Anastasia Kobzeva
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Anastasia Kobzeva has a total of 9 co-authors. Here are the top co-authors by number of shared publications:
Dave Kush
6 shared publicationsSuhas Arehalli
3 shared publications- Dave Kush
University of Toronto Scarborough, ON, Canada
2 shared publications - Tal Linzen
Department of Linguistics, New York University
2 shared publications - Charlotte Sant
Arctic University of Norway
1 shared publication - Myrte Vos
UiT The Arctic University of Norway
1 shared publication
Latest Publications
- Open Access
- 20 July 2025
- Published by Elsevier in Journal of Memory and Language
- Vol. 144
Learning filler-gap dependencies with neural language models: Testing island sensitivity in Norwegian and English
- 20 July 2025
- Published by Elsevier in Journal of Memory and Language
- Vol. 144
Open AccessLearning filler-gap dependencies with neural language models: Testing island sensitivity in Norwegian and English
- Open Access
- 13 March 2025
- Published by Taylor & Francis in Language Acquisition
- Vol. 33 (1) , 39-82
Acquiring constraints on filler-gap dependencies from structural collocations: Assessing a computational learning model of island-insensitivity in Norwegian
- 13 March 2025
- Published by Taylor & Francis in Language Acquisition
- Vol. 33 (1) , 39-82
Open AccessAcquiring constraints on filler-gap dependencies from structural collocations: Assessing a computational learning model of island-insensitivity in Norwegian
- Open Access
- 25 November 2024
- Published by Center for Open Science in PsyArXiv
Acquiring Constraints on Filler-Gap Dependencies from Structural Collocations: Assessing a Computational Learning Model of Island-Insensitivity in Norwegian
Children induce complex syntactic knowledge from their native language input. A long-standing discussion focuses on types of learning biases that help them arrive at correct generalization and solve induction problems posed by impoverished input. Studies employing computational models for learning specific language phenomena serve as testing grounds for evaluating types of biases required for successful acquisition. Recent work by Pearl and Sprouse (2013b) demonstrates that a distributional learner that tracks trigrams over structurally annotated input can acquire wh-filler-gap dependencies and island constraints on them in English. While intriguing, it is unclear yet whether a similar distributional learning model is a viable mechanism for learning island facts in other languages given the possibility of cross-linguistic variation. In this study, we explore whether a distributional learner can acquire wh- and relative clause filler-gap dependencies and island constraints in Norwegian from child-directed annotated text. We find that the proposed learning strategy can capture some patterns of island-insensitivity in Norwegian while failing to learn others due to a lack of relevant data in the input. Our findings suggest that given limited input data, a simple n-gram-based distributional learning over structured representations may not be sufficient to fully recover human-like knowledge of filler-gap dependency relations and island constraints cross-linguistically.
- 25 November 2024
- Published by Center for Open Science in PsyArXiv
Open AccessAcquiring Constraints on Filler-Gap Dependencies from Structural Collocations: Assessing a Computational Learning Model of Island-Insensitivity in Norwegian
Children induce complex syntactic knowledge from their native language input. A long-standing discussion focuses on types of learning biases that help them arrive at correct generalization and solve induction problems posed by impoverished input. Studies employing computational models for learning specific language phenomena serve as testing grounds for evaluating types of biases required for successful acquisition. Recent work by Pearl and Sprouse (2013b) demonstrates that a distributional learner that tracks trigrams over structurally annotated input can acquire wh-filler-gap dependencies and island constraints on them in English. While intriguing, it is unclear yet whether a similar distributional learning model is a viable mechanism for learning island facts in other languages given the possibility of cross-linguistic variation. In this study, we explore whether a distributional learner can acquire wh- and relative clause filler-gap dependencies and island constraints in Norwegian from child-directed annotated text. We find that the proposed learning strategy can capture some patterns of island-insensitivity in Norwegian while failing to learn others due to a lack of relevant data in the input. Our findings suggest that given limited input data, a simple n-gram-based distributional learning over structured representations may not be sufficient to fully recover human-like knowledge of filler-gap dependency relations and island constraints cross-linguistically.
