The Indus Valley civilization — Harappa, Mohenjo-daro, and an undeciphered script

Intuition Beginner

The companion unit 32.04.01 introduced the Indus Valley Civilization: a Bronze-Age urban culture that covered more than a million square kilometers, built grid-planned cities, and left a script no one can read. This unit goes deeper into the three puzzles that make the Harappans unlike any other early civilization.

First, their engineering. Between roughly 2,600 and 1,900 BCE, cities such as Mohenjo-daro and Harappa were laid out on grid plans, built from standardized baked bricks, supplied by hundreds of private wells, and served by covered drains running beneath nearly every street. No contemporary civilization in Egypt or Mesopotamia matched this level of civic sanitation. A traveler walking through Mohenjo-daro around 2,500 BCE would have found cleaner streets than many European cities two thousand years later.

Second, the script. Over 4,000 inscribed objects survive, carrying about 400 distinct symbols. Yet not one text runs longer than 26 signs, and most carry only five. No bilingual inscription has ever been found. The central question — whether these symbols encode a spoken language at all, or whether they are a non-linguistic sign system closer to medieval heraldry — is still open after a century of effort.

Third, the ending. Around 1,900 BCE the great cities emptied out over generations, not in a single catastrophe. The old colonial story blamed a violent "Aryan invasion," but archaeology finds no destroyed cities and no mass graves. The real picture is slower: rivers shifted, the climate dried, trade with Mesopotamia faltered, and people dispersed east into the Ganges plain.

What makes the Indus case instructive is that we must reason about a complex society almost entirely from things — bricks, seals, weights, drains — rather than from words. Every claim about Harappan kings, gods, or laws is a reading of material evidence, not of texts.

Visual Beginner

Feature	Mature Harappan	Mesopotamia	Egypt
Mature urban phase	c. 2,600–1,900 BCE	c. 3,500–1,600 BCE	c. 3,000–1,070 BCE
Known sites	~1,500	several hundred	a few hundred
Street plan	Grid-aligned, hierarchical widths	Organic, irregular	Temple/palace-focused
Drainage	City-wide covered sewers	Limited, palace areas	Limited
Largest city (population)	Mohenjo-daro, ~40,000	Uruk, ~50,000–80,000	Thebes, ~40,000+
Writing	Undeciphered, ~400 signs, max 26 signs	Cuneiform, deciphered, long texts	Hieroglyphs, deciphered, long texts
Royal monuments	None found	Ziggurats, palaces	Pyramids, temples

Worked example Beginner

Three numbers anchor the Indus case.

First, the scale. Roughly 1,500 Harappan sites have now been identified, scattered across the Indus floodplain, the Ghaggar-Hakra basin, Gujarat, and the highlands of Balochistan. The Mature Harappan phase — the window during which the great cities were occupied and the script was in use — runs from about 2,600 to 1,900 BCE, a span of some 700 years. That is comparable to the lifetime of the Roman Empire, and longer than the whole history of the United States.

Second, the Great Bath of Mohenjo-daro. This tank measures about 12 meters long, 7 meters wide, and 2.4 meters deep. Its volume is roughly $12\times 7\times 2.4=201.6$ cubic meters, about 200,000 liters of water. The tank was built from fitted bricks sealed with bitumen to hold water, with adjacent rooms for changing or ritual use. The engineering precision required to keep a 200,000-liter tank watertight, on a mud-brick platform, is what makes the structure exceptional.

Third, the script in aggregate. With about 4,000 surviving inscribed objects averaging close to 5 signs each, the entire legible Harappan corpus amounts to roughly 20,000 individual sign-instances — drawn from a repertoire of about 400 distinct signs. Compare that with Mesopotamia, where single archive complexes contain tens of thousands of long tablets. The brevity of the Indus texts is itself a clue: a civilization of this scale that used writing mainly for short seal-labels was using it very differently from its neighbors.

Check your understanding Beginner

Formal definition Intermediate+

The working terms of this unit, fixed against the modern periodisation in Wright (2009) ^{[Wright 2009]} and Possehl (2002) ^{[Possehl 2002]}.

