SynFlow: A Multidimensional Diachronic Semantic Analysis Toolkit

# SynFlow: A Multidimensional Diachronic Semantic Analysis Toolkit
Source: [https://arxiv.org/html/2608.19472](https://arxiv.org/html/2608.19472)
Kris HeylenAffiliation:\[1ex\] Department of Linguistics, KU LeuvenAffiliation:Instituut voor de Nederlandse TaalDirk GeeraertsAffiliation:\[1ex\] Department of Linguistics, KU LeuvenStefano De PascaleAffiliation:\[1ex\] Department of Linguistics, KU LeuvenAffiliation:Vrije Universiteit Brussel\[1ex\]Correspondence:[ttbach\.phan@kuleuven\.be](mailto:[email protected])Dirk SpeelmanAffiliation:\[1ex\] Department of Linguistics, KU Leuven

###### Abstract

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 presentSynFlow, 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 adjectiveviral, 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\.

## 1Introduction

Recent computational work on lexical semantic change \(LSC\) is dominated by vector\-space modelling approaches that compare vector representations of target words across historical periods[20](https://arxiv.org/html/2608.19472#bib.bib10);[29](https://arxiv.org/html/2608.19472#bib.bib13);[30](https://arxiv.org/html/2608.19472#bib.bib21)\. While effective on shared tasks, such methods are difficult to interpret[17](https://arxiv.org/html/2608.19472#bib.bib9), making their outputs difficult to relate directly to linguistic theories of semantic change: a model may indicate that a word has changed without directly revealing which aspects of its usage account for that change\.

We introduce SynFlow, an open\-source toolkit for multidimensional diachronic semantic analysis\. SynFlow represents each dimension as a period\-specific distribution of observable values and quantifies semantic change as differences between the diachronic distributions within a dimension\. It natively supports automatic extraction and quantification of dependency\-based co\-occurrences, morphological features, and constructional configurations, while externally derived representations such as Frame Semantics can be incorporated through the same interface\. This allows different dimensions to be analysed within a common diachronic workflow rather than requiring a separate analytical procedure for each representation\.

SynFlow compares these distributions across periods using cosine distance, Jensen–Shannon divergence \(JSD\)[18](https://arxiv.org/html/2608.19472#bib.bib15), or total variation distance \(TVD\), and decomposes change scores into contributions from individual values, making it possible to determinewhich dimensions changeandwhat drives the change\. For lexical fillers, it additionally provides incremental clustering to trace broader thematic developments\. The system therefore preserves a direct connection between quantitative change signals and the linguistic evidence underlying them\.

The main contribution of SynFlow is ashared workflow for multidimensional diachronic semantic analysis, providing temporal comparison, statistical testing, and value\-level decomposition\.

We demonstrate111https://www\.youtube\.com/watch?v=CRlG2kgdfTESynFlow with a qualitative case study of the German adjectiveviral, showing how one semantic development is reflected across multiple linguistic dimensions\. We also report previously published SemEval\-2020 Task 1 results[32](https://arxiv.org/html/2608.19472#bib.bib28);[31](https://arxiv.org/html/2608.19472#bib.bib27)to situate the performance of these representations relative to existing LSC systems\.

## 2Related work

### 2\.1Linguistic dimensions in diachronic analysis

Theory\-driven studies of semantic change have long examined linguistically motivated dimensions such as prototype structure, metaphor, frame semantics and constructions\([8](https://arxiv.org/html/2608.19472#bib.bib30);[11](https://arxiv.org/html/2608.19472#bib.bib11);[15](https://arxiv.org/html/2608.19472#bib.bib29);[21](https://arxiv.org/html/2608.19472#bib.bib26);[26](https://arxiv.org/html/2608.19472#bib.bib25);[27](https://arxiv.org/html/2608.19472#bib.bib24);[28](https://arxiv.org/html/2608.19472#bib.bib12);[35](https://arxiv.org/html/2608.19472#bib.bib23), e\.g\.,\)\. While these approaches provide detailed accounts of semantic development, their analytical dimensions are often tailored to particular lexical items or change phenomena, limiting systematic comparison across targets and corpora\. Computational work has begun to operationalise structured linguistic information at a larger scale\([14](https://arxiv.org/html/2608.19472#bib.bib14);[23](https://arxiv.org/html/2608.19472#bib.bib19), e\.g\.,\)\. SynFlow extends this direction by providing a reusable workflow for comparing the temporal distributions of such structured usage dimensions within a common framework\.

### 2\.2Computational tools for linguistic profiling

SynFlow is related to corpus tools that profile lexical behaviour, including Sketch Engine’s Word Sketch Difference \(WSD\)[36](https://arxiv.org/html/2608.19472#bib.bib18), Korp[2](https://arxiv.org/html/2608.19472#bib.bib31), and DiaCollo[12](https://arxiv.org/html/2608.19472#bib.bib22)\. DiaCollo models diachronic change through linear\-window collocations which often suffer from accidental co\-occurrences[7](https://arxiv.org/html/2608.19472#bib.bib16)\. WSD and Korp provide predefined syntactic profiles and association measures that rank individual collocates whereas SynFlow supportsconfigurable dependency paths, quantifies redistribution ofcomplete profiles, and decomposes change into individual contributions\. SynFlow also processes multiple temporal periods withina single passrather than requiring repeated pairwise comparisons like WSD\.

