Pith. sign in

REVIEW 4 major objections 5 minor 12 references

A History of Philosophy in Colombia through Topic Modelling

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Value theory and epistemology dominate Colombia's philosophy journal, with no decline in historical articles since 1951.

desk verdict A useful first topic-modeling map of a Colombian philosophy journal, with a solid descriptive core and a historically-focused claim that needs sensitivity work before it can be trusted. read the letter →

arxiv 2412.04236 v1 pith:RZIUM52Q submitted 2024-12-05 cs.LG cs.CLcs.DL

Juan R. Loaiza , Miguel González-Duque This is my paper · ORCID
classification cs.LGcs.CLcs.DL
keywords topicmodellingdynamicmodelsIdeasyValoresColombianphilosophyvaluetheoryhistoryofSpanishnaturallanguageprocessingdigitalhumanities
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to show what a data-driven history of philosophy looks like outside the Anglophone mainstream by topic-modelling every Spanish-language article published in Ideas y Valores, Colombia's leading philosophy journal, between 1951 and 2022. Its central empirical claims are that value theory (ethics, political philosophy, aesthetics) and epistemology and philosophy of science dominate the journal's output, that German thinkers such as Kant, Husserl, and Hegel are constant reference points, and that historically focused articles did not significantly decline despite an editorial turn toward 'original' work. A sympathetic reader would take these findings as evidence that the journal's history is one of stable engagement with the history of philosophy and a lasting ethical-political orientation, and as a demonstration that the tools of computational text analysis transfer to a Spanish-language corpus.

What carries the argument

The central object is the Dynamic Topic Model, a temporal extension of Latent Dirichlet Allocation in which topic-word distributions evolve by small Gaussian steps from one time slice to the next. The paper fits K=90 topics to 875 Spanish-language articles, labels each topic with a PhilPapers main-area and subarea tag plus an optional '#historical' tag, and assigns documents to topics by taking the top documents that cumulatively cover 50% of the likelihood sum for each topic. This machinery produces the area shares and the yearly historical-ratio series that are then tested by linear regression.

What would settle it

Re-run the analysis with independent raters assigning '#historical' tags to the same 90 topics, or with cutoff values of 30% and 70%; a statistically significant negative correlation between the historical-article ratio and year under any of those replications would show that the reported non-decline is an artifact of the tagging and cutoff choices.

Watch

Extended reading notes

Core claim

Across Ideas y Valores from 1951 to January 2022, the journal's most prominent philosophical areas are value theory (571 documents, led by ethics and political philosophy) and metaphysics and epistemology (240 documents), with a stable emphasis on German philosophers, especially Kant, Hegel, Husserl, and Heidegger. Against the expectation raised by editorial debates, the share of documents assigned to historically focused topics shows no significant decline (r(38)=0.20, p=0.8885 on a one-sided test), and the fitted slope is slightly positive (0.0020). The authors conclude that editorial pressure toward original work did not measurably reduce historical scholarship.

Load-bearing premise

The claim that historical articles never declined rests on the authors' manually assigned '#historical' topic labels and their 50% likelihood cutoff for assigning documents to topics; if either misclassifies documents, the flat trend could vanish.

Editorial extensions

If this is right

  • If the flat historical ratio holds, the journal's 1980s policy shift toward 'original' work did not crowd out interpretive and historical scholarship.
  • The concentration in value theory suggests that Colombian philosophy's public identity is tied to ethics, political philosophy, and aesthetics more than to other areas.
  • German idealism and phenomenology, centered on Kant, Hegel, Husserl, and Heidegger, remain stable reference points across many topics throughout the journal's lifetime.
  • The same Dynamic Topic Model pipeline can be applied to other Latin American journals, allowing comparable regional histories of philosophy to be constructed.
  • The Spanish-language preprocessing workflow, including custom OCR correction and expanded stopword lists, is a necessary part of making topic modelling work on non-English philosophical corpora.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the 50% likelihood cutoff for assigning documents to topics is admitted to be arbitrary, the absolute document counts per main area are less robust than the qualitative topic patterns, and small changes in that cutoff could shift the area shares.
  • The '#historical' tag groups author-focused topics (Kant, Hegel, Husserl) together with exegetical method, so a finer distinction between 'about a historical figure' and 'using historical interpretation' could reveal a decline that the current binary tag hides.
  • If other Latin American journals show the same value-theory dominance, this would suggest a regional pattern in philosophy publication rather than a peculiarity of Ideas y Valores.
  • The stable ratio of historical articles may be tied to the persistent presence of German philosophy in the journal; testing author affiliations and reference networks could show whether the apparent stability is driven by a specific scholarly community.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper applies Dynamic Topic Models to 816 Spanish-language articles from the Colombian journal Ideas y Valores (1951-2022), labels the 90 learned topics using the PhilPapers taxonomy, assigns documents to topics via a 50% cumulative-likelihood cutoff, and reports descriptive trends: value theory, epistemology, and philosophy of science are the most prominent areas, German philosophers such as Kant, Hegel, and Husserl are strongly represented, and the ratio of articles in manually tagged "#historical" topics shows no significant decline (r(38)=0.20, p=0.8885). An appendix provides a tutorial-level derivation of LDA and DTM.