Mature Harappan phase. The interval c. 2,600–1,900 BCE during which the characteristic urban assemblage — grid-planned cities, standardised baked bricks in the ratio 4:2:1 (length to width to height), a binary-cum-decimal system of stone weights with a standard unit near 13.7 g, the Indus script on seals and tablets, and a uniform repertoire of ceramic and metal forms — appears across the Indus and Ghaggar-Hakra basins. The preceding Early Harappan (c. 3,300–2,600 BCE) and following Late Harappan (c. 1,900–1,300 BCE) phases are treated in the companion unit 32.04.01.

Harappan urban planning. A settlement pattern in which an elevated western "citadel" or acropolis (carrying public structures such as the Great Bath, granaries, and assembly halls) is separated by open ground from a larger "lower city" laid out on a grid of arterial streets. Street widths follow a hierarchy: main avenues roughly 9 m, secondary streets 4–5 m, residential lanes 1.5–2 m. Domestic units open onto interior courtyards and connect through clay pipes to brick-lined, covered sewers with maintenance access.

Meluhha. The toponym used in Akkadian-period Mesopotamian texts (c. 2334–2,000 BCE) for a distant trading partner at the head of the Persian Gulf trade, almost universally identified with the Mature Harappan civilization on the combined evidence of Harappan seals, carnelian beads, etched carnelian, and weight-stones recovered at Mesopotamian sites (Ur, Lagash, Eshnunna, Kish) and Mesopotamian cylinder seals and goods at Harappan ports (Lothal).

De-urbanisation. The protracted process (c. 2,000–1,700 BCE) by which the Mature Harappan cities were depopulated, the script fell out of use, long-distance trade contracted, and settlement shifted eastward toward the Ganges-Yamuna doab and southward into Gujarat. The term is preferred to "collapse," which implies a sudden catastrophic event the archaeological record does not support.

The Indus script (corpus statistics). The inscribed corpus comprises $N\approx 4,000$ objects carrying $V\approx 400$ distinct signs. Denote by $S=(s_1,s_2,\dots ,s_{\ell })$ a sequence of signs of length $\ell$. The empirical distribution of $\ell$ is heavily concentrated at small values (median near 5, a single outlier at $\ell =26$). The conditional entropy of a sign given its predecessor,

$$H(S_n\mid S_{n-1})=-\sum _{i,j}p(s_i,s_j)\log p(s_i\mid s_j),$$

is the quantity at the center of the decipherment debate: writing systems that encode natural language exhibit conditional entropy in a characteristic band, and the dispute concerns whether the Indus corpus falls in that band.

Case study Intermediate+

Two positions on the script question

The single most contested question in Indus scholarship is whether the symbols constitute a writing system at all. Two live positions structure the literature, and the evidence bears on each in distinct ways.

Position 1 — the linguistic position (Parpola; Mahadevan; Rao et al.). Asko Parpola (1994) and the late Iravatham Mahadevan devoted decades to compiling the sign corpus and arguing, on the basis of sign frequencies, positional regularities, and the structure of terminal "suffix" signs, that the Indus script encodes a language — most probably an early Dravidian language. The most influential recent statement is Rao, Yadav, Vahia, Joglekar, Adhikari and Mahadevan's 2009 paper in Science ^{[Rao 2009]}, which computed the conditional entropy of sign-ordering in the Indus corpus and compared it against natural-language corpora (Sumerian, English, Tamil), Type 1 non-linguistic systems (DNA, protein sequences), and Type 2 non-linguistic systems (Vinca signs, medieval heraldry). The Indus conditional entropy fell firmly within the band occupied by natural-language writing systems and outside the band of the non-linguistic comparanda. The authors concluded that the sign sequences carry the kind of sequential structure characteristic of language.

