Can a computer help resolve the problem of ethnographic description?

James Kippen & Bernard Bel

Anthropological Quarterly (1989), 62, 3: 131-144

Abstract

A major prob­lem in ethno­graph­ic descrip­tion may be summed up as the search for ways to dis­en­tan­gle folk from ana­lyt­i­cal mod­els. Knowledge-based sys­tems have con­tributed to the devel­op­ment of for­mal struc­tures for the manip­u­la­tion of sym­bols asso­ci­at­ed with par­tic­u­lar phys­i­cal and con­cep­tu­al phe­nom­e­na. Importantly, their out­put can be inter­pret­ed by experts in the domain. This pro­vides eval­u­a­tion pro­ce­dures for mod­els elab­o­rat­ed joint­ly by ana­lysts and infor­mants. This paper describes an inter­ac­tive sys­tem in which knowl­edge is rep­re­sent­ed as pro­duc­tion rules in a for­mat derived from the the­o­ry of for­mal lan­guages. Modus ponens and modus tol­lens are explained and com­pared to deriva­tion schema­ta in first-order pred­i­cate log­ic. The results of an appli­ca­tion to the study of North Indian tabla drum­ming are assessed. We con­clude that (1) knowl­edge rep­re­sent­ed at a low the­o­ret­i­cal lev­el fails to descrim­i­nate between the input from infor­mants and the intu­itive assump­tions of ana­lysts, (2) exper­i­men­tal pro­ce­dures can be improved con­sid­er­ably if the sys­tem is designed to per­form auto­mat­ed knowl­edge acqui­si­tion (using prob­a­bilis­tic gram­mars and induc­tive learning).

Excerpts of an AI review of this paper (Academia, June 2025)

Summary of the Work

The arti­cle focus­es on the use of com­pu­ta­tion­al approach­es — par­tic­u­lar­ly knowledge-based sys­tems — to address the chal­lenge of pro­duc­ing con­sis­tent and sys­tem­at­ic ethno­graph­ic descrip­tions. The authors draw on their research with the Bol Processor (BP), a for­mal language-based sys­tem orig­i­nal­ly devel­oped to rep­re­sent and gen­er­ate sequences of ver­bal drum syl­la­bles (bols) in North Indian tabla per­for­mance. They describe how the BP’s gram­mar for­mal­ism, infer­ence engine, and mem­ber­ship tests facil­i­tate iter­a­tive inter­ac­tion between an ana­lyst and infor­mants (expert musi­cians). By demon­strat­ing the system’s abil­i­ty to pro­duce, rec­og­nize, and eval­u­ate per­mis­si­ble melod­ic or rhyth­mic vari­a­tions, the authors aim to illu­mi­nate how com­put­ers can be more deeply inte­grat­ed into anthro­po­log­i­cal and eth­no­mu­si­co­log­i­cal research.

Key Contributions

Formal Grammar and Pattern Representation: The paper offers a clear expla­na­tion of how context-free and context-sensitive gram­mars can be used to spec­i­fy cul­tur­al­ly valid musi­cal struc­tures. In par­tic­u­lar, the authors high­light the impor­tance of rep­re­sent­ing pat­terns and con­straints on musi­cal impro­vi­sa­tion in a gen­er­a­tive format.

Interactive Methodology: A cen­tral point is the iter­a­tive feed­back loop that allows infor­mants to assess computer-generated sequences and pro­vide cor­rec­tions. This “apprentice-like” inter­ac­tion under­scores how com­pu­ta­tion­al tools can help researchers refine the­o­ret­i­cal mod­els of cul­tur­al knowl­edge in near-real-time.

Probabilistic Grammars: The authors incor­po­rate a weight­ing sys­tem for pro­duc­tion rules, there­by account­ing for the rel­a­tive like­li­hood of dif­fer­ent musi­cal deriva­tions. This approach not only adds real­ism to the gen­er­a­tive process but also tack­les com­mon issues with pure­ly enu­mer­a­tive or ran­dom output.

Membership Testing: The mem­ber­ship test, which decides whether a new­ly pro­posed sequence belongs to a giv­en gram­mar, is described as an effi­cient, deter­min­is­tic bottom-up pars­er. This fea­ture is sig­nif­i­cant because it allows infor­mants to offer nov­el vari­a­tions while enabling the sys­tem to quick­ly judge their adher­ence to the emer­gent rule set.

Reflections on Folk vs. Analytical Models: The arti­cle rais­es impor­tant anthro­po­log­i­cal ques­tions about the bound­ary between infor­mants’ inter­nal­ized (folk) knowl­edge and the analyst’s for­mal (re)construction. The authors’ frank dis­cus­sion of the chal­lenges involved, such as the risk of over­for­mal­iz­ing or “antic­i­pat­ing” infor­mant knowl­edge, is method­olog­i­cal­ly relevant.

Strengths

• The paper empha­sizes the val­ue of com­bin­ing field­work with com­pu­ta­tion­al exper­i­men­ta­tion. This dual approach—grounded in real ethno­graph­ic and musi­cal practice—provides a com­pelling demon­stra­tion of how arti­fi­cial intel­li­gence tech­niques can advance the study of music and culture.
• Clear exam­ples illus­trate code struc­tures, pars­ing mechan­ics, and the gram­mar design. These con­crete details will help oth­er researchers adapt or repli­cate the Bol Processor approach in dif­fer­ent cul­tur­al or musi­cal contexts.
• The account of iter­a­tive knowl­edge acqui­si­tion high­lights a col­lab­o­ra­tive research dynam­ic, show­ing how insight flows back and forth between infor­mants and com­pu­ta­tion­al mod­els, rather than being a uni­di­rec­tion­al exer­cise of extract­ing information.

Conclusion

This study offers a thought­ful and inno­v­a­tive approach to bridg­ing ethno­graph­ic inquiry with com­pu­ta­tion­al mod­els. By show­ing how arrange­ments of sym­bol­ic musi­cal data can be sys­tem­at­i­cal­ly gen­er­at­ed, test­ed, and refined in con­junc­tion with human experts, the authors illus­trate a nov­el and promis­ing method for ethno­graph­ic descrip­tion. The paper encour­ages a nuanced under­stand­ing of both the poten­tial and lim­i­ta­tions of using expert sys­tems to cap­ture the com­plex­i­ty of cul­tur­al knowledge.

Download this paper