Significance. This is a valuable and unusually transparent contribution to the computational history of Latin American philosophy. It ships code and data, documents the Spanish preprocessing pipeline, uses an external public taxonomy to guide topic labeling, and is candid about several limitations. If the descriptive findings are robust, they provide a useful baseline for future journal-level studies in non-English philosophy. However, the main inferential claim about historical decline rests on the least validated part of the pipeline, namely the authors' manual #historical tags and an admitted ad-hoc document-assignment threshold, so the central conclusion is currently weaker than the descriptive results.

major comments (4)
  1. [§4.3 with §3.3 and §3.4] The central non-decline result in Figure 6 is a regression of the yearly ratio of "historical" documents, a variable constructed from (i) the authors' manual "#historical" topic tags (§3.3) and (ii) the 50% cumulative-likelihood document-assignment cutoff (§3.4). The authors explicitly call both choices ad hoc and concede that the tags "may be open to discussion" (§5). Because a document can be assigned to several topics and the ratio is therefore a nonlinear function of these choices, the reported r(38)=0.20, p=0.8885 is not a stable estimate without inter-annotator reliability or a sensitivity analysis. I ask for sensitivity analyses at other cutoffs (e.g., 25% and 75%), an examination of leave-one-out removal of contested tags, and, if possible, a small validation study with an independent rater for the #historical labels.
  2. [§3.1 and Table 1] The methods state that the corpus consists of 875 full-text articles, but after preprocessing "we obtained a corpus of 816 documents" with no explanation for the 59-document drop. Notes 4 and 5 later refer to 875 as the corpus size. Since all proportions in §4.1–§4.3 use document counts, the manuscript must specify whether the 875 or 816 base is used, why some documents were excluded, and whether the exclusions could affect the historical ratio.
  3. [§3.2] The choice K=90 is made using quantitative metrics combined with manual inspection, and no robustness check with respect to K is reported. Topic granularity directly determines which topics receive a "#historical" tag and hence the ratio in Figure 6; the main descriptive percentages and the non-decline finding could change at other values of K. I request that the central claims be rechecked at adjacent choices of K (e.g., 60 and 120), or that a stability analysis of the historical ratio across K be reported.
  4. [§4.3 and Figure 6] The one-sided test H1:r<0 returns p=0.8885, which only fails to reject the null; it does not establish that the historical ratio is constant, especially with 40 yearly observations and a 95% confidence interval that includes both negative and positive slopes. The sentence "it appears more likely that the ratio is constant" is stronger than the evidence justifies. I recommend reporting the slope and confidence interval, and an explicit equivalence test or power analysis if the claim is that there was no meaningful decline.
minor comments (5)
  1. [§4.3] The text after Table 4 contains the typo "Riceour"; it should be "Ricoeur".
  2. [§4.1] The heading-like phrase "Metaphyics and Epistemology" contains a typo; it should be "Metaphysics and Epistemology".
  3. [Figure 6] The y-axis label "Historical Docs. / T otal Docs." has an unnecessary space in "T otal".
  4. [Table 3 note] The note that two subareas have exactly the same number of items and most probable words is ambiguous; if this arises from one single topic tagged with multiple subarea labels, the table should state that directly rather than presenting them as separate aggregated subareas.
  5. [§3.4] The assignment procedure would be clearer if the manuscript stated how documents assigned to more than one #historical topic are counted in the numerator of the yearly historical ratio; the current description leaves this ambiguity.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the topic-model outputs are unsupervised, the main-area labels come from the external PhilPapers taxonomy, and the no-decline regression is computed from data rather than forced by the manual #historical tag or the ad-hoc 50% cutoff.