Position 2 — the skeptical position (Farmer, Sproat, Witzel). Steve Farmer, Richard Sproat, and Michael Witzel argued in 2004 ^{[FarmerSproatWitzel 2004]} that the Indus symbols are not a full writing system and may not encode language at all. Their argument rests on three observations. First, the texts are extraordinarily short — a mean near five signs, a maximum of 26, against the thousands of signs routinely found on a single Mesopotamian or Egyptian text. Second, no long inscriptions exist on any of the durable media (stone, metal, terracotta) that would have preserved them, even though the civilization was plainly capable of fine carving; a literate society of this scale, they contend, would have left longer records. Third, the symbols show little of the repetitive, grammatical patterning (fixed word boundaries, recurring suffix strings, sign-reduplication encoding plurality or tense) that genuine writing systems exhibit at this corpus size. Farmer, Sproat, and Witzel propose instead that the signs are a system of political and religious symbols — emblems of office, clan markers, ritual icons — comparable in function to medieval European heraldry or Near Eastern pot-marks: ordered for visual display, not for encoding speech.

The two positions need not be exhaustive, and a careful reading keeps both open. The conditional-entropy result of Rao et al. shows that Indus sign-sequences are not random and share statistical structure with language; it does not, by itself, prove the signs encode a specific spoken language, because some non-linguistic systems can also produce non-random ordering. The Farmer–Sproat–Witzel critique correctly stresses the brevity and the absence of long texts, which is genuinely anomalous for a putative full script of a complex society; but the inference from "no long texts survive" to "no long texts ever existed" requires an assumption about preservation that perishable writing media (bark, palm-leaf, cloth, wood) would defeat. Kenoyer (1998) ^{[Kenoyer 1998]} and McIntosh (2008) ^{[McIntosh 2008]} present the matter as unresolved, treating the script as a writing system by analogy while conceding that decisive proof awaits either a bilingual text or a substantially longer inscription.

What the evidence settles is narrower than either camp's rhetoric. The corpus is too small, too short, and too little understood to be deciphered; the underlying language family is unknown; and no claim about Harappan governance, religion, or personal names rests on a reading of the text. The honest position, reflected in the standard syntheses, is that the script's status remains the single largest obstacle to reconstructing Harappan civilization on its own terms.

Bridge. The script debate builds toward every later unit on the relation between literacy and state power, because it isolates the question of whether writing is a necessary component of urban civilization; it appears again in the Mesopotamia unit 32.02.02, where cuneiform's administrative function is uncontested precisely because long bilingual and trilingual texts survive. The central insight is that the strength of any claim about an ancient society is bounded by the evidence that constrains it: with deciphered text, claims about governance and religion become precise, and without it they remain probabilistic. This is exactly the methodological asymmetry that separates Indus scholarship from Egyptology and Assyriology, and the bridge is the recognition that the script's resistance to decipherment is itself a fact about the civilization rather than a gap waiting to be filled.

Comparative framework Intermediate+

Two positions on the decline

The end of the Mature Harappan urban phase is the second contested question, and here the historiography is heavier because colonial politics distorted it for a century. Two defensible positions remain live; the violent "Aryan invasion" story is not among them.

Position 1 — environmental and economic multi-causation. The dominant scholarly position explains de-urbanisation as the cumulative effect of climatic and hydrological stress. Paleoenvironmental proxies (lake sediments, speleothems, pollen cores) register a weakening of the Indian monsoon beginning around 2,200–2,000 BCE, part of the global 4.2-kiloyear aridification event that also coincides with the collapse of the Akkadian Empire in Mesopotamia 32.02.02 and the political fragmentation of Old Kingdom Egypt. Concurrently, tectonic shifts appear to have diverted the Yamuna eastward toward the Ganges and the Sutlej westward toward the Indus, stripping the Ghaggar-Hakra system (plausibly the Saraswati of the Rigveda) of the perennial flow that had sustained a dense cluster of Harappan settlements; satellite imagery confirms large paleochannels consistent with a once-larger river. To this hydrological stress is added the contraction of the Mesopotamian trade network after the fall of Akkad (c. 2,154 BCE) and the Ur III state (2,004 BCE), which reduced demand for Harappan exports. On this reading the cities did not fall; they became economically and agriculturally unsustainable, and their populations dispersed over generations.