There are other tools operating on structured linguistic information in parsed corpora\. ANNIS[4](https://arxiv.org/html/2608.19472#bib.bib36), Grew\-match[9](https://arxiv.org/html/2608.19472#bib.bib35), and the FrameNet Brasil WebTool[34](https://arxiv.org/html/2608.19472#bib.bib34)support querying, visualisation, or annotation; STARK[13](https://arxiv.org/html/2608.19472#bib.bib4)extracts configurable dependency patterns; Profiling\-UD[3](https://arxiv.org/html/2608.19472#bib.bib3)derives multi\-level linguistic profiles; ComparaTree[33](https://arxiv.org/html/2608.19472#bib.bib2)compares linguistic properties across treebanks; and DELTA[6](https://arxiv.org/html/2608.19472#bib.bib1)combines dependency\-pattern extraction with multi\-level diversity measurement across corpora\. However, these systems primarily support querying, extraction, profiling, annotation, or corpus\-level comparison rather thantarget\-oriented diachronic semantic analysis\. SynFlow complements them by utilising structured linguistic information to model diachronic semantics\. It models linguistic observations as target\-conditioned temporal distributions and analyses their redistribution through statistical testing and value\-level decomposition\.

## 3System architecture

SynFlow takes annotated diachronic subcorpora as input, with each directory representing one temporal period\. Given a target lemma, POS, and selected dimension types, it extracts the corresponding linguistic observations, converts them into period\-specific distributions, measures change across periods, and decomposes the resulting scores into contributions from individual values\. The overall workflow is demonstrated in figure[1](https://arxiv.org/html/2608.19472#S3.F1)\.

Query\(e\.g\., viral/ADJ\)DimensionExtraction\(Output: dataframe\)DistributionConstructionChangeDetection\(cosine distance, JSD, TVD\)ChangeDecomposition\(item\-level contributions\)

Figure 1:Overview of the workflow\.SynFlow distinguishes between adimension type, which specifies the linguistic information to extract, and an individualdimension, whose observed values form a distribution\. For targetww, dimensiondd, and periodtt, usage is represented asPt​\(v∣d,w\)P\_\{t\}\(v\\mid d,w\)over valuesvv\. Some dimension types such asconstructiondefine a single distribution, while others define multiple dimensions, such as those conditioned on a specific dependency slot or morphological feature type\.

SynFlow natively extracts dependency\-based, constructional, and morphological information, while externally annotated dimensions can be analysed through the same downstream workflow provided their dataframe format matches the output of the Dimension Extraction step\. These dimensions provide complementary views of linguistic usage\. Slot and feature types describe broader structural profiles, slot fillers and morphological features capture variation within particular categories, and constructional configurations capture how multiple syntactic relations combine within the same usage instance\. Dedicated notebook templates are provided for the supported analysis types\.

### 3\.1Dimension extraction

##### Input format

SynFlow uses a CoNLL\-U\-like format in which each sub\-corpus is divided into individual sentences\. Each sentence begins with <s=sentence\_id\> and ends with </s\>\. Seven tab\-separated columns are required in the following order:token, lemma, pos, id, head\_id, deprel, feats\. We have also provided a notebook and scripts for parsing the raw data \(1 sentence/line\) into the correct SynFlow format using Stanza[22](https://arxiv.org/html/2608.19472#bib.bib8)\.

Table 1:Dimensions and example values natively extracted by SynFlow\.
##### Dimensions

SynFlow natively extracts five types of linguistic information \(Table[1](https://arxiv.org/html/2608.19472#S3.T1)\) using different tree\-traversal algorithms\.Slot typesare individual dependency relations involving the target, with direction encoded aschi\_for children andpa\_for parents\. For example, inthe boy eats the cake,eatis associated withchi\_nsubjandchi\_objas it is the head ofboyandcake, whileboyis associated withpa\_nsubjandchi\_det\.Slot fillersare the lexical items occupying a dependency slot, with each slot defining a separate dimension\.Constructional configurationsrepresent combinations of dependency slots co\-occurring around the same target instance, e\.g\.\[chi\_nsubj \+ chi\_obj\], within a user\-specified dependency\-path depth\.Morphological feature typesare categories extracted from the target’sfeatsfield, whilemorphological featuresrepresent the observed values of each category as separate dimensions\.

