Sensemaking
Interpreting language involves word structure, grammar, meaning, and context. The source uses these levels to explain why making sense of text requires more than identifying individual words.
THE IDEA
Sensemaking
Teaching computers to interpret language makes familiar human skills explicit. Natural language processing examines how words are formed, how they relate within sentences, and how their meaning changes with context. Figure 1.2 presents a generalized text-mining process.
The source distinguishes these levels of analysis:
- Phonology: interpreting sounds within and across words.
- Morphology: identifying morphemes, the smallest grammatical units.
- Lexicology: studying words and their meanings.
- Syntax: identifying grammatical relationships within a sentence.
- Semantics: interpreting meaning, including resolving ambiguous words.
- Discourse: connecting meaning across sentences.
- Pragmatics: interpreting meaning in its situation or context.
FIGURE 1.2 Generalized Text Mining Process.
These are related aspects of language, rather than mandatory stages that every system performs in this order. Human readers often handle them without conscious attention; designing a language-processing system requires deciding which distinctions matter for its task.
When to use it
Teaching a chatbot to understand and respond to user queries by processing language through the sensemaking steps.
Analyzing legal documents to extract relevant clauses and relationships using syntactic and discourse analysis.
Building a recommendation system that interprets user reviews by disambiguating sentiment and context.
Developing a translation system that goes beyond word-for-word translation by incorporating pragmatics and discourse.
FROM THEORY TO EVERYDAY LIFE
In practice
Text Mining Customer Reviews
A company analyzes customer reviews. Word forms and sentence structure help identify what is being discussed; context helps distinguish praise from a sarcastic complaint. Spoken reviews also require interpreting sounds. The task determines which language-processing methods are needed.
A MOMENT TO REFLECT
Put it to work.
How does the sensemaking process in natural language processing mirror or differ from human learning?
What challenges arise when automating the more subtle steps like discourse and pragmatics?
Can you identify a decision that depends on interpreting context across several sentences, rather than counting individual words?
CONTINUE THE CONVERSATION
Source texts
- Evidence-Based Decision-Making How to Leverage Available Data and Avoid Cognitive Biases (Andrew D. Banasiewicz)