Reading a conversation by building it up, one thought at a time
AWAI splits each utterance into thoughts and analyzes the conversation by accumulating their functions and connections. This approach is grounded in international standards for dialogue acts and discourse relations, and in research into memory and cognition.
A single utterance as an example
Three friends are planning a trip to Tokyo. Hayato speaks Japanese, Emma English, and Diego Spanish (Hayato's and Diego's lines are shown in English translation).
- Diego
And for the evening, can we hit some izakayas in a small alleyway? None of those tourist traps, please.
- Hayato
Omoide Yokocho in Shinjuku is perfect for that, the alleyway vibe is the best.Oh, and what's your budget looking like? Prices can vary a lot.
- Emma
Let's keep it mid-range — splurge on one amazing sushi night, and stay cheap and local the rest.
AWAI reads Hayato's utterance as two thoughts.
Omoide Yokocho in Shinjuku is perfect for that, the alleyway vibe is the best.
- Function
- Request / proposal. It also serves a turn-keeping function.
- Connection
- Similarity to Diego's proposal of a back-alley izakaya
Oh, and what's your budget looking like? Prices can vary a lot.
- Function
- Question and Inform (one thought carrying two functions)
- Connection
- Expansion of Diego's proposal
- Points
- “Confirming the budget” and “explaining the wide range of price points”
Emma's reply is read as a thought with two functions, Response and Request / proposal, and it is connected to the question in Thought 2 by dependence.
This example is the actual analysis from our public meeting demo.
What has been read becomes the context for what comes next
Each extracted thought is retained in a form that allows semantic similarity to be measured (a 3072-dimensional vector). Together with their functions and connections, these thoughts accumulate as the conversation progresses, becoming the context for reading the next utterance: the flow shared in the room, past thoughts recalled by association, and the preconceptions each speaker brings.
AWAI's analysis operates on these accumulated thoughts and connections.
Three pillars
- FunctionsWhat a thought in an utterance is doing in the conversation. Read using the 63 communicative functions of ISO 24617-2.
- ConnectionsHow thoughts are connected to one another. Read using the 19 logical relations built on ISO 24617-8 (DR-core) and response dependence.
- ContextHow the conversation so far is maintained while thoughts are summarized and their connections are sought. Based on common ground, working memory, schemas, and ACT-R.
Functions and connections are what is retained; context is how the conversation is read.
Five products stand on one analysis
Every AWAI product runs the same analysis underneath. What changes is only what gets fed into it, which is why they all read the same way.
In the space between thoughts, structure emerges.
In the space between minds, shared understanding is born.
"AWAI" is an archaic Japanese reading of "between." In the uncertainty before things are settled, possibility lives — AWAI makes that space visible.