### 3\.2Change detection

Counts within each dimension and period are normalised to sum to one, yielding probability distributions\. This differs from occurrence\-normalised frequencies used in some descriptive visualisations, where counts are divided by the number of target occurrences\. SynFlow then measures distribution change between consecutive periods using cosine distance, JSD, or TVD\. Because distances estimated from sparse observations can be unstable, SynFlow optionally applies support weighting:

W​e​i​g​h​t​e​d​D​i​s​t=R​a​w​D​i​s​t×min⁡\(1,ck\),WeightedDist=RawDist\\times\\min\\left\(1,\\frac\{c\}\{k\}\\right\),wherec=min⁡\(c1,c2\)c=\\min\(c\_\{1\},c\_\{2\}\)is the smaller number of observations in the two periods, andkkis a user\-defined support threshold\. Distances receive full weight \(i\.e\., 1\) when both periods contain at leastkkobservations and are linearly down\-weighted otherwise\.

SynFlow also supports permutation testing to help distinguish meaningful changes from random fluctuations\. Alongside the magnitude of the detected change, it provides app\-value indicating how likely a change of similar or greater size would be observed by chance\. This gives users an interpretable measure of whether a detected change is statistically significant\.

For multi\-period analyses, SynFlow applies the Benjamini–Yekutieli \(BY\) procedure[1](https://arxiv.org/html/2608.19472#bib.bib33)to control the False Discovery Rate \(FDR\)\. For each dimension, correction is applied jointly to thepp\-values from all adjacent\-period comparisons\.

#### 3\.2\.1Incremental slot filler clustering

This section only applies to slot fillers\. To distinguish lexical replacement from broader thematic change, SynFlow provides incremental clustering of slot fillers\. The analysis uses static word2vec embeddings[19](https://arxiv.org/html/2608.19472#bib.bib7)trained separately for each period and Procrustes aligned[25](https://arxiv.org/html/2608.19472#bib.bib6)to a common space, following diachronic embedding approaches\([10](https://arxiv.org/html/2608.19472#bib.bib20)\)\. Fillers in the first period are grouped using hierarchical agglomerative clustering, with each cluster represented by the frequency\-weighted centroid of its members\. In subsequent periods, fillers are assigned to sufficiently similar existing centroids, while unmatched fillers form new clusters\. Active centroids are then updated from the current\-period fillers\. Previously established clusters are retained when temporarily inactive, allowing them to be reactivatable in later periods\. This incremental memory enables SynFlow to trace the emergence, persistence, decline, and reactivation of broader filler themes over time\. A dedicated notebook is provided for training and aligning the required word2vec embeddings\.

### 3\.3Change decomposition

After computing the distance between two period\-specific distributions, SynFlow can further decompose the observed change into contributions from individual values\. This allows users to move from a single aggregate change score to the specific linguistic items responsible for that difference\. For a dimensiondd, SynFlow reports how much each valuevvcontributes to the overall distance between two consecutive periods\. For example, if the distribution ofFILLER\[chi\_amod\]changes substantially between two periods, decomposition can reveal which adjectival fillers account for most of that shift\. This step is particularly useful for qualitative analysis, because it links the numerical change score back to interpretable linguistic evidence\. Rather than only indicating that a dimension has changed, SynFlow shows which individual values drive the observed redistribution\.

### 3\.4Output and visualisation

Most intermediate and final outputs produced by SynFlow are stored as comma\-separated values \(CSV\) files\. This allows users to inspect the extracted information directly, verify intermediate processing steps, and reuse the outputs in external analyses\. The accompanying notebooks also provide visualisations at different stages of the workflow to facilitate exploration and interpretation\. These include time\-series plots for tracking changes across periods, heatmaps for comparing patterns across dimensions or time, and bar plots for inspecting distributions and item\-level contributions\. Together, the tabular outputs and visualisations allow users to move between quantitative summaries and the linguistic evidence underlying them\.

### 3\.5Availability and execution

SynFlow is freely available as open\-source software under the MIT license\. The repository222https://github\.com/phantatbach/SynFlowprovides installation instructions, dependency specifications, documentation of the expected corpus format, and ready\-to\-use notebook templates for the main analysis types\. SynFlow is currently distributed as a notebook\-based Python workflow\. Users install the system by cloning the repository, creating a Python environment, installing the packages specified inrequirements\.txt, and running the notebooks\. Python 3\.10 or newer is recommended\.

There are dedicated notebooks333Because we cannot upload the whole dataset, we have provided dataframes of different dimensions so that users can explore Change Detection and Change Decomposition\.forslot\-level,constructional,morphological\-feature,diachronic embeddingandqualitative analysis\. Within each notebook, users specify the corpus location, target lemma and POS, temporal periods, and relevant analysis parameters\. Input corpora are organised into directories corresponding to temporal periods and follow the seven\-column CoNLL\-U\-like format described in[3\.1](https://arxiv.org/html/2608.19472#S3.SS1.SSS0.Px1)\. The notebooks then guide users through the complete workflow from dimension extraction and distribution construction to change measurement, decomposition, statistical analysis, and visualisation\. This design allows the different linguistic dimensions to be analysed through a common workflow without requiring users to modify SynFlow’s underlying extraction and analysis functions\.