full rationale

The paper's central quantitative claims are descriptive outputs of an unsupervised Dynamic Topic Model, not derivations from the target conclusions. The main-area labels are anchored to the external PhilPapers taxonomy (Sec. 3.3), so the finding that value theory and epistemology are prominent is an empirical labeling result rather than a definitional tautology. The historical-decline test uses a manual '#historical' topic tag, applied before the time-trend regression; the tag encodes whether a topic has a dominant historical emphasis, not whether such topics decline over time. The regression coefficient r(38) = 0.20 is therefore computed from data and is not identical to the tag definition. The 50% cumulative-likelihood document-assignment cutoff is explicitly admitted to be ad hoc (Sec. 3.4), and the authors concede that the historical tags 'may be open to discussion' (Sec. 5). These are measurement-validity and robustness limitations, not circular reductions: changing the threshold or the tags would change the estimated ratio, but the result would still be an empirical quantity, not a restatement of the choice itself. No load-bearing self-citations appear; the DTM implementation is credited to Blei and Gerrish and the comparative topic-modeling studies are external. Thus no specific equation or construction step can be quoted where a claimed prediction reduces to its input by definition.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The paper introduces no new theoretical entities, forces, or mechanisms. Its central claims rest on the choice K=90, the 50% document-assignment threshold, manual topic labeling, and assumptions about OCR correction and journal representativeness.

free parameters (3)
  • Number of topics K = 90
    Chosen from grid K in {50, 60, ..., 150} using coherence, perplexity, empty-topic counts, and manual inspection (Sec. 3.2). All topic counts and ratios depend on this choice.
  • Document assignment cutoff = 50% of cumulative topic likelihood
    Documents are assigned to a topic if they account for the top 50% of that topic's total likelihood; the authors call this choice ad-hoc (Sec. 3.4). It determines the document counts in Tables 3 and 4 and the historical ratio.
  • DTM time-slice structure = Not stated in main text
    The number and width of time slices affect how smoothly topics evolve, but the paper does not specify whether slices are single years or multi-year bins.
assumptions (5)
  • standard math Bag-of-words representation and the LDA/DTM generative assumptions (exchangeable words, Gaussian random-walk topic drift) adequately model philosophical text.
    Invoked in Sec. 2.2 and Appendix A; if these assumptions fail, topic estimates inherit the error.
  • domain assumption The single journal Ideas y Valores is an adequate lens on Colombian and Latin American academic philosophy.
    Stated in the Introduction as 'a lens through which we can observe' the development of philosophical ideas; regional generalizations rest on this assumption.
  • ad hoc to paper Manual topic labeling by the authors using the PhilPapers taxonomy produces valid main-area and #historical tags.
    Sec. 3.3 acknowledges that topic interpretation requires researcher judgment; the historical-decline regression in Sec. 4.3 uses these tags without inter-annotator validation.
  • domain assumption Orthographic correction with PySpellChecker and a custom dictionary improves OCR text without systematically changing topic word distributions.
    Sec. 3.1 reports improved word recognition rates but does not test whether corrections bias older documents differently from newer HTML articles.
  • ad hoc to paper The 50% cumulative-likelihood cutoff for assigning documents to topics is appropriate.
    Sec. 3.4 explicitly labels this as an ad-hoc choice and leaves neutral criteria for future research.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A History of Philosophy in Colombia through Topic Modelling." pith.science (2026). https://pith.science/paper/RZIUM52Q

@misc{pith2026241204236,
  author       = {Pith},
  title        = {Pith review of: A History of Philosophy in Colombia through Topic Modelling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RZIUM52Q}},
  note         = {Machine review of arXiv:2412.04236}
}
read the original abstract

Data-driven approaches to philosophy have emerged as a valuable tool for studying the history of the discipline. However, most studies in this area have focused on a limited number of journals from specific regions and subfields. We expand the scope of this research by applying dynamic topic modelling techniques to explore the history of philosophy in Colombia and Latin America. Our study examines the Colombian philosophy journal Ideas y Valores, founded in 1951 and currently one of the most influential academic philosophy journals in the region. By analyzing the evolution of topics across the journal's history, we identify various trends and specific dynamics in philosophical discourse within the Colombian and Latin American context. Our findings reveal that the most prominent topics are value theory (including ethics, political philosophy, and aesthetics), epistemology, and the philosophy of science. We also trace the evolution of articles focusing on the historical and interpretive aspects of philosophical texts, and we note a notable emphasis on German philosophers such as Kant, Husserl, and Hegel on various topics throughout the journal's lifetime. Additionally, we investigate whether articles with a historical focus have decreased over time due to editorial pressures. Our analysis suggests no significant decline in such articles. Finally, we propose ideas for extending this research to other Latin American journals and suggest improvements for natural language processing workflows in non-English languages.

