Lumpers and Splitters: Understanding the Great Debate of Classification
From the way we organize our digital files to how scientists categorize the natural world, humans have an innate drive to classify. However, the method of classification is rarely universal. In almost every academic discipline that requires placing examples into defined categories, a fundamental tension exists between two opposing philosophical approaches: lumping and splitting.
At its core, the lumper-splitter problem arises when researchers must decide whether to emphasize the similarities between objects or their differences. A "lumper" is someone who assigns examples broadly, believing that signature similarities are more important than minor differences. Conversely, a "splitter" creates precise definitions and establishes new categories to classify samples that differ in key ways.
Key Facts
- Lumpers prioritize similarities to create broad, inclusive categories.
- Splitters prioritize differences to create narrow, precise categories.
- The terms were used as early as 1845 by naturalist Edward Newman.
- In software modeling, it is generally easier to lump existing splits than to split existing lumps.
- The conflict appears across diverse fields, including biology, linguistics, psychiatry, and neuroscience.
The Origin of the Terms
While many previously believed Charles Darwin first used these terms in an 1857 letter to Joseph Dalton Hooker, research by Glenn Branch of the NCSE indicates the credit belongs to naturalist Edward Newman. Writing in 1845, Newman argued for the necessity of discarding "imaginary species" while admitting new ones, noting that the talents of "hair-splitting" and "lumping" must yield to the "power of Truth."
The terms gained wider academic traction through George G. Simpson's 1945 work, The Principles of Classification and a Classification of Mammals. Simpson highlighted the extremes of both sides: critics of splitters argued that if they could tell two animals apart, they placed them in different genera; critics of lumpers joked that if a carnivore wasn't a dog or a bear, a lumper would simply call it a cat.
By 1969, the terminology had reached medical genetics via Victor McKusick, and by 1975, it entered the humanities. During a debate in the Times Literary Supplement, J.H. Hexter accused historian Christopher Hill of "lumping"—ignoring evidence that didn't fit a coherent pattern and emphasizing similarities to support a specific theory, whereas "splitters" preferred to embrace incoherent complexity.
Lumping and Splitting Across Disciplines
Biology and Taxonomy
In biology, naming a species is essentially a hypothesis about evolutionary relationships. When taxonomists realize two differently named organisms are actually the same species, they perform synonymisation—a form of lumping where the older name is retained based on the "priority of nomenclature." Splitting occurs when a single taxon is divided into multiple new ones.
This tension is evident in the classification of ferns. The Pteridophyte Phylogeny Group I (PPG I) uses 319 fern genera, a number defended by researchers like Schuettpelz et al. (2018) as a reflection of improved data and diversity appreciation. However, critics like Christenhusz and Chase argue that this "excessive splitting" destabilizes nomenclature and makes identification difficult because the resulting genera have few distinguishing characteristics.
History and the Arts
In historical study, lumpers create broad definitions covering vast time periods, which can lead to unwieldy categories where members have little in common. Splitters, meanwhile, create tight groups of inter-relationships, which can result in "distinctions without difference" and a failure to see overarching patterns.
A prime example is the term "Romantic." To a splitter, it might refer strictly to German poetry between 1780 and 1810. To a lumper, it could encompass every composer from Hummel to Rachmaninoff and beyond.
Software Modeling
In software engineering, specifically model-driven architecture, lumpers generalize by creating a few broadly defined objects. Splitters avoid generalization, producing many narrowly defined objects. A key principle here is that "splits can be lumped more easily than lumps can be split." For instance, combining two narrow classes with identical error messages is simple, but splitting a broad class requires examining every single object to ensure the split is accurate.
Language Classification
Historical linguists often clash over how much evidence is required to group languages into a family. Splitters rely strictly on the comparative method—the reconstruction of a common ancestor or protolanguage—to prove kinship. This approach is common among recent American linguists.
Lumpers are more open to lexicostatistics (the quantitative comparison of basic vocabulary) and mass typological comparison. They are more tolerant of uncertainty regarding whether similarities result from a common ancestor (divergence) or borrowing (convergence). This approach is prominent in the Moscow School of Comparative Linguistics, led by figures such as Vladislav Illich-Svitych and Sergei Starostin.
Religious Studies and Philosophy
In the study of early Christian liturgy, lumpers seek a single line of successive texts from the apostolic age, while splitters see overlapping, intermingling strands. Similarly, the concept of a single "Hindu religion" is a lumper's view (Smartism), whereas splitters identify as adherents of specific traditions like Shaivism, Vaishnavism, or Shaktism.
In philosophy, physicist Freeman Dyson suggested a broad division: splitters correspond to materialists (who see the world as divided into atoms), while lumpers correspond to Platonists (who see the world as composed of ideas).
Psychiatry and Neuroscience
Psychiatric diagnosis is often a battle between these two views. Splitters emphasize the heterogeneity (diversity) within diagnostic categories to justify more specific labels. Lumpers focus on similarities to create broader entities. For example, a lumper might categorize a patient's state as "stress," while a splitter would distinguish between worry, grief, or a specific anxiety disorder.
Neuroscience links these tendencies to "uncertainty aversion" (splitters) and "uncertainty tolerance" (lumpers). Neuroscientist Marc-Lluís Vives explains that while we must categorize stimuli to navigate the world (e.g., recognizing all mugs as "mugs"), the degree to which we distinguish a glass mug from a ceramic one affects how we map the stimulus to a response.
Computational Linguistics and NLP
In Natural Language Processing (NLP), the debate appears in Word-Sense Disambiguation (WSD)—the process of identifying which meaning of a word is used in a specific context. Researchers in WSD tend to be lumpers, often ignoring subtle dictionary distinctions that aren't salient to the computer program. Tools like Word2Vec are used to quantify the overlap or distinction between these semantic categories in general usage.
Comparison Summary
| Feature | Lumpers | Splitters |
|---|---|---|
| Primary Focus | Similarities and patterns | Differences and precision |
| Category Size | Broad and inclusive | Narrow and exclusive |
| Risk | Over-generalization / Unwieldy definitions | Over-specification / "Distinctions without difference" |
| Biological Action | Synonymisation | Taxonomic splitting |
| Philosophical Lean | Platonism / Uncertainty tolerance | Materialism / Uncertainty aversion |
Frequently Asked Questions
What is the main difference between a lumper and a splitter?
A lumper focuses on the similarities between examples to create broad, inclusive categories, while a splitter focuses on the differences to create narrow, precise, and numerous categories.
Why is it easier to lump than to split in software modeling?
Lumping is simpler because if two narrow classes behave identically, they can be merged immediately. Splitting a broad class is more complex because every object within that class must be analyzed to determine where the boundaries of the new, narrower classes should be.
How does the lumper-splitter debate affect biology?
It affects how species are named and grouped. Lumpers may merge multiple species into one through synonymisation, while splitters may divide a single group into several new taxa based on new data or a greater appreciation for diversity.
What is the "comparative method" in linguistics?
The comparative method is a rigorous technique used by linguistic splitters to prove kinship between languages by reconstructing a common ancestor or protolanguage, rather than relying on general similarities.
How does neuroscience explain these two tendencies?
Neuroscience suggests that lumping is associated with "uncertainty tolerance," while splitting is associated with "uncertainty aversion," affecting how the brain represents semantic categories and distinguishes between similar stimuli.