### 3\.6Runtime and scalability

SynFlow supports multiprocessing at every possible step\. As a practical benchmark, we ran SynFlow with a representative target \(plane/NOUN\) from Corpus of Historical American English[5](https://arxiv.org/html/2608.19472#bib.bib5)\(1810s \- 2000s, divided into 20 periods and containing 116,613 parsed files\), on a personal laptop \(32 GB RAM, AMD Ryzen 7 8845HS\)\. Direct slot\-level exploration, which reads the parsed corpus, counts dependency slots within each period, and produces the corresponding visualisations, took approximately 7 minutes 30 seconds\. Extracting all slot fillers and constructing the slot\-filler dataframe took 4 minutes 55 seconds\. For each adjacent\-period transition, we ran 1,000 permutations for each pair of slot\-filler distributions of every extracted slot\. Processing all slot\-specific permutation tests for one transition required only 3 minutes\. These results indicate that SynFlow remains practical for large diachronic corpora on consumer hardware\.

## 4Evaluation

### 4\.1Demonstrative case study

We demonstrate SynFlow with the German adjectiveviralin the Leipzig German News corpus[16](https://arxiv.org/html/2608.19472#bib.bib32), covering 1995–2025 and divided into six five\-year periods\. Over this period,viralshifts from predominantly virus\-related uses, as invirale Infektion‘viral infection’, towards the newer sense ‘widely circulated or popular’, especially in online and social\-media contexts, as inviral gehen‘go viral’\.

In theslot typedimension, occurrence\-normalised frequencies of different slots \(Figure[2](https://arxiv.org/html/2608.19472#S4.F2)\) remain largely stable except for a marked shift around 2011–2015:pa\_amoddecreases whilepa\_advmodincreases, reflecting the growing use ofviralas an adverbial modifier in expressions such asviral gehen\.

![Refer to caption](https://arxiv.org/html/2608.19472v1/slot_types_diachronic_freq.png)Figure 2:Change of normalised frequencies of all slot types ofviral\.This redistribution produces a large TVD between the two periods, and decomposition shows that the redistribution ofpa\_amodandpa\_advmodaccounts for much of the overall TVD \(Figure[3](https://arxiv.org/html/2608.19472#S4.F3)\)\.

![Refer to caption](https://arxiv.org/html/2608.19472v1/slot_type_contrib.png)Figure 3:Contributions of individual slot types to TVD ofviral\. Green bars indicate an increase in relative frequency; red bars indicate a decrease\.![Refer to caption](https://arxiv.org/html/2608.19472v1/diachronic_tvd_all_slots.png)Figure 4:Weighted TVD trajectories for slot\-filler distributions across dependency slots\.![Refer to caption](https://arxiv.org/html/2608.19472v1/incremental_clustering.png)Figure 5:Incremental clustering results with social\-media\-related clusters growing rapidly from 2011\.Theslot\-fillerdimensions ofpa\_amodandpa\_advmodshow the largest weighted TVD increases around 2011–2015 \(Figure[4](https://arxiv.org/html/2608.19472#S4.F4)\), while most other slots remain comparatively stable\. Further decomposition and clustering analysis \(Figure[5](https://arxiv.org/html/2608.19472#S4.F5)\) reveal that these changes are increasingly associated with fillers from online and social\-media discourse, includingHit, Marketing, Video, TikTokinpa\_amod, andgeheninpa\_advmod\. Slot\-filler analysis therefore complements the structural signal by showing the lexical contexts accompanying the newer meaning ofviral\.

The same development is visible at theconstructionallevel, showing that the rise of the adverbialviralis not limited to an isolated dependency relation, but also affects the larger combinations of grammatical relations in which the target occurs\. From around the same period, configurations containingpa\_advmodbecome increasingly frequent \(Figure[6](https://arxiv.org/html/2608.19472#S4.F6)\)\. For example, the configuration\[chi\_advmod \+ pa\_advmod\]occurs in expressions such asgerade viral gehen‘to be going viral right now’, whereviralis linked togehenthroughpa\_advmodand togeradethroughchi\_advmod\.

![Refer to caption](https://arxiv.org/html/2608.19472v1/construction_freq_comparison.png)Figure 6:Frequencies of different constructions aroundviralAt themorphologicallevel, the frequency ofVariantincreases around the same period aspa\_advmod\. These uninflected adverbial uses are annotated asVariant=Short\. However,FEAT\[Variant\]contains only the valueShort, meaning that the change is visible at the feature\-type level, but not as redistribution within the corresponding feature\-value dimension\. This contrasts with the dependency dimensions, where change is visible both at the broader structural level and within individual slot\-filler distributions\.

Overall, the case ofviralillustrates how one semantic development can leave complementary traces across several linguistic dimensions\. Rather than collapsing these signals into a single score, SynFlow allows users to identify where change occurs and inspect the linguistic evidence underlying each dimension\.