- Open Access
- 14 October 2024
- Published by Wiley in Cognitive Science
- Vol. 48 (10) , e13501
Grammar and Expectation in Active Dependency Resolution: Experimental and Modeling Evidence from Norwegian
Filler‐gap dependency resolution is often characterized as an active process. We probed the mechanisms that determine where and why comprehenders posit gaps during incremental processing using Norwegian as our test language. First, we investigated why active filler‐gap dependency resolution is suspended inside island domains like embedded questions in some languages. Processing‐based accounts hold that resource limitations prevent gap‐filling in embedded questions across languages, while grammar‐based accounts predict that active gap‐filling is only blocked in languages where embedded questions are grammatical islands. In a self‐paced reading study, we find that Norwegian participants exhibit filled‐gap effects inside embedded questions, which are not islands in the language. The findings are consistent with grammar‐based, but not processing, accounts. Second, we asked if active filler‐gap processing can be understood as a special case of probabilistic ambiguity resolution within an expectation‐based framework. To do so, we tested whether word‐by‐word surprisal values from a neural language model could predict the location and magnitude of filled‐gap effects in our behavioral data. We find that surprisal accurately tracks the location of filled‐gap effects but severely underestimates their magnitude. This suggests either that mechanisms above and beyond probabilistic ambiguity resolution are required to fully explain active gap‐filling behavior or that surprisal values derived from a long short‐term memory neural language model are not good proxies for humans' incremental expectations during filler‐gap resolution.
- 14 October 2024
- Published by Wiley in Cognitive Science
- Vol. 48 (10) , e13501
Open AccessGrammar and Expectation in Active Dependency Resolution: Experimental and Modeling Evidence from Norwegian
Filler‐gap dependency resolution is often characterized as an active process. We probed the mechanisms that determine where and why comprehenders posit gaps during incremental processing using Norwegian as our test language. First, we investigated why active filler‐gap dependency resolution is suspended inside island domains like embedded questions in some languages. Processing‐based accounts hold that resource limitations prevent gap‐filling in embedded questions across languages, while grammar‐based accounts predict that active gap‐filling is only blocked in languages where embedded questions are grammatical islands. In a self‐paced reading study, we find that Norwegian participants exhibit filled‐gap effects inside embedded questions, which are not islands in the language. The findings are consistent with grammar‐based, but not processing, accounts. Second, we asked if active filler‐gap processing can be understood as a special case of probabilistic ambiguity resolution within an expectation‐based framework. To do so, we tested whether word‐by‐word surprisal values from a neural language model could predict the location and magnitude of filled‐gap effects in our behavioral data. We find that surprisal accurately tracks the location of filled‐gap effects but severely underestimates their magnitude. This suggests either that mechanisms above and beyond probabilistic ambiguity resolution are required to fully explain active gap‐filling behavior or that surprisal values derived from a long short‐term memory neural language model are not good proxies for humans' incremental expectations during filler‐gap resolution.
- Open Access
- 9 September 2024
- Published by Center for Open Science in PsyArXiv
Grammar and Expectation in Active Dependency Resolution: Experimental and Modeling Evidence from Norwegian
Filler-gap dependency resolution is often characterized as an active process. We probed the mechanisms that determine where and why comprehenders posit gaps during incremental processing using Norwegian as our test language. First, we investigated why active filler-gap dependency resolution is suspended inside island domains like embedded questions in some languages. Processing-based accounts hold that resource limitations prevent gap-filling in embedded questions across languages, while grammar-based accounts predict that active gap-filling is only blocked in languages where embedded questions are grammatical islands. In a self-paced reading study, we find that Norwegian participants exhibit filled-gap effects inside embedded questions, which are not islands in the language. The findings are consistent with grammar-based, but not processing, accounts. Second, we asked if active filler-gap processing can be understood as a special case of probabilistic ambiguity resolution within an expectation-based framework. To do so, we tested whether word-by-word surprisal values from a neural language model could predict the location and magnitude of filled-gap effects in our behavioral data. We find that surprisal accuratelytracks the location of filled-gap effects, but severely underestimates their magnitude. This suggests either that mechanisms above and beyond probabilistic ambiguity resolution are required to fully explain active gap-filling behavior or that surprisal values derived from LSTMs are not good proxies for humans' incremental expectations during filler-gap resolution.
- 9 September 2024
- Published by Center for Open Science in PsyArXiv
Open AccessGrammar and Expectation in Active Dependency Resolution: Experimental and Modeling Evidence from Norwegian
Filler-gap dependency resolution is often characterized as an active process. We probed the mechanisms that determine where and why comprehenders posit gaps during incremental processing using Norwegian as our test language. First, we investigated why active filler-gap dependency resolution is suspended inside island domains like embedded questions in some languages. Processing-based accounts hold that resource limitations prevent gap-filling in embedded questions across languages, while grammar-based accounts predict that active gap-filling is only blocked in languages where embedded questions are grammatical islands. In a self-paced reading study, we find that Norwegian participants exhibit filled-gap effects inside embedded questions, which are not islands in the language. The findings are consistent with grammar-based, but not processing, accounts. Second, we asked if active filler-gap processing can be understood as a special case of probabilistic ambiguity resolution within an expectation-based framework. To do so, we tested whether word-by-word surprisal values from a neural language model could predict the location and magnitude of filled-gap effects in our behavioral data. We find that surprisal accuratelytracks the location of filled-gap effects, but severely underestimates their magnitude. This suggests either that mechanisms above and beyond probabilistic ambiguity resolution are required to fully explain active gap-filling behavior or that surprisal values derived from LSTMs are not good proxies for humans' incremental expectations during filler-gap resolution.