Position 2 — gradual migration and cultural recombination. A second position, sharpened by the ancient-DNA revolution since 2018, holds that the de-urbanisation of the Indus core was followed — not caused — by the gradual southward movement of steppe-pastoralist-descended populations carrying Indo-Aryan languages. Narasimhan et al. (2019) ^{[Narasimhan 2019]} and the Rakhigarhi genome (Shinde et al., 2019) together show that Indus-periphery individuals of the third millennium BCE lack the steppe genetic component, while South Asian genomes of the second millennium BCE begin to carry it, consistent with a gradual migration from the steppe through Central Asia. Crucially, this genetic signal arrives after the Harappan cities were already in decline, and no archaeological evidence of violent conquest accompanies it. The migration was a multi-generational admixture, not an invasion; the cities were not sacked because they had largely emptied before the migrants reached the northwest.

Retiring the colonial myth

The violent "Aryan invasion" thesis is a third position in the older literature, and it must be addressed because it dominated textbooks for decades. Its strongest statement was Mortimer Wheeler's, who in the 1940s attributed skeletal remains at Mohenjo-daro to a massacre and quipped that the Rigvedic god Indra "stands accused" of destroying the Harappan cities. The thesis collapsed under three blows. First, the skeletons Wheeler cited were re-examined and shown to belong to different strata and periods, not to a single violent event. Second, no excavation at any Mature Harappan site has produced a destruction layer, burned fortifications, or the mass graves that conquest leaves. Third, the Rigveda, composed centuries after the urban decline, cannot serve as eyewitness testimony to that decline. The colonial thesis projected a much later text onto a much earlier archaeology, and it did so within a racial framework ("Aryan" versus "Dravidian") that modern genetics has dissolved: both steppe-descended and older South Asian populations contributed to later South Asians, and neither maps onto the colonial racial categories. The defensible synthesis treats the urban decline as primarily environmental and economic, and the subsequent Indo-Aryan presence as a gradual migration whose linguistic consequences are real but whose military character is an invention.

This matters beyond antiquarian accuracy because the invasion myth was politically weaponized in both colonial and post-colonial India. Treating it as empirically dead is the precondition for reading the actual evidence, which is slower, less dramatic, and better attested.

The Mesopotamian comparison

The 4.2-kiloyear synchronicity gives the controlled comparison its force. Mesopotamia 32.02.02 and Egypt both experienced political fragmentation under the same climatic stress, yet both reconsolidated under new dynasties within a few centuries. The Indus did not. The difference is informative: where Mesopotamian and Egyptian centralised states could rebuild institutional capacity after a stress pulse, the Harappan urban system — which appears to have coordinated its cities without visible monarchy — had no institutional core to reconstruct. This suggests that the same environmental shock can produce reconsolidation in one political configuration and irreversible de-urbanisation in another, a comparative result that disciplines any single-cause account of either the rise or the fall of pristine civilizations.

Exercises Intermediate+

Advanced results Master

What the script statistics can and cannot show

The technical heart of the script debate is the conditional-entropy comparison of Rao et al. (2009), and its interpretation requires care that the popular summaries often skip. Conditional entropy measures the reduction in uncertainty about a sign once its predecessor is known; natural-language writing systems, computer languages, and musical notation all share a characteristic band precisely because all of them are sequentially structured. The Indus corpus falls within that band, which rules out the hypothesis that the signs are placed at random. What it does not rule out is the hypothesis that the signs are a non-linguistic but sequentially constrained system — ritual sequences, heraldic ordering, or a constrained sign-list used for display. Farmer, Sproat, and Witzel (2004) press exactly this point: their Type 2 comparanda (Vinca signs, heraldic emblems) are non-linguistic yet non-random, and they argue that the Indus statistics are consistent with that class. The defensible reading of the controversy is that the entropy result is necessary but not sufficient for the linguistic thesis: it removes one objection (randomness) without supplying the positive identification (a specific language) that would constitute decipherment. The gap between "non-random" and "linguistic" is the space in which the dispute lives, and on the present corpus it cannot be closed by statistics alone.