### 4\.2Experimental benchmark

We previously benchmarked SynFlow on SemEval\-2020 Task 1[24](https://arxiv.org/html/2608.19472#bib.bib17)using two linguistic representations: dependency\-based slot fillers and Frame Semantics[32](https://arxiv.org/html/2608.19472#bib.bib28);[31](https://arxiv.org/html/2608.19472#bib.bib27)\. Both representations proved effective for lexical semantic change detection and ranked above many neural\-network\-based systems on the benchmark \(Table[2](https://arxiv.org/html/2608.19472#S4.T2)\), while retaining direct interpretability of the linguistic evidence underlying the change scores\.

Table 2:Ranking of previously evaluated SynFlow representations against systems reported for the English SemEval\-2020 benchmark\. Full results are reported in[32](https://arxiv.org/html/2608.19472#bib.bib28);[31](https://arxiv.org/html/2608.19472#bib.bib27)\.

## 5Conclusion and future work

We presented SynFlow, an open\-source toolkit for multidimensional diachronic semantic analysis\. It natively supports dependency\-based, constructional, and morphological information while allowing externally derived representations to be analysed through the same interface\.

Theviralcase study illustrates how a single semantic development can be traced across multiple dimensions, while previous benchmark results show competitive performance against embedding\-based methods\.

Future work will extend SynFlow to additional linguistic dimensions and integrate further parsers and LLM\-based annotation methods into the extraction pipeline\.

