Leonie Weissweiler
@weissweiler.bsky.social
Postdoc at Uppsala University Computational Linguistics with Joakim Nivre
PhD from LMU Munich, prev. UT Austin, Princeton, @ltiatcmu.bsky.social, Cambridge
computational linguistics, construction grammar, morphosyntax
leonieweissweiler.github.io
PhD from LMU Munich, prev. UT Austin, Princeton, @ltiatcmu.bsky.social, Cambridge
computational linguistics, construction grammar, morphosyntax
leonieweissweiler.github.io
Oops, here are the three caused-motion examples that were meant to go in the first post:
November 19, 2025 at 6:20 PM
Oops, here are the three caused-motion examples that were meant to go in the first post:
We find that a range of models indeed struggle with this, but Gemma 27B solves it almost perfectly!
In grey are cases where the model struggles to answer both questions, in red the cases where it would have needed to reply on the caused-motion semantics.
🧵6/7
In grey are cases where the model struggles to answer both questions, in red the cases where it would have needed to reply on the caused-motion semantics.
🧵6/7
November 19, 2025 at 1:57 PM
We find that a range of models indeed struggle with this, but Gemma 27B solves it almost perfectly!
In grey are cases where the model struggles to answer both questions, in red the cases where it would have needed to reply on the caused-motion semantics.
🧵6/7
In grey are cases where the model struggles to answer both questions, in red the cases where it would have needed to reply on the caused-motion semantics.
🧵6/7
Once we have manually annotated the final dataset in this way, we return to the original question and test if LLMs find the third question below easier than the second, which would indicate difficulties in making use of the semantics of the caused-motion construction.
🧵5/7
🧵5/7
November 19, 2025 at 1:57 PM
Once we have manually annotated the final dataset in this way, we return to the original question and test if LLMs find the third question below easier than the second, which would indicate difficulties in making use of the semantics of the caused-motion construction.
🧵5/7
🧵5/7
But this leaves many FPs, and filtering them by hand would be expensive.
To reduce this cost, we use few-shot prompt-based filtering, which greatly reduces the number of FPs that our human annotator will have to sift through, and therefore the annotation cost.
🧵3/7
To reduce this cost, we use few-shot prompt-based filtering, which greatly reduces the number of FPs that our human annotator will have to sift through, and therefore the annotation cost.
🧵3/7
November 19, 2025 at 1:57 PM
But this leaves many FPs, and filtering them by hand would be expensive.
To reduce this cost, we use few-shot prompt-based filtering, which greatly reduces the number of FPs that our human annotator will have to sift through, and therefore the annotation cost.
🧵3/7
To reduce this cost, we use few-shot prompt-based filtering, which greatly reduces the number of FPs that our human annotator will have to sift through, and therefore the annotation cost.
🧵3/7
They are all instances of the so-called caused-motion construction, and collecting enough instances for testing was a challenge, given its rarity!
To construct our dataset, we first create a dependency filter based on the syntactic side of the construction.
🧵2/7
To construct our dataset, we first create a dependency filter based on the syntactic side of the construction.
🧵2/7
November 19, 2025 at 1:57 PM
They are all instances of the so-called caused-motion construction, and collecting enough instances for testing was a challenge, given its rarity!
To construct our dataset, we first create a dependency filter based on the syntactic side of the construction.
🧵2/7
To construct our dataset, we first create a dependency filter based on the syntactic side of the construction.
🧵2/7
📢Out now in NEJLT!📢
In each of these sentences, a verb that doesn't usually encode motion is being used to convey that an object is moving to a destination.
Given that these usages are rare, complex, and creative, we ask:
Do LLMs understand what's going on in them?
🧵1/7
In each of these sentences, a verb that doesn't usually encode motion is being used to convey that an object is moving to a destination.
Given that these usages are rare, complex, and creative, we ask:
Do LLMs understand what's going on in them?