Why decipherment claims fail

A dozen or more decipherments have been proposed — Sanskrit, several Dravidian languages, Munda, Sumerian, even a proto-Indo-European reading. None has secured assent, and the reasons are structural rather than merely circumstantial. A decipherment requires three things simultaneously: a sufficiently long text to exhibit grammar, a bilingual key to anchor sign-to-sound correspondences, and a known candidate language. The Indus corpus supplies none reliably. The longest text (26 signs) is too short to test a grammar; no bilingual text exists; and the underlying language is unknown, so any proposed reading is underdetermined — the same short sequence can be "decoded" into several unrelated languages with equal plausibility, which is itself a sign that the decoding is projecting the decoder's language onto the signs rather than reading them. Possehl (2002) ^{[Possehl 2002]} catalogues the major attempts and the grounds on which each was rejected; the recurring failure mode is that a short, sign-poor corpus admits too many mutually incompatible readings for any one to be constraint-satisfying.

The de-urbanisation evidence in detail

The settlement-survey evidence assembled since the 1970s has reframed the decline as a regional transformation rather than a collapse. Three findings are load-bearing. First, the Mature Harappan settlement density along the Ghaggar-Hakra paleochannel was comparable to that along the Indus itself, and the abandonment of those eastern sites in the Late Harappan period coincides with the desiccation of the channel as the Yamuna shifted east to the Ganges and the Sutlej west to the Indus. Second, the eastern Punjab and western Ganges-Yamuna doab show a rise in settlement numbers during the Late Harappan, consistent with population relocation rather than disappearance; the people did not vanish, they moved. Third, material culture shows continuity across the transition — Late Harappan pottery (the Cemetery H complex at Harappa, the Jhukar complex in Sindh) develops from Mature Harappan forms rather than appearing abruptly, which counts against any model of population replacement at the urban core. The synthesis is a slow, multi-generational reorganization driven by hydrology and climate, in which the institutional apparatus of the Mature cities (standardised weights, the script, long-distance trade) was not rebuilt elsewhere because the specific agricultural and commercial base that sustained it had moved.

The genetic revolution and its limits

The ancient-DNA results since 2018 are the most consequential new evidence in a generation, and their correct interpretation requires distinguishing what they settle from what they do not. Narasimhan et al. (2019) ^{[Narasimhan 2019]} profiled individuals across Central and South Asia and identified three ancestral components in modern South Asians: Ancestral South Indians (indigenous to the subcontinent for tens of thousands of years), Iranian-related farmers (present by the fourth millennium BCE), and Steppe pastoralists carrying Y-haplogroup R1a-Z93 (detectable from about 2,000–1,500 BCE). Indus-periphery individuals from the third millennium BCE, including the Rakhigarhi genome, cluster with the first two components and lack the third. This establishes that the Mature Harappan population did not carry the steppe component, and that the steppe component arrived later, consistent with a migration in the second millennium BCE. What the genetics does not settle is the cultural mechanism: aDNA tracks biological ancestry, not language spread with full precision, and it does not describe the political or military character of the contact. The steppe signal is consistent with the spread of Indo-Aryan languages by a migration that was substantial enough to leave a genomic signature yet gradual enough to leave no archaeological trail of conquest — which is exactly the profile the Mesopotamian and Harappan archaeological records indicate. The genetics retires the racial framing of the colonial debate and supplies a dated population movement; it does not, and cannot, resuscitate the invasion narrative.

