticing an influx of Spanish searches from our customers as COVID-19 pandemic restrictions increased, and in-store customers began shifting to online shopping. At the height of the pandemic, Spanish queries across the Walmart app and website increased more than five times their pre-pandemic levels.

This rapid shift in behavior highlighted the need to provide these customers with a Spanish-first online shopping experience. We quickly mobilized to determine how to address their needs, and were able to roll out a new product with basic features, i.e., a minimum viable product (MVP), a few months later. 

How did Walmart determine an in-house solution would be the best option? 

This solution was built in-house by Walmart Global Tech associates, with the Spanish Experience team leading the initiative with support from the U.S. Omni Tech Customer Experiences Search & Personalization team. 

The most recent generic machine translation (MT) models provide a good starting point to lift relevance by translating the Spanish queries to English before delivering them to the search engine. However, challenges specific to the e-commerce domain make generic MT an impractical solution.

The challenges include non-translatable entities, like "guerrero" ("warrior" in English), which should not be translated when referring to the tortilla brand; ambiguous queries like ‘calabaza’ which means both ‘pumpkin’ and ‘butternut squash’ in Spanish; and mixed language queries, colloquially known as Spanglish. Hispanic customers in the U.S. frequently mix Spanish and English in the same query, such as inputting ‘cake de fresa’ for ‘strawberry cake.’  

In addition, there is cross-language ambiguity While in English, "pan" relates to cookware, the same morpheme means "bread" in Spanish. There is also the issue of Spanish dialects. With Hispanic immigration in U.S. coming from over 20 Spanish-speaking countries, different queries sometimes converge to the same intention. For example, both "aguacate" from Cuba and "palta" from Chile and Peru mean "avocado.” 

And of course, for online stores serving millions of customers, the real-time speed of the translation is as important as its accuracy. 

Our Spanish Experience team includes natural language processing (NLP) experts with Hispanic and e-commerce backgrounds. The solution built in-house uses techniques known as fine-tuningadjusting parameters of a model to fit within certain observationsand domain adaptiontraining a neural network with different target data distributions to adapt to another model—to transform the translation models into a suitable solution that can be used to address challenges specific to the e-commerce domain.

Can you give a description of what NLP technology is and how it enables in-depth Spanish language search on 

Natural language processing (NLP) includes a broad spectrum of techniques we apply at Walmart Search. Query classification to identify the product type that satisfies the customer intention the best, as well as entity recognition to spot, in the query text, the sizes, colors, product lines, and other facets that the user may be interested in, are two examples of NLP techniques that we leverage.

In addition to those, our NLP at includes elements of Cross Lingual Information Retrieval (CLIR) and Neural Machine Translation to detect queries in Spanish and translate them to English to get relevant results.

Thanks to the growing awareness of the support for Spanish search throughout our Hispanic customers, and the organic increase in the frequency of Spanish terms in our search box, our NLP pipelines are constantly detecting new Spanish queries with low conversion rates that can benefit from a better translation.  

For example, we’ve started noticing that the Cuban ‘frutabomba’ already renders as good results as the more generic "papaya", and both ‘piscina’ and ‘alberca’ return pools. Moreover, recognizing that a person is searching for popcorn even if they type palomita de maiz, rositas de maiz, or pororo despite regional variations is key to ensuring that all of our customers are able to search for items in a way that feels natural to them.

The recently released opt-in/opt-out feature will also provide valuable feedback on this regard, as customers can explicitly validate or dislike translations. 

How have customers responded? 

Since we released the first versions of Spanish search at we have seen a growing frequency of Spanish queries in our search box, showing that the Hispanic community is organically embracing this convenient capacity of searching in native language.  

Can you discuss any future plans for this search translation technology? 

This solution is continuously evolving with the goal of providing a full end-to-end Spanish language experience. We will continue to adapt it to our customers’ needs by releasing additional features in the coming months. 


This ad will auto-close in 10 seconds