## References

- Benjamini and Yekutieli \(2001\)Y\. Benjamini and D\. YekutieliThe control of the false discovery rate in multiple testing under dependency\.The Annals of Statistics29\(4\)\.External Links:ISSN 0090\-5364,[Link](https://projecteuclid.org/journals/annals-of-statistics/volume-29/issue-4/The-control-of-the-false-discovery-rate-in-multiple-testing/10.1214/aos/1013699998.full),[Document](https://dx.doi.org/10.1214/aos/1013699998)Cited by:[§3\.2](https://arxiv.org/html/2608.19472#S3.SS2.p3.1)\.
- Borinet al\.\(2025\)L\. Borin, M\. Forsberg, M\. Hammarstedt, L\. Holmer, and A\. MatssonKorp: Språkbanken’s word research platform\.InSixty years of Swedish computational lexicography,D\. Dannélls, K\. Blensenius, and L\. Borin \(Eds\.\),Digital Linguistics, Vol\.3,pp\. 175–193\.External Links:[Document](https://dx.doi.org/10.1515/9783111577234-010)Cited by:[§2\.2](https://arxiv.org/html/2608.19472#S2.SS2.p1.1)\.
- Brunatoet al\.\(2020\)D\. Brunato, A\. Cimino, F\. Dell’Orletta, G\. Venturi, and S\. MontemagniProfiling\-UD: a tool for linguistic profiling of texts\.InProceedings of the Twelfth Language Resources and Evaluation Conference,N\. Calzolari, F\. Béchet, P\. Blache, K\. Choukri, C\. Cieri, T\. Declerck, S\. Goggi, H\. Isahara, B\. Maegaard, J\. Mariani, H\. Mazo, A\. Moreno, J\. Odijk, and S\. Piperidis \(Eds\.\),Marseille, France,pp\. 7145–7151\(eng\)\.External Links:[Link](https://aclanthology.org/2020.lrec-1.883/),ISBN 979\-10\-95546\-34\-4Cited by:[§2\.2](https://arxiv.org/html/2608.19472#S2.SS2.p2.1)\.
- Chiarcos and Lüdeling \(2009\)C\. Chiarcos and A\. LüdelingANNIS: A Search Tool for Multi\-Layer Annotated Corpora\.InProceedings of the Corpus Linguistics Conference 2009 \(CL2009\),Vol\.358\.Cited by:[§2\.2](https://arxiv.org/html/2608.19472#S2.SS2.p2.1)\.
- Davies \(2010\)M\. DaviesThe corpus of historical american english \(coha\)\.External Links:[Link](https://www.english-corpora.org/coha/)Cited by:[§3\.6](https://arxiv.org/html/2608.19472#S3.SS6.p1.1)\.
- Estève and Dobrovoljc \(2026\)L\. Estève and K\. DobrovoljcDELTA: a toolkit for measuring linguistic diversity in dependency\-parsed corpora\.InProceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics \(Volume 3: System Demonstrations\),D\. Croce, J\. Leidner, and N\. S\. Moosavi \(Eds\.\),Rabat, Marocco,pp\. 75–85\.External Links:[Link](https://aclanthology.org/2026.eacl-demo.6/),[Document](https://dx.doi.org/10.18653/v1/2026.eacl-demo.6),ISBN 979\-8\-89176\-382\-1Cited by:[§2\.2](https://arxiv.org/html/2608.19472#S2.SS2.p2.1)\.
- Evert \(2008\)S\. EvertCorpora and collocations\.InCorpus Linguistics\. An International Handbook,Cited by:[§2\.2](https://arxiv.org/html/2608.19472#S2.SS2.p1.1)\.
- Geeraerts \(1993\)D\. GeeraertsVagueness’s puzzles, polysemy’s vagaries\.Cognitive Linguistics4\(3\),pp\. 223–272\.External Links:[Link](https://doi.org/10.1515/cogl.1993.4.3.223),[Document](https://dx.doi.org/doi%3A10.1515/cogl.1993.4.3.223)Cited by:[§2\.1](https://arxiv.org/html/2608.19472#S2.SS1.p1.1)\.
- Guillaume \(2021\)B\. GuillaumeGraph Matching and Graph Rewriting: GREW tools for corpus exploration, maintenance and conversion\.InProceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: System Demonstrations,D\. Gkatzia and D\. Seddah \(Eds\.\),Online,pp\. 168–175\.External Links:[Link](https://aclanthology.org/2021.eacl-demos.21/),[Document](https://dx.doi.org/10.18653/v1/2021.eacl-demos.21)Cited by:[§2\.2](https://arxiv.org/html/2608.19472#S2.SS2.p2.1)\.
- Hamiltonet al\.\(2016\)W\. L\. Hamilton, J\. Leskovec, and D\. JurafskyDiachronic Word Embeddings Reveal Statistical Laws of Semantic Change\.InProceedings of the 54th Annual Meeting of the Association for Computational Linguistics \(Volume 1: Long Papers\),K\. Erk and N\. A\. Smith \(Eds\.\),Berlin, Germany,pp\. 1489–1501\.External Links:[Link](https://aclanthology.org/P16-1141/),[Document](https://dx.doi.org/10.18653/v1/P16-1141)Cited by:[§3\.2\.1](https://arxiv.org/html/2608.19472#S3.SS2.SSS1.p1.1)\.
- Jansegers and Gries \(2020\)M\. Jansegers and S\. Th\. GriesTowards a dynamic behavioral profile: A diachronic study of polysemous sentir in Spanish\.Corpus Linguistics and Linguistic Theory16\(1\),pp\. 145–187\(en\)\.External Links:ISSN 1613\-7035, 1613\-7027,[Link](https://www.degruyter.com/document/doi/10.1515/cllt-2016-0080/html),[Document](https://dx.doi.org/10.1515/cllt-2016-0080)Cited by:[§2\.1](https://arxiv.org/html/2608.19472#S2.SS1.p1.1)\.
- Jurish \(2015\)B\. JurishDiaCollo: On the trail of diachronic collocations\.Cited by:[§2\.2](https://arxiv.org/html/2608.19472#S2.SS2.p1.1)\.