🧵1/7
November 19, 2025 at 1:57 PM
📢Out now in NEJLT!📢
In each of these sentences, a verb that doesn't usually encode motion is being used to convey that an object is moving to a destination.
Given that these usages are rare, complex, and creative, we ask:
Do LLMs understand what's going on in them?
🧵1/7
In each of these sentences, a verb that doesn't usually encode motion is being used to convey that an object is moving to a destination.
Given that these usages are rare, complex, and creative, we ask:
Do LLMs understand what's going on in them?
🧵1/7
📢Life update📢
🥳I'm excited to share that I've started as a postdoc at Uppsala University NLP @uppsalanlp.bsky.social, working with Joakim Nivre on topics related to constructions and multilinguality!
🙏Many thanks to the Walter Benjamin Programme of the DFG for making this possible.
🥳I'm excited to share that I've started as a postdoc at Uppsala University NLP @uppsalanlp.bsky.social, working with Joakim Nivre on topics related to constructions and multilinguality!
🙏Many thanks to the Walter Benjamin Programme of the DFG for making this possible.
September 15, 2025 at 3:10 PM
📢Life update📢
🥳I'm excited to share that I've started as a postdoc at Uppsala University NLP @uppsalanlp.bsky.social, working with Joakim Nivre on topics related to constructions and multilinguality!
🙏Many thanks to the Walter Benjamin Programme of the DFG for making this possible.
🥳I'm excited to share that I've started as a postdoc at Uppsala University NLP @uppsalanlp.bsky.social, working with Joakim Nivre on topics related to constructions and multilinguality!
🙏Many thanks to the Walter Benjamin Programme of the DFG for making this possible.
✨New paper ✨
RoBERTa knows the difference between "so happy that you're here", "so certain that I'm right" and "so happy that I cried"!
Exciting result (and more) from Josh Rozner along with @coryshain.bsky.social, @kmahowald.bsky.social and myself, go check it out!
RoBERTa knows the difference between "so happy that you're here", "so certain that I'm right" and "so happy that I cried"!
Exciting result (and more) from Josh Rozner along with @coryshain.bsky.social, @kmahowald.bsky.social and myself, go check it out!
March 12, 2025 at 2:44 PM
✨New paper ✨
RoBERTa knows the difference between "so happy that you're here", "so certain that I'm right" and "so happy that I cried"!
Exciting result (and more) from Josh Rozner along with @coryshain.bsky.social, @kmahowald.bsky.social and myself, go check it out!
RoBERTa knows the difference between "so happy that you're here", "so certain that I'm right" and "so happy that I cried"!
Exciting result (and more) from Josh Rozner along with @coryshain.bsky.social, @kmahowald.bsky.social and myself, go check it out!
✨New paper✨
Linguistic evaluations of LLMs often implicitly assume that language is generated by symbolic rules.
In a new position paper, @adelegoldberg.bsky.social, @kmahowald.bsky.social and I argue that languages are not Lego sets, and evaluations should reflect this!
arxiv.org/pdf/2502.13195
Linguistic evaluations of LLMs often implicitly assume that language is generated by symbolic rules.
In a new position paper, @adelegoldberg.bsky.social, @kmahowald.bsky.social and I argue that languages are not Lego sets, and evaluations should reflect this!
arxiv.org/pdf/2502.13195
February 20, 2025 at 3:06 PM
✨New paper✨
Linguistic evaluations of LLMs often implicitly assume that language is generated by symbolic rules.
In a new position paper, @adelegoldberg.bsky.social, @kmahowald.bsky.social and I argue that languages are not Lego sets, and evaluations should reflect this!
arxiv.org/pdf/2502.13195
Linguistic evaluations of LLMs often implicitly assume that language is generated by symbolic rules.
In a new position paper, @adelegoldberg.bsky.social, @kmahowald.bsky.social and I argue that languages are not Lego sets, and evaluations should reflect this!
arxiv.org/pdf/2502.13195