Relationships exist within contexts—who the kids are, what the job stress is, what anniversary you're approaching—and these details matter for communication. AI that tracks these named entities (people, places, events) helps you avoid forgetting crucial context that shapes what someone is actually dealing with.
Named Entity Recognition (NER) is a natural language processing technique that automatically identifies and classifies specific things mentioned in text: people, places, organizations, events, and other named entities. In the context of relationships, NER systems can automatically extract and track the key actors, incidents, and emotional events in your relational narratives, creating a structured understanding of your situation without requiring you to manually tag everything.
When you tell an AI about a conflict with your partner, you mention people ("my mom said..."), events ("when we went to that dinner"), timeframes ("last Tuesday"), and recurring issues. A basic chatbot threads together this information as it processes your words, but an NER-enabled system explicitly identifies and categorizes these entities, building a relational map.
The mechanism works through machine learning classifiers trained on annotated text. The model reads your conversation and labels each word or phrase by entity type: "Person," "Event," "Location," "Date," etc. When you describe a family dinner gone wrong, the NER system identifies: who attended (Person entities), when it occurred (Date), where (Location), and what specific comments triggered hurt (extracting the Event). It then links these into a coherent narrative structure.
This is more sophisticated than simple keyword search. The system understands that "my sister Rachel" and "Rachel" refer to the same entity even if mentioned differently. It recognizes that "the meeting with his parents" and "that awkward dinner" are the same event. Over time, with enough conversation, NER can build a knowledge graph—a visual map showing how different people, events, and issues in your life relate to each other.
Relationships are contextual webs. A current conflict with your partner might be connected to an unresolved issue with an in-law, a past betrayal by a friend, or a pattern established in your family of origin. When you're in the emotional middle of a conflict, you might not see these connections. An NER-enabled system can automatically identify recurring people and issues across conversations, surfacing patterns: "You've mentioned feeling unsupported by your partner in three separate conversations about different topics. Let's look at whether there's a deeper trust issue here."
Conversation analysis tools use NER to generate summaries that actually capture your relational landscape. Instead of vague notes like "we talked about conflict," the system generates: "You discussed tension with your partner about household division of labor, mentioning that your mother also criticized your division of housework. Is there a family-of-origin pattern here?"
NER also powers relationship memory tools. When you return to an AI coaching tool weeks later, the system can search not just for keywords but for entity-based relevance: "You asked about communication with your sister last month. Here's that conversation—is that relevant to today's question?"
NER is powerful but imperfect. It struggles with ambiguous pronouns: when you say "she told him that she was angry," the system must correctly assign "she" and "him" to the right people—a challenge called pronoun resolution. It also struggles with implicit entities (when you refer to someone as "my ex" without naming them), with entities that are concepts rather than named things ("the way my family handles conflict"), and with entities specific to your relationship (inside jokes or nicknames that aren't in training data).
NER trained on general text data might misclassify relationships: it might tag "the attachment theory discussion with my therapist" and miss that the key entity is the therapeutic relationship itself, not just the concept. The system needs to be fine-tuned on relationship communication patterns to work optimally.
There's also a privacy consideration: NER-enabled systems are tracking who appears in your conversations, which could raise consent questions if not handled carefully. If your coaching tool is sharing de-identified patterns across users ("many people mention their mother-in-law in conflict conversations"), it's extracting entities from intimate conversations.
NER works well with established therapy frameworks. Family systems therapy, for instance, focuses on patterns across family members and across time—exactly what NER can surface. Attachment theory analysis benefits from NER tracking how specific people (your partner, your parents, your ex) appear across different conversations and emotional contexts. The AI can prompt: "You mention your partner withdrawing during conflicts and your father being emotionally unavailable during your childhood. Are you seeing a pattern in how you experience intimacy?"
Try this: In your next conversation with ChatGPT or Claude about a relationship situation, explicitly ask the AI to identify all the people involved, what happened, when, and what the emotional core of the issue is. Then ask it to look back through your conversation and note if you've mentioned similar situations or people before (manually referencing past conversations you want it to consider). This gives you a sense of how NER-based analysis could surface hidden connections in your relational patterns.
Peri can explain this concept, give practical examples, help you decide whether it applies to your situation, or recommend a journey if appropriate.
Explore related journeys or tell Peri what you're working through.