- Krsnik and Dobrovoljc \(2025\)L\. Krsnik and K\. DobrovoljcSTARK: a toolkit for dependency \(sub\)tree extraction and analysis\.InProceedings of the 23rd International Workshop on Treebanks and Linguistic Theories \(TLT, SyntaxFest 2025\),S\. Jablotschkin, S\. Kübler, and H\. Zinsmeister \(Eds\.\),Ljubljana, Slovenia,pp\. 44–51\.External Links:[Link](https://aclanthology.org/2025.tlt-1.5/),ISBN 979\-8\-89176\-291\-6Cited by:[§2\.2](https://arxiv.org/html/2608.19472#S2.SS2.p2.1)\.
- Kutuzovet al\.\(2021\)A\. Kutuzov, L\. Pivovarova, and M\. GiulianelliGrammatical Profiling for Semantic Change Detection\.InProceedings of the 25th Conference on Computational Natural Language Learning,Online,pp\. 423–434\(en\)\.External Links:[Link](https://aclanthology.org/2021.conll-1.33),[Document](https://dx.doi.org/10.18653/v1/2021.conll-1.33)Cited by:[§2\.1](https://arxiv.org/html/2608.19472#S2.SS1.p1.1)\.
- Law \(2019\)J\. LawDiachronic frame analysis\.Constructions and Frames11\(1\),pp\. 43–78\.Note:Type: Journal ArticleExternal Links:ISSN 1876\-1933,[Link](https://www.jbe-platform.com/content/journals/10.1075/cf.00023.law),[Document](https://dx.doi.org/https%3A//doi.org/10.1075/cf.00023.law)Cited by:[§2\.1](https://arxiv.org/html/2608.19472#S2.SS1.p1.1)\.
- Leipzig \(2026\)W\. LeipzigDownloads – German – Wortschatz Leipzig\.External Links:[Link](https://wortschatz-leipzig.de/en/download/deu)Cited by:[§4\.1](https://arxiv.org/html/2608.19472#S4.SS1.p1.1)\.
- Lenciet al\.\(2022\)A\. Lenci, M\. Sahlgren, P\. Jeuniaux, A\. Cuba Gyllensten, and M\. MilianiA comparative evaluation and analysis of three generations of Distributional Semantic Models\.Language Resources and Evaluation56\(4\),pp\. 1269–1313\(en\)\.External Links:ISSN 1574\-020X, 1574\-0218,[Link](https://link.springer.com/10.1007/s10579-021-09575-z),[Document](https://dx.doi.org/10.1007/s10579-021-09575-z)Cited by:[§1](https://arxiv.org/html/2608.19472#S1.p1.1)\.
- Menéndezet al\.\(1997\)M\.L\. Menéndez, J\.A\. Pardo, L\. Pardo, and M\.C\. PardoThe Jensen\-Shannon divergence\.Journal of the Franklin Institute334\(2\),pp\. 307–318\.External Links:ISSN 0016\-0032,[Link](https://www.sciencedirect.com/science/article/pii/S0016003296000634),[Document](https://dx.doi.org/10.1016/S0016-0032%2896%2900063-4)Cited by:[§1](https://arxiv.org/html/2608.19472#S1.p3.1)\.
- Mikolovet al\.\(2013\)T\. Mikolov, K\. Chen, G\. Corrado, and J\. DeanEfficient estimation of word representations in vector space\.External Links:1301\.3781,[Link](https://arxiv.org/abs/1301.3781)Cited by:[§3\.2\.1](https://arxiv.org/html/2608.19472#S3.SS2.SSS1.p1.1)\.
- Periti and Montanelli \(2024\)F\. Periti and S\. MontanelliLexical Semantic Change through Large Language Models: a Survey\.ACM Computing Surveys56\(11\),pp\. 1–38\(en\)\.External Links:ISSN 0360\-0300, 1557\-7341,[Link](https://dl.acm.org/doi/10.1145/3672393),[Document](https://dx.doi.org/10.1145/3672393)Cited by:[§1](https://arxiv.org/html/2608.19472#S1.p1.1)\.
- Pettersson\-Traba \(2016\)D\. Pettersson\-TrabaAnalyzing the behavioral profiles of sets of near\-synonyms in American English from a diachronic perspective\.External Links:[Link](https://api.semanticscholar.org/CorpusID:56020073)Cited by:[§2\.1](https://arxiv.org/html/2608.19472#S2.SS1.p1.1)\.
- Qiet al\.\(2020\)P\. Qi, Y\. Zhang, Y\. Zhang, J\. Bolton, and C\. D\. ManningStanza: a python natural language processing toolkit for many human languages\.InProceedings of the 58th Annual Meeting of the Association for Computational Linguistics: System Demonstrations,A\. Celikyilmaz and T\. Wen \(Eds\.\),Online,pp\. 101–108\.External Links:[Link](https://aclanthology.org/2020.acl-demos.14/),[Document](https://dx.doi.org/10.18653/v1/2020.acl-demos.14)Cited by:[§3\.1](https://arxiv.org/html/2608.19472#S3.SS1.SSS0.Px1.p1.1)\.
- Ryzhovaet al\.\(2021\)A\. Ryzhova, D\. Ryzhova, and I\. SochenkovDetection of Semantic Changes in Russian Nouns with Distributional Models and Grammatical Features\.External Links:[Link](https://api.semanticscholar.org/CorpusID:240067372)Cited by:[§2\.1](https://arxiv.org/html/2608.19472#S2.SS1.p1.1)\.
- Schlechtweget al\.\(2020\)D\. Schlechtweg, B\. McGillivray, S\. Hengchen, H\. Dubossarsky, and N\. TahmasebiSemEval\-2020 Task 1: Unsupervised Lexical Semantic Change Detection\.InProceedings of the Fourteenth Workshop on Semantic Evaluation,Barcelona \(online\),pp\. 1–23\(en\)\.External Links:[Link](https://aclanthology.org/2020.semeval-1.1),[Document](https://dx.doi.org/10.18653/v1/2020.semeval-1.1)Cited by:[§4\.2](https://arxiv.org/html/2608.19472#S4.SS2.p1.1)\.
