Yuqing Zhang
@yuqing0304.bsky.social
Come check out our poster at EMNLP 2025 tomorrow (Fri, Nov 7, 12:30 - 13:30)!
Paper: aclanthology.org/2025.finding...
Code: github.com/yuqing0304/N...
Many thanks to my wonderful co-authors @ecesuurker.bsky.social, Tessa Verhoef, @gboleda.bsky.social, and @arianna-bis.bsky.social
Paper: aclanthology.org/2025.finding...
Code: github.com/yuqing0304/N...
Many thanks to my wonderful co-authors @ecesuurker.bsky.social, Tessa Verhoef, @gboleda.bsky.social, and @arianna-bis.bsky.social
November 6, 2025 at 4:16 PM
Come check out our poster at EMNLP 2025 tomorrow (Fri, Nov 7, 12:30 - 13:30)!
Paper: aclanthology.org/2025.finding...
Code: github.com/yuqing0304/N...
Many thanks to my wonderful co-authors @ecesuurker.bsky.social, Tessa Verhoef, @gboleda.bsky.social, and @arianna-bis.bsky.social
Paper: aclanthology.org/2025.finding...
Code: github.com/yuqing0304/N...
Many thanks to my wonderful co-authors @ecesuurker.bsky.social, Tessa Verhoef, @gboleda.bsky.social, and @arianna-bis.bsky.social
Most studies on semantic change rely on corpus analyses or experiments with humans. But corpora can’t reveal causal mechanisms, and experiments are hard to scale.
Why do words shift meaning and how do communicative needs shape lexicons? To explore that, we built a simulation: NeLLCom-Lex.
Why do words shift meaning and how do communicative needs shape lexicons? To explore that, we built a simulation: NeLLCom-Lex.
November 6, 2025 at 3:39 PM
Most studies on semantic change rely on corpus analyses or experiments with humans. But corpora can’t reveal causal mechanisms, and experiments are hard to scale.
Why do words shift meaning and how do communicative needs shape lexicons? To explore that, we built a simulation: NeLLCom-Lex.
Why do words shift meaning and how do communicative needs shape lexicons? To explore that, we built a simulation: NeLLCom-Lex.
NeLLCom-Lex consists of a two-phase learning setup:
1) Supervised learning (SL): agents “learn to speak” an existing human lexicon.
2) Reinforcement learning (RL): they then communicate and adapt it to communicative needs.
Question: Can such agents start evolving their lexicon like humans do?
1) Supervised learning (SL): agents “learn to speak” an existing human lexicon.
2) Reinforcement learning (RL): they then communicate and adapt it to communicative needs.
Question: Can such agents start evolving their lexicon like humans do?
November 6, 2025 at 3:39 PM
NeLLCom-Lex consists of a two-phase learning setup:
1) Supervised learning (SL): agents “learn to speak” an existing human lexicon.
2) Reinforcement learning (RL): they then communicate and adapt it to communicative needs.
Question: Can such agents start evolving their lexicon like humans do?
1) Supervised learning (SL): agents “learn to speak” an existing human lexicon.
2) Reinforcement learning (RL): they then communicate and adapt it to communicative needs.
Question: Can such agents start evolving their lexicon like humans do?
We use a color naming task 🎨
Each agent must identify a target color among distractors.
In far contexts, general words suffice.
In close contexts, more informative words are necessary.
This is where pragmatic adaptation comes into play: humans choose more informative words when context demands it.
Each agent must identify a target color among distractors.
In far contexts, general words suffice.
In close contexts, more informative words are necessary.
This is where pragmatic adaptation comes into play: humans choose more informative words when context demands it.
November 6, 2025 at 3:35 PM
We use a color naming task 🎨
Each agent must identify a target color among distractors.
In far contexts, general words suffice.
In close contexts, more informative words are necessary.
This is where pragmatic adaptation comes into play: humans choose more informative words when context demands it.
Each agent must identify a target color among distractors.
In far contexts, general words suffice.
In close contexts, more informative words are necessary.
This is where pragmatic adaptation comes into play: humans choose more informative words when context demands it.
But can neural agents learn to be pragmatic and efficient?
It turns out yes! Agents exposed to context adapt. They use more informative words in harder contexts (β). Interaction with context leads to richer lexicons (|W|) with lower system-level informativeness (I_L) and moderate drift (D_L).
It turns out yes! Agents exposed to context adapt. They use more informative words in harder contexts (β). Interaction with context leads to richer lexicons (|W|) with lower system-level informativeness (I_L) and moderate drift (D_L).
November 6, 2025 at 3:21 PM
But can neural agents learn to be pragmatic and efficient?
It turns out yes! Agents exposed to context adapt. They use more informative words in harder contexts (β). Interaction with context leads to richer lexicons (|W|) with lower system-level informativeness (I_L) and moderate drift (D_L).
It turns out yes! Agents exposed to context adapt. They use more informative words in harder contexts (β). Interaction with context leads to richer lexicons (|W|) with lower system-level informativeness (I_L) and moderate drift (D_L).
We then expose agents to varying communicative needs.
Agents facing more challenging contexts (AllClose) develop stronger context sensitivity and reshape word meanings like "mint" to denote narrower color regions.
Agents facing more challenging contexts (AllClose) develop stronger context sensitivity and reshape word meanings like "mint" to denote narrower color regions.
November 6, 2025 at 3:19 PM
We then expose agents to varying communicative needs.
Agents facing more challenging contexts (AllClose) develop stronger context sensitivity and reshape word meanings like "mint" to denote narrower color regions.
Agents facing more challenging contexts (AllClose) develop stronger context sensitivity and reshape word meanings like "mint" to denote narrower color regions.
Moreover, communicating in harder contexts (AllClose) allows agents to develop more efficient lexicons, achieving high accuracy with a smaller vocabulary.
November 6, 2025 at 3:18 PM
Moreover, communicating in harder contexts (AllClose) allows agents to develop more efficient lexicons, achieving high accuracy with a smaller vocabulary.
💡 Takeaway:
Neural agents can reproduce human-like pragmatic behavior and develop efficient lexicons.
Our framework, NeLLCom-Lex, provides a tool to simulate semantic change, offering insights into how word meanings broaden, narrow, or shift under new communicative pressures.
Neural agents can reproduce human-like pragmatic behavior and develop efficient lexicons.
Our framework, NeLLCom-Lex, provides a tool to simulate semantic change, offering insights into how word meanings broaden, narrow, or shift under new communicative pressures.
November 6, 2025 at 2:30 PM
💡 Takeaway:
Neural agents can reproduce human-like pragmatic behavior and develop efficient lexicons.
Our framework, NeLLCom-Lex, provides a tool to simulate semantic change, offering insights into how word meanings broaden, narrow, or shift under new communicative pressures.
Neural agents can reproduce human-like pragmatic behavior and develop efficient lexicons.
Our framework, NeLLCom-Lex, provides a tool to simulate semantic change, offering insights into how word meanings broaden, narrow, or shift under new communicative pressures.