SynFlow: A Multidimensional Diachronic Semantic Analysis Toolkit
An open-source tool that breaks down how a word's meaning changed, not just that it changed
SynFlow is an open-source toolkit that tracks lexical semantic change across multiple linguistic angles, such as syntax, morphology, and constructional patterns, instead of collapsing everything into a single vector-based score. A case study on the German adjective viral shows how its shift from 'virus-related' to 'going viral online' shows up consistently across several linguistic dimensions at once. On the SemEval-2020 Task 1 benchmark, its representations ranked above many neural-network-based systems while staying interpretable.
What they did
- Vector-space methods for lexical semantic change (LSC) can flag that a word changed but rarely reveal which aspects of its usage actually drove that change
- SynFlow turns dependency-based co-occurrences, morphological features, constructional configurations, and externally derived representations like Frame Semantics into period-specific distributions processed through one shared workflow
- It supports multiple distance measures (cosine distance, Jensen-Shannon divergence, total variation distance/TVD), permutation testing for statistical significance, and value-level decomposition of change scores
- An incremental clustering feature built on word2vec embeddings helps distinguish simple lexical replacement from broader thematic shifts over time
- In the case study, the German adjective viral shows a shift around 2011-2015 from adjectival to adverbial use (as in viral gehen, 'to go viral'), visible consistently across syntactic slots, lexical fillers, constructional configurations, and morphological features

| Type | Dimension | Example values | Interpretation |
|---|---|---|---|
| Slot type | SlotType | chi_nsubj, chi_obj | grammatical profile |
| Slot filler | FILLER[chi_obj] | cake | lexical preferences |
| Construction | Construction | [chi_nsubj + chi_obj] | multi-slot configurations |
| Feature type | FeatureType | Number, Tense | morphological profile |
| Feature | FEAT[Number] | Sing, Plur | variation within feature |

| Representation | Subtask 1 | Subtask 2 |
|---|---|---|
| Slot filler | 4th /28 | 12th /28 |
| Frame Semantics | 9th /28 | 9th /28 |

Why it matters
Researchers previously had to use separate tools for syntactic profiling, morphological analysis, and vector-based semantic change detection; SynFlow unifies these into one workflow, making analysis faster and results easier to interpret. It also demonstrated practical runtime on a large historical corpus (20 periods, over 116,000 parsed files) using just a consumer laptop, showing it's ready for real research use.

Terms in this paper
- Lexical semantic change (LSC) · the study of how a word's meaning shifts over time
- Total variation distance (TVD) · a statistical measure of how different two probability distributions are
- Jensen-Shannon divergence (JSD) · a symmetric statistical measure of the difference between two probability distributions
- Permutation testing · a statistical method that shuffles data randomly to check whether an observed change could have occurred by chance
- Frame Semantics · a linguistic theory that describes word meaning in terms of roles within a structured situational scenario

Original abstract (English)
Lexical semantic change (LSC) is commonly modelled through vector-space representations, but these approaches often provide limited insight into which aspects of usage are changing. Diachronic corpus research instead examines interpretable dimensions such as syntactic behaviour, morphology, and constructional patterns, but typically through separate analytical workflows. We present SynFlow, an open-source toolkit for multidimensional diachronic analysis of linguistic usage. SynFlow converts linguistic observations into period-specific distributions and applies a shared workflow across dependency-based co-occurrences, morphological features, constructional configurations, and externally derived representations such as Frame Semantics. It supports different distance measures, together with value-level decomposition, statistical testing, and incremental clustering of lexical fillers. We demonstrate SynFlow through a qualitative case study of the German adjective viral, showing how a single semantic development is reflected across syntactic, lexical, constructional, and morphological dimensions. We further report previously published results on SemEval-2020 Task 1 to situate the performance of these representations relative to existing lexical semantic change detection systems.
Read on arXivLatest papers
- LoRA-GA$^2$: Low Rank Adaptation with Multi-step Gradient Adaptive AlignmentPeeking at a few early training gradients before fine-tuning starts to set up LoRA smarter
- Remember, Verify, or Ask? Cross-Family Evaluation of Memory Commitment in LLM AgentsAI assistants would rather double-check facts than ask you a question, even when asking is the right call
- Robust Incomplete Multimodal Sentiment Analysis via Iterative Proxy CorrectionWhen text input is missing or broken, this AI doesn't guess once and move on—it revises its guess step by step to read emotions more reliably
- Generating Diverse Personas for User Simulators to Test Interview Dialogue SystemsTo test interview-style chatbots you need many different fake users, so this work has an LLM automatically generate those fake user personalities
- Rethinking Patch Based Multivariate Time Series Forecasting with Semantic Structured PartitioningA new way to slice time series into meaningful chunks instead of arbitrary equal-length pieces
- Reliable Financial Named Entity Recognition under Domain ShiftAn AI's confidence trained on formal filings turns unreliable once it reads tweets
- Bringing analytic rigor to agentic AI for science: The Brain Researcher platform for neuroimaging data analysisA system that makes AI show its work when analyzing brain-imaging data, not just deliver an answer
- GenMatch: An End-to-End Generative Matching Framework for Micro-View Order-Dispatching in Ride-HailingDiDi replaced its multi-step ride-hailing dispatch pipeline with one generative model and saw real-world gains
Latest from METAL LAB
- Google Discover adds chatbot that adjusts your feed based on spoken preferences
- OpenAI Closes In on Anthropic Again in Enterprise Spending Share
- Meta Unveils First 10 Tasks in WildArtifactBench, a Benchmark for AI Agents
- Musk: "Optimus + Grok will one day handle healthcare for all humanity"
- 35% of Web Pages Published Since ChatGPT Show Signs of AI Authorship
Figures: Bach Phan-Tat et al., arXiv:2608.19472, CC BY 4.0