- Smithet al\.\(2017\)S\. L\. Smith, D\. H\. Turban, S\. Hamblin, and N\. Y\. HammerlaOffline bilingual word embeddings\.InInternational Conference on Learning Representations \(ICLR\),Cited by:[§3\.2\.1](https://arxiv.org/html/2608.19472#S3.SS2.SSS1.p1.1)\.
- Sovran \(2004\)T\. SovranPolysemy and Semantic Frames \- A diachronic study of "Truth" in Hebrew\.26\(1\),pp\. 89–100\.External Links:[Document](https://dx.doi.org/10.3406/verbu.2004.1773)Cited by:[§2\.1](https://arxiv.org/html/2608.19472#S2.SS1.p1.1)\.
- Sullivan \(2007\)K\. SullivanMetaphoric Extension and Invited Inferencing in Semantic Change\.Culture, Language and RepresentationSpecial Issue: Metaphor and Discourse,pp\. 257–274\.Cited by:[§2\.1](https://arxiv.org/html/2608.19472#S2.SS1.p1.1)\.
- Sweetser \(1990\)E\. SweetserFrom Etymology to Pragmatics: Metaphorical and Cultural Aspects of Semantic Structure\.Cambridge Studies in Linguistics,Cambridge University Press,Cambridge\.External Links:ISBN 978\-0\-521\-42442\-4,[Link](https://www.cambridge.org/core/books/from-etymology-to-pragmatics/F1454419128B2D24A372DD0808B51198),[Document](https://dx.doi.org/10.1017/CBO9780511620904)Cited by:[§2\.1](https://arxiv.org/html/2608.19472#S2.SS1.p1.1)\.
- Tahmasebiet al\.\(2021\)N\. Tahmasebi, L\. Borin, A\. Jatowt, Y\. Xu, and S\. HengchenComputational approaches to semantic change\.\(en\)\.Cited by:[§1](https://arxiv.org/html/2608.19472#S1.p1.1)\.
- Tahmasebi and Dubossarsky \(2023\)N\. Tahmasebi and H\. DubossarskyComputational modeling of semantic change\.Note:\_eprint: 2304\.06337External Links:[Link](https://arxiv.org/abs/2304.06337)Cited by:[§1](https://arxiv.org/html/2608.19472#S1.p1.1)\.
- Tatet al\.\(2026a\)B\. P\. Tat, K\. Heylen, S\. De Pascale, D\. Geeraerts, and D\. SpeelmanReFRAME or remain: unsupervised lexical semantic change detection with frame semantics\.InProceedings of the 15th Joint Conference on Lexical and Computational Semantics \(\*SEM 2026\),S\. M\. Mohammad and N\. Ousidhoum \(Eds\.\),San Diego, California, United States,pp\. 83–97\.External Links:[Link](https://aclanthology.org/2026.starsem-conference.5/),[Document](https://dx.doi.org/10.18653/v1/2026.starsem-conference.5),ISBN 979\-8\-89176\-413\-2Cited by:[§1](https://arxiv.org/html/2608.19472#S1.p5.1),[§4\.2](https://arxiv.org/html/2608.19472#S4.SS2.p1.1),[Table 2](https://arxiv.org/html/2608.19472#S4.T2)\.
- Tatet al\.\(2026b\)B\. P\. Tat, K\. Heylen, D\. Geeraerts, S\. De Pascale, and D\. SpeelmanTransparent semantic change detection with dependency\-based profiles\.InThe Proceedings for the 6th International Workshop on Computational Approaches to Language Change \(LChange’26\),N\. Tahmasebi, P\. Cassotti, S\. Montariol, A\. Kutuzov, N\. Huebscher, E\. Spaziani, and N\. Baes \(Eds\.\),Rabat, Morocco,pp\. 97–109\.External Links:[Link](https://aclanthology.org/2026.lchange-1.8/),[Document](https://dx.doi.org/10.18653/v1/2026.lchange-1.8),ISBN 979\-8\-89176\-362\-3Cited by:[§1](https://arxiv.org/html/2608.19472#S1.p5.1),[§4\.2](https://arxiv.org/html/2608.19472#S4.SS2.p1.1),[Table 2](https://arxiv.org/html/2608.19472#S4.T2)\.
- Terčon and Dobrovoljc \(2025\)L\. Terčon and K\. DobrovoljcComparaTree: a multi\-level comparative treebank analysis tool\.InProceedings of the 23rd International Workshop on Treebanks and Linguistic Theories \(TLT, SyntaxFest 2025\),S\. Jablotschkin, S\. Kübler, and H\. Zinsmeister \(Eds\.\),Ljubljana, Slovenia,pp\. 129–139\.External Links:[Link](https://aclanthology.org/2025.tlt-1.15/),ISBN 979\-8\-89176\-291\-6Cited by:[§2\.2](https://arxiv.org/html/2608.19472#S2.SS2.p2.1)\.
- Torrentet al\.\(2025\)T\. T\. Torrent, E\. E\. D\. S\. Matos, A\. D\. D\. Costa, M\. A\. Gamonal, S\. Peron\-Corrêa, and V\. M\. R\. L\. PaivaA flexible tool for a qualia\-enriched FrameNet: the FrameNet Brasil WebTool\.Language Resources and Evaluation59\(4\),pp\. 4529–4557\(en\)\.External Links:ISSN 1574\-020X, 1574\-0218,[Link](https://link.springer.com/10.1007/s10579-023-09714-8),[Document](https://dx.doi.org/10.1007/s10579-023-09714-8)Cited by:[§2\.2](https://arxiv.org/html/2608.19472#S2.SS2.p2.1)\.
- Vais \(2022\)C\. M\. VaisA Diachronic Application of Frame Semantics to Meaning Change in Prepositions: The Case of German mit\.Doctoral dissertation,University of California, Berkeley,Berkeley, CA\.External Links:[Link](https://escholarship.org/uc/item/3fv9m9b0)Cited by:[§2\.1](https://arxiv.org/html/2608.19472#S2.SS1.p1.1)\.
- Word\-Sketch\-Difference \(2019\)Word\-Sketch\-DifferenceWord sketch difference\.\(en\-GB\)\.External Links:[Link](https://www.sketchengine.eu/guide/word-sketch-difference-compare-words/)Cited by:[§2\.2](https://arxiv.org/html/2608.19472#S2.SS2.p1.1)\.