Why the Indus disciplines the convergent-bundle thesis

The comparative significance of the Indus case is that it decouples variables that the Mesopotamian and Egyptian cases bundle together. Mesopotamia and Egypt each present urban scale, monumental architecture, centralised monarchy, and (deciphered) full writing as a single package, which makes it tempting to treat that package as the definition of civilization. The Indus achieves urban scale, long-distance trade, and engineering standardization without monumental royal architecture, without decipherable evidence of monarchy, and — on the skeptical reading — possibly without full writing. It is therefore the controlled case that prevents the Mesopotamian bundle from being mistaken for a universal requirement. This is why every general theory of pristine civilization is tested against the Indus: a theory that explains Mesopotamia but cannot accommodate the Indus is a theory of Mesopotamia, not of civilization.

Synthesis. Putting these together — the engineering uniformity of Mature Harappan cities, the statistical ambiguity of the script, the environmental drivers of de-urbanisation, and the genetic evidence against any violent invasion — yields a picture in which the Indus case generalises across the curriculum as the controlled experiment that tests every general theory of pristine civilization: a society that achieved urban scale and long-distance trade without visible monarchy, monumental temples, or (on the skeptical reading) full writing. The foundational reason the Indus matters comparatively is that it decouples the variables Mesopotamia and Egypt bundle together; this is exactly why it disciplines the convergent-bundle thesis, and the bridge is that the same civilizational outcome can be reached by more than one institutional configuration. The central insight is that complexity does not require a single causal engine, and the Indus appears again in every comparative framework 32.02.02, 32.03.01 as the case that refuses to fit the Mesopotamian template.

Connections Master

Prehistory and the deep demographic substrate 32.01.01. The Mature Harappan population rests on a substrate whose formation the prehistory unit traces: the Ancestral South Indian and Iranian-related farmer components were already in place in the subcontinent millennia before the urban phase, and the Mehrgarh Neolithic tradition (from c. 7,000 BCE) supplies the indigenous agricultural base on which the cities were built. The genetic continuity between the Indus-periphery individuals and these earlier components is what makes the later steppe signal visible as an addition rather than a baseline.

The companion survey unit 32.04.01. This unit supplies depth on three questions — the script debate, the decline, and the comparative significance — that the survey unit introduces but cannot develop. The survey unit owns the Vedic corpus, the Upanishadic philosophical tradition, and the Mauryan transition; the two are designed to be read together, the survey for the regional and textual frame and this unit for the engineering, statistical, and historiographical depth.

Parallel pristine civilizations — Mesopotamia 32.02.02 and Egypt 32.03.01. The 4.2-kiloyear synchronicity gives these comparisons their force: the same climatic pulse that fragmented Mesopotamia and Egypt also stressed the Indus, yet only the Indus failed to reconsolidate. Reading the three cases together is what disciplines single-cause accounts of both rise and decline, and the Mesopotamia unit 32.02.02 supplies the deciphered-writing and imperial-administration baseline against which the Indus script and the absence of monarchy become legible as variations rather than deficits.

Historical & philosophical context Master

The modern recovery of the Indus Valley Civilization begins with Alexander Cunningham's 1875 publication of a Harappan seal he did not recognise, and takes institutional form in John Marshall's announcement of the civilization in 1924 and his three-volume report Mohenjo-daro and the Indus Civilization (1931) ^{[Marshall 1931]}. Marshall's report fixed the civilization in the scholarly imagination, but it did so within a colonial frame that measured the Indus against Mesopotamia and Egypt and found it wanting in precisely the categories — monumental royal architecture, deciphered inscriptions — those civilizations supplied as standards. The early literature treated the absence of palaces and pyramids as a deficit to be explained rather than as a positive feature of a different institutional configuration, an asymmetry Wright (2009) ^{[Wright 2009]} and Kenoyer (1998) ^{[Kenoyer 1998]} explicitly work to correct.

The decline question acquired its distorting political charge in the 1940s, when Mortimer Wheeler attributed the Mohenjo-daro skeletons to an "Aryan" massacre and cast the Rigvedic god Indra as the destroyer. This was an inference from a much later text to a much earlier event, made within a colonial racial taxonomy that divided "Aryan" invaders from "Dravidian" victims, and it entered textbooks for decades before the archaeological re-examination of the skeletons, the absence of destruction layers at any Mature Harappan site, and the genetic dissolution of the racial categories combined to retire it. Possehl (2002) ^{[Possehl 2002]} traces the thesis and its collapse in detail; the point for this unit is that a historiographical error, once institutionalised, can outlive its evidence by two generations.