Figures

Figures reproduced from arXiv: 2412.04236 by the authors.

Figure 3
Figure 3. Ratio of words recognized by PySpellChecker before and after orthographical correction. A word is recognized if it is contained in PySpellChecker’s default Spanish, English or German dictionaries, RAE’s 2020 frequency list, or if it is contained in articles that were available in HTML format. Each point represents a single document, and its location represents how many words are recognized before and after correctio… view at source ↗
Figure 7
Figure 7. Samples from a Dirichlet distribution with [PITH_FULL_IMAGE:figures/full_fig_p018_7.png] view at source ↗
Figure 8
Figure 8. How topic proportions evolve over time slices. At each time step, the new topic proportions are slight modifications of the previous ones, de￾fined by sampling from a Gaussian distribution in a high-dimensional space. This can be expressed mathematically with the relationship αt | αt−1 ∼ N (αt−1, δ2 I). To transform these real numbers αt into probabil￾ity vectors, LDA relied on considering them as pa￾rameters of a D… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

12 extracted references · 11 canonical work pages

  1. [1]

    Deciding the document’s length, or number of words N

    work page

  2. [2]

    Say we have K topics in total, then this would correspond to specifying a vector θ = (θ1,

    Deciding the proportions of each topic to use. Say we have K topics in total, then this would correspond to specifying a vector θ = (θ1, . . . , θK) such that all entries add up to 1.0. In other words,θk corresponds to the proportion (or percentage) of topick in the document. Each topic β1, . . . ,βK is a vector of percentages over all words

    work page

  3. [3]

    a”, “b”, and “c

    For each of the N words, the author first decides which topic it belongs to according to the pro- portions in θ. Say topic k was selected. The author then decides on which word to choose next according to the vector of proportionsβk = (βk1, . . . , βkV ), where V is the total number of unique words in all documents. As a minimal example, say we have two t...

    work page

  4. [7]

    Deciding on the document’s length N

    work page

  5. [8]

    , θK) from a Dirichlet(α) where α = (α1,

    Sampling topic proportions θ = ( θ1, . . . , θK) from a Dirichlet(α) where α = (α1, . . . , αK)

    work page

  6. [9]

    ,βK are estimated from all doc- uments using statistical inference

    For word wn, the author decides on a topic by choosing on an index k according to the proba- bilities in θ, and then choosing on a word from the vocabulary according to the probabilities in βk.6 In LDA, the Dirichlet parameter α and the topic proportions β1, . . . ,βK are estimated from all doc- uments using statistical inference . These parame- ters dete...

    work page 2003

  7. [10]

    Evolves all topics from the previous time slice: βk,t | βk,t−1 ∼ N(βk,t−1, σ2I)

    work page

  8. [11]

    Evolves the topic proportions from the previous time slice: αt | αt−1 ∼ N(αt−1, δ2I)

    work page

Show all 12 references
  1. [12]

    (b) For each word: i

    For each document: (a) Chooses the topic proportions for the doc- ument by drawing η ∼ N(αt, a2I) and mapping it to a probability vector using Softmax. (b) For each word: i. Decides on a topic k according to the proportions on Softmax(η). ii. Chooses a word w from topick by sa...

  2. [210]

    doi: 10.1007/978-3-031-48956-3 10

    Springer International Publishing, 2024. doi: 10.1007/978-3-031-48956-3 10. C. A. L´opez. Cultura y tecnificaci´on en los primeros fil´osofos colombianos de la normalizaci ´on. Ideas y Valores , 70:79–109, 2021. doi: 10.15446/ ideasyvalores.v70n7Supl.95336. C. Malaterre, J.-F....

    2024 doi

  3. [2010]

    http://is.muni.cz/publication/ 884893/en

    ELRA. http://is.muni.cz/publication/ 884893/en. L. Richardson. Beautifulsoup4: Screen-scraping library, 2020. URL http://www.crummy.com/ software/BeautifulSoup/bs4/. SCImago. SJR - SCImago journal & country rank. Available at: http://www.scimagojr.com (Ac- cessed: 3 December 2...

    2020

  4. [2018]

    URL https://gitlab.com/pdftools/ python-ghostscript. D. Greene, J. O’Sullivan, and D. O’Reilly. Topic mod- elling literary interviews from The Paris Review. Digital Scholarship in the Humanities , 2024. doi: 10.1093/llc/fqad098. M. Grootendorst. Bertopic: Neural topic modeling...

    2024 arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.