"What does this mean?"
Yuki showed Mira's note. It read "Meaning beyond bits."
Aoi nodded. "Good question. Shannon's information theory handles symbol transmission. But not meaning."
"Doesn't handle meaning?" Yuki was surprised.
"Shannon himself acknowledged it. 'The semantic problem is irrelevant to engineering.'"
Riku interrupted. "But if meaning doesn't get through, communication is meaningless."
"Exactly. That's why semantic communication is getting attention recently."
Aoi drew a diagram on the whiteboard. Sender, channel, receiver.
"In Shannon's model, if bits transmit accurately, it's success. But in human communication, the same bit sequence can have different meanings depending on context."
"For example?" Yuki asked.
Riku answered. "The word 'sick'. Can mean good or bad."
"Perfect example," Aoi acknowledged. "Same symbol sequence, but meaning reverses with context."
"So information theory is useless?" Yuki asked anxiously.
"No, it needs extension. Consider knowledge shared between sender and receiver."
Mira drew a new diagram. Two overlapping circles.
"Common ground," Aoi explained. "Knowledge, experience, culture shared by sender and receiver. This determines meaning."
"The larger the common ground, the more you can convey with fewer bits?" Yuki inferred.
"Exactly. Couples can converse with just 'that' and 'it' because they have huge common ground."
Riku thought. "So with people you just met, you need lots of explanation?"
"Yes. Because common ground is small, you need to explicitly send information."
Yuki looked at her notebook. "Is this related to conditional entropy?"
Aoi's eyes lit up. "Sharp. Exactly. H(message|context). Knowing context reduces message uncertainty."
"So hearing 'sick' without context, you don't understand the meaning," Riku summarized.
"But," Yuki said. "Context is also information, right? How do you convey it?"
"Good question. Actually, much context is implicitly shared. Physical environment, past conversations, social norms."
Mira showed a note. "Pragmatics - language in use"
"Pragmatics," Aoi translated. "How language is actually used. Not just formal meaning, but context of use."
Riku gave an example. "'Pass the salt' is a command, but sometimes takes question form."
"Indirect speech act. Surface form and actual intent differ."
Yuki was confused. "So how do you measure information content?"
"Difficult problem. Can't measure by bit count. Need to consider how much receiver's knowledge state changes."
Aoi wrote a new equation. "ΔK = K_after - K_before. Amount of knowledge change."
"But quantifying knowledge is hard."
"That's why semantic information theory is still developing. But with machine learning progress, meaning representation is becoming possible."
"Representation learning?" Yuki asked.
"Embed words and sentences in vector space. Semantically close words are also close in vector space."
Mira drew a complex diagram. Points scattered in multidimensional space.
"This is semantic space," Aoi explained. "'King' minus 'man' plus 'woman' becomes close to 'queen'."
Riku was amazed. "You can compute meaning?"
"To some extent. Not perfect, but useful."
Yuki pondered. "So in the future, we can compress meaning?"
"Theoretically possible. For example, saying 'the matter we discussed in yesterday's meeting' can reference long context briefly."
"Like a pointer," Riku said.
"Exactly. Common ground is a pointer to memory. References already stored information."
Yuki laughed. "Human conversation is efficient."
"Very. But that efficiency depends on common ground. That's why cross-cultural communication is difficult."
Mira said quietly. "Translation loses meaning"
"Translation loses meaning," Aoi nodded. "Conveying cultural context across languages is difficult."
"So it's not just noise," Yuki summarized. "Lack of common ground also blocks communication."
"Exactly. Reasons for not getting through aren't just technical problems. Also problems of meaning."
The four quietly contemplated the layers of meaning behind words. Information cannot be measured by bits alone.