- Open Access
- 26 April 2023
- Published by Center for Open Science in PsyArXiv
Neural Networks Can Learn Patterns of Island-insensitivity in Norwegian
Recent research suggests that Recurrent Neural Networks (RNNs) can capture abstract generalizations about filler-gap dependencies (FGDs) in English and so-called island constraints on their distribution (Wilcox et al., 2018; 2021). These results have been interpreted as evidence that it is possible, in principle, to induce complex syntactic knowledge from the input without domain-specific learning biases. However, the English results alone do not establish that island constraints were induced from distributional properties of the training data instead of simply reflecting architectural limitations independent of the input to the models. We address this concern by investigating whether such models can learn the distribution of acceptable FGDs in Norwegian, a language that is sensitive to fewer islands than English (Christensen, 1982). Results from five experiments show that Long Short-Term Memory (LSTM) RNNs can (i) learn that Norwegian FGD formation is unbounded, (ii) recover the island status of temporal adjunct and subject islands, and (iii) learn that Norwegian, unlike English, permits FGDs into two types of embedded questions. The fact that LSTM RNNs can learn cross-linguistic differences in island facts therefore strengthens the claim that RNN language models can induce the constraints from patterns in the input.
- 26 April 2023
- Published by Center for Open Science in PsyArXiv
Open AccessNeural Networks Can Learn Patterns of Island-insensitivity in Norwegian
Recent research suggests that Recurrent Neural Networks (RNNs) can capture abstract generalizations about filler-gap dependencies (FGDs) in English and so-called island constraints on their distribution (Wilcox et al., 2018; 2021). These results have been interpreted as evidence that it is possible, in principle, to induce complex syntactic knowledge from the input without domain-specific learning biases. However, the English results alone do not establish that island constraints were induced from distributional properties of the training data instead of simply reflecting architectural limitations independent of the input to the models. We address this concern by investigating whether such models can learn the distribution of acceptable FGDs in Norwegian, a language that is sensitive to fewer islands than English (Christensen, 1982). Results from five experiments show that Long Short-Term Memory (LSTM) RNNs can (i) learn that Norwegian FGD formation is unbounded, (ii) recover the island status of temporal adjunct and subject islands, and (iii) learn that Norwegian, unlike English, permits FGDs into two types of embedded questions. The fact that LSTM RNNs can learn cross-linguistic differences in island facts therefore strengthens the claim that RNN language models can induce the constraints from patterns in the input.
- Open Access
Comparing Island Effects for Different Dependency Types in Norwegian
Recent research suggests that island effects may vary as a function of dependency type, potentially challenging accounts that treat island effects as reflecting uniform constraints on all filler-gap dependency formation. Some authors argue that cross-dependency variation is more readily accounted for by discourse-functional constraints that take into account the discourse status of both the filler and the constituent containing the gap. We ran a judgment study that tested the acceptability of wh-extraction and relativization from nominal subjects, embedded questions (EQs), conditional adjuncts, and existential relative clauses (RCs) in Norwegian. The study had two goals: (i) to systematically investigate cross-dependency variation from various constituent types and (ii) to evaluate the results against the predictions of the Focus Background Conflict constraint (FBCC). Overall we find some evidence for cross-dependency differences across extraction environments. Most notably wh-extraction from EQs and conditional adjuncts yields small but statistically significant island effects, but relativization does not. The differential island effects are potentially consistent with the predictions of the FBCC, but we discuss challenges the FBCC faces in explaining finer-grained judgment patterns.
Open AccessComparing Island Effects for Different Dependency Types in Norwegian
Recent research suggests that island effects may vary as a function of dependency type, potentially challenging accounts that treat island effects as reflecting uniform constraints on all filler-gap dependency formation. Some authors argue that cross-dependency variation is more readily accounted for by discourse-functional constraints that take into account the discourse status of both the filler and the constituent containing the gap. We ran a judgment study that tested the acceptability of wh-extraction and relativization from nominal subjects, embedded questions (EQs), conditional adjuncts, and existential relative clauses (RCs) in Norwegian. The study had two goals: (i) to systematically investigate cross-dependency variation from various constituent types and (ii) to evaluate the results against the predictions of the Focus Background Conflict constraint (FBCC). Overall we find some evidence for cross-dependency differences across extraction environments. Most notably wh-extraction from EQs and conditional adjuncts yields small but statistically significant island effects, but relativization does not. The differential island effects are potentially consistent with the predictions of the FBCC, but we discuss challenges the FBCC faces in explaining finer-grained judgment patterns.
