Context Matters - What Are Your Patients Really Saying?

09.13.2018 • Zach Childers, PhD

iStock Blog Photo

Precision in language analysis is an inherently complex and difficult goal. It is necessary, however, to know precisely what a patient is communicating as well as what can reasonably be inferred from the patients’ language in order to hear the voice of your patients and effectively improve their experiences.

The goal of natural language processing (NLP), when applied to comments from patient surveys or social media reviews, is to decipher patients’ own words about their care experiences consistently over large volumes of text. Language experts tap into two domains of linguistic knowledge for this task: Lexical semantics, which is the study of word meaning, and syntax, which is the study of grammatical structure.

Communication never occurs in a vacuum. Whether in spoken conversation, in a social media post or in a written survey response, the meaning of the words we use depend only partially on grammatical rules.

The meaning of the language we use is largely driven by context. In conversation, in social media or in surveys, our language use presupposes a context, which restricts the meaning of our utterances. For example, there is an implied restriction in most uses of universal quantification words like “every” and “all.” “Every room has a TV” usually means something like “Every room in this hospital has a TV.” Also, saying “I hated all the meals” would be consistent with meals being served in the care setting.

We interpret what we read or hear based on our presuppositions about the context. For example, the meaning of a sentence like “A truck is under the tree” can be very different depending on whether the context is Christmas morning or a logging work site.  

The context that supplements meaning also comes with its own restrictions. You are well within reasonable expectations to infer that a crime has been revealed from the utterance "Jane spilled the beans" if the context is testimony at a congressional hearing. That inference, however, would often be inappropriate if the context were a supermarket.

Contextual interpretation of language often goes beyond the difference between idiomatic and literal meaning. Context can also influence specific kinds of inferences we can make from a single utterance.

In the context of patient experience, "There was someone in my room when I arrived" might be a simple statement of fact or a comment on room privacy. If the context of experience is a hotel stay, the same sentence would involve a negative inference regarding cleaning scheduling, booking logistics or security.

This is the reason why a dictionary and a grammar are not enough to fully understand the intended meaning of an utterance. You must also know the context of an utterance and the nature of the relationships between ideas that might be expressed in an individual sentence.

This is also the reason why you need to have a domain-specific ontology in order to make sense of meaning relationships in specific contexts when you are analyzing language. An ontology is a structured network of categories or nodes with a variety of different logical relationships between these nodes. A domain-specific ontology, when built right on a large body of data to reflect idiosyncratic meaning relationships in that domain, is very large and complex.

The ontology on which we built the NarrativeDx patient experience platform undergoes continual expansion and refinement. Our ontology contains thousands of categorical nodes and many connected layers within a rich network of inferential relationships of language use in the context of patient experience. This allows you to understand that "Sugar levels were too high" is about meal content in the context of a description of diet or menus, but the same sentence would be about blood glucose in the context of treatments or admission. Similarly, our domain-specific ontology helps us identify that the adjective "positive" is a good thing in the context of nurse attitudes, but it is very often a bad thing in the context of medical test results.

With our patient experience-specific ontology, we help you understand the specific meaning behind your patients’ words in specific contexts of care experiences. Our hard work and expertise in building and growing our ontology allow our clients to get to the root causes of important issues and take targeted actions.

You can read our case study on how we helped HackensackUMC Palisades identify areas for improvement and determine the actions needed once they were able to more accurately hear their patients' voices. 

Read Case Study


Zach Childers, PhD

Zach Childers, PhD

Linguistics Engineer, NarrativeDx

Get ”Insights into Patient Experience“ delivered to your inbox