The script question carries its own historiography. A century of decipherment claims — Sanskrit, Dravidian, Munda, Sumerian — has produced no consensus, and Farmer, Sproat, and Witzel's 2004 intervention ^{[FarmerSproatWitzel 2004]} reframed the question by asking whether the signs are writing at all. The ensuing exchange with Rao et al. (2009) ^{[Rao 2009]} is methodologically instructive: it is one of the few cases in ancient-history scholarship where a statistical argument and a philological argument have been brought into direct contact, and where the limits of each are visible. The honest position, reflected in the standard syntheses, is that the Indus script remains undeciphered, that its status as full writing is unresolved, and that this unresolvedness is a fact about the evidence rather than a failure of effort.

Bibliography Master

@book{wright2009ancient,
  author    = {Wright, Rita P.},
  title     = {The Ancient Indus: Urbanism, Economy, and Society},
  publisher = {Cambridge University Press},
  year      = {2009}
}

@book{possehl2002indus,
  author    = {Possehl, Gregory L.},
  title     = {The Indus Civilization: A Contemporary Perspective},
  publisher = {AltaMira Press},
  year      = {2002}
}

@book{kenoyer1998ancient,
  author    = {Kenoyer, Jonathan Mark},
  title     = {Ancient Cities of the Indus Valley Civilization},
  publisher = {Oxford University Press},
  year      = {1998}
}

@book{mcintosh2008ancient,
  author    = {McIntosh, Jane R.},
  title     = {The Ancient Indus Valley: New Perspectives},
  publisher = {ABC-CLIO},
  year      = {2008}
}

@article{farmersproatwitzel2004collapse,
  author  = {Farmer, Steve and Sproat, Richard and Witzel, Michael},
  title   = {The Collapse of the Indus-Script Thesis: The Myth of a Literate {H}arappan Civilization},
  journal = {Electronic Journal of Vedic Studies},
  volume  = {11},
  number  = {2},
  pages   = {19--57},
  year    = {2004}
}

@article{rao2009entropic,
  author  = {Rao, Rajesh P. N. and Yadav, Nisha and Vahia, Mayank N. and Joglekar, H. and Adhikari, R. and Mahadevan, Iravatham},
  title   = {Entropic Evidence for Linguistic Structure in the {I}ndus Script},
  journal = {Science},
  volume  = {324},
  number  = {5931},
  pages   = {1165},
  year    = {2009}
}

@article{narasimhan2019formation,
  author  = {Narasimhan, Vagheesh M. and Patterson, Nick and Moorjani, Priya and others},
  title   = {The Formation of Human Populations in {S}outh and {C}entral {A}sia},
  journal = {Science},
  volume  = {365},
  number  = {6457},
  pages   = {eaat7487},
  year    = {2019}
}

@article{shinde2019ancient,
  author  = {Shinde, Vasant and Narasimhan, Vagheesh M. and Rohland, Nadin and others},
  title   = {An Ancient {H}arappan Genome Lacks {A}ncestry from Steppe Pastoralists or {I}ranian Farmers},
  journal = {Cell},
  volume  = {179},
  number  = {3},
  pages   = {729--735.e10},
  year    = {2019}
}

@book{marshall1931mohenjo,
  author    = {Marshall, John},
  title     = {Mohenjo-daro and the Indus Civilization},
  publisher = {Probsthain},
  year      = {1931}
}

@book{parpola1994deciphering,
  author    = {Parpola, Asko},
  title     = {Deciphering the Indus Script},
  publisher = {Cambridge University Press},
  year      = {1994}
}

@book{wheeler1968indus,
  author    = {Wheeler, Mortimer},
  title     = {The Indus Civilization},
  edition   = {3rd},
  publisher = {Cambridge University Press},
  year      = {1968}
}