- Open Access
- 12 May 2022
- Published by Center for Open Science in PsyArXiv
LSTMs Can Learn Basic Wh- and Relative Clause Dependencies in Norwegian
One of the key features of natural languages is that they exhibit long-distance filler-gap dependencies (FGDs): In the sentence "What do you think the pilot sent __?" the wh-filler "what" is interpreted as the object of the verb "sent" across multiple words. The ability to establish FGDs is thought to require hierarchical syntactic structure. However, recent research suggests that recurrent neural networks (RNNs) without specific hierarchical bias can learn complex generalizations about wh-questions in English from raw text data (Wilcox et al. 2018; 2019). Across two experiments, we probe the generality of this result by testing whether a long short-term memory (LSTM) RNN model can learn basic generalizations about FGDs in Norwegian. Testing Norwegian allows us to assess whether previous results were due to distributional statistics of the English input or whether models can extract similar generalizations in languages with different syntactic distributions. We also test the model's performance on two different FGDs: wh-questions and relative clauses, allowing us to determine if the model learns abstract generalizations about FGDs that extend beyond a single construction type. Results from Experiment 1 suggest that the model expects fillers to be paired with gaps and that this expectation generalizes across different syntactic positions. Results from Experiment 2 suggest that the model's expectations are largely unaffected by the increased linear distance between the filler and the gap. Our findings provide support for the conclusion that LSTM RNN's ability to learn basic generalizations about FGDs is robust across dependency type and language.
- 12 May 2022
- Published by Center for Open Science in PsyArXiv
Open AccessLSTMs Can Learn Basic Wh- and Relative Clause Dependencies in Norwegian
One of the key features of natural languages is that they exhibit long-distance filler-gap dependencies (FGDs): In the sentence "What do you think the pilot sent __?" the wh-filler "what" is interpreted as the object of the verb "sent" across multiple words. The ability to establish FGDs is thought to require hierarchical syntactic structure. However, recent research suggests that recurrent neural networks (RNNs) without specific hierarchical bias can learn complex generalizations about wh-questions in English from raw text data (Wilcox et al. 2018; 2019). Across two experiments, we probe the generality of this result by testing whether a long short-term memory (LSTM) RNN model can learn basic generalizations about FGDs in Norwegian. Testing Norwegian allows us to assess whether previous results were due to distributional statistics of the English input or whether models can extract similar generalizations in languages with different syntactic distributions. We also test the model's performance on two different FGDs: wh-questions and relative clauses, allowing us to determine if the model learns abstract generalizations about FGDs that extend beyond a single construction type. Results from Experiment 1 suggest that the model expects fillers to be paired with gaps and that this expectation generalizes across different syntactic positions. Results from Experiment 2 suggest that the model's expectations are largely unaffected by the increased linear distance between the filler and the gap. Our findings provide support for the conclusion that LSTM RNN's ability to learn basic generalizations about FGDs is robust across dependency type and language.
- Open Access
- 15 November 2016
- Published by UiT The Arctic University of Norway in Poljarnyj vestnik
- Vol. 19
Four Ways to Tremble in Russian: A Case Study of a Radial Category
Many dictionaries and other lexicographic resources often describe near-synonyms using one another, which makes it difficult for a foreign language learner to make a choice between them. Although it is true that near-synonyms can be used when referring to the same state of affairs, they often involve different construals, insofar as they represent different ways of viewing the situation in question. Quite often, not all alternatives are possible in a given context, but it is not an easy task to pinpoint which alternative is possible in which context. In this study I consider a group of nearly synonymous verbs with the meaning ‘tremble’ in Russian and attempt to build a radial category network of meanings based on corpus data. I show that the network accommodates the similarities among the verbs, but at the same time also brings out the differences among them.
- 15 November 2016
- Published by UiT The Arctic University of Norway in Poljarnyj vestnik
- Vol. 19
Open AccessFour Ways to Tremble in Russian: A Case Study of a Radial Category
Many dictionaries and other lexicographic resources often describe near-synonyms using one another, which makes it difficult for a foreign language learner to make a choice between them. Although it is true that near-synonyms can be used when referring to the same state of affairs, they often involve different construals, insofar as they represent different ways of viewing the situation in question. Quite often, not all alternatives are possible in a given context, but it is not an easy task to pinpoint which alternative is possible in which context. In this study I consider a group of nearly synonymous verbs with the meaning ‘tremble’ in Russian and attempt to build a radial category network of meanings based on corpus data. I show that the network accommodates the similarities among the verbs, but at the same time also brings out the differences among them.