The Khipu Database Project began in the fall of 2002, with the goal of collecting all known information about khipu into one centralized repository. Having the data in digital form allows researchers to ask questions about khipu which up until
now would have been very difficult, if not impossible, to answer. The Khipu Database Project was funded 2002-2004 by the National Science Foundation and Harvard University, and in 2004-2005 is funded by the National Science Foundation.
The KDB and its associated data entry application were designed and implemented specifically for the use of this project. The khipu data schema is modeled on the physical structure of khipu. The overall structure of a khipu is that of a branching network in which the number of branching levels is highly variable, but in which components at every level share certain characteristics. The data schema for the KDB embraces the following critical facts about khipu construction: the interlocking relationships between khipu components, the branching or tree-like structure of khipu, the similarity of certain components, and the multi-dimensionality of khipu variables.
In a relational database, each table may be linked to one or many different tables by defining correspondences between data fields in each table. These relationships can be complex, including restrictions on the possible data in one record given the data in another. Such a structure is ideal for describing a flexible object such as a khipu. Khipu components are specified in detail in their own records and linked into their proper places in the entire object through carefully designed relationships. In this way, the database builds a network or web of correspondences between khipu parts. This allows the database to mimic the physical structures of a khipu without loss of accuracy. It should be noted that the current design allows complete freedom in capturing khipu structure; the number of pendants that belong to a primary cord or knots that belong to a pendant are infinitely variable. Similarly, the database can accommodate any number of levels of subsidiaries.
Certain aspects of khipu share many characteristics. For example, pendant cords at any level (top cords, pendants, subsidiaries, etc) have variables of fiber, final twist, end treatment, length, and color. Similarly, all knots on a khipu have a position on a particular string, a type, directionality, and a numerical value. By creating tables that incorporate these common elements for cords or knots at all levels, we increase the efficiency of our data structure while still allowing it to be extensible. Finally, some variables may themselves have many dimensions; color is the most obvious example. One cord may be composed of several different colors, and may even change color along its length. The database effectively and accurately contains color information by allowing many different color records for one cord. As other variables become known and are recorded, the database can be easily extended to completely contain new information, without compromising existing data.
This diagram shows the main tables in the Khipu Database and their relationships.
Because of the comprehensiveness and flexibility of the KDB, we have been able for the first time to ask questions about aspects of the khipu corpus as a whole.Our first queries have focused on three areas: searching for numerical similarities
between khipus, statistical analysis of khipu characteristics, and searching for correlations between khipu variables.
We began querying the KDB by asking “Are there khipus in which the sequence of cord total values match?” We were able to discover several previously unknown matches, that is, pairs of khipu in which portions of the sequence of total cord values are the same. In analyzing these exact matches, it became clear that there were larger portions of each khipu that were very similar though not exactly equal. This led us to investigate non-exact matches. For instance, using the database, we can ask questions such as “Show all khipu with cord total sequences in which a value on the first khipu is within 1 of the value on the second khipu.” (This would find, for example, a correspondence between 10–16–4–5–5 and 9–17–3–6–6.) Or, “Show all khipu with numeric sequences in which a number on the first khipu is within 10% of a number on the second.” Using this approach, we were able to find several other khipu with significant similarities. It should be emphasized that these searches will find matching cord value sequences in any position on any khipu; that is, a sequence near the beginning of one khipu may appear in the middle of another. Reversed sequences can also be found. We have also examined proportions between succeeding numbers in a sequence, and expect that as more khipu are entered, many more such matches will appear.
The database allows us easily to group together similar elements of all khipu and determine overall distribution of khipu characteristics. For instance, we can ask “What is the most frequently occurring total cord value?” For the entire body of khipu data, it turns out so far that 1 is overwhelmingly the most frequently occurring value. The next most frequently occurring numbers, in order, are 2, 3, 4, 10, and 5. Interestingly, this distribution may change when we look at khipu from one particular area: in khipu from Pachacamac, for instance, the most frequently occurring total cord values are 20, 5, 10, 1, and 50.
One portion of our color investigations to date has focused on color distribution in cords with more than one hue. We asked “Which colors can occur together in barber-pole cords?” As with the question about number frequency, the answer depends on the provenance of khipus included. Overall, 31 colors appear in barber-pole cords, with white, light brown, and light yellowish brown being found most frequently. Most other colors appear combined with one of these. In Pachacamac, by contrast, only ten distinct colors appear in barber pole cords, and of those, only one (dark grayish blue) ever appears with a color other than white. Similar queries were run for Ica and Leymebamba, and each region shows distinct differences in the use of color in barber-pole and mottled cords. The differences between regions allow us to better understand variations in khipus from different areas, which may provide the foundation for defining khipu typologies and classifications.
A very important question is whether there is a correlation between two or several khipu construction variables. Because the Laguna de los Cóndores collection has the most complete information thus far on attachment, cord twist, and knot direction, we began our inquiries with those khipu. We found that in general, long knots occur with approximately equal likelihood on a recto- or verso- attached pendant. However, long knots (S or Z) with a value of 5 show a marked difference between the percentage occurring on recto- and the percentage occurring on verso- attached cords. This may indicate a correlation between long knots of value 5 and pendant cord attachment. To determine whether such a correlation is supportable, we will compare these results to attachment data from other khipu when they become available.