Hierarchical data and the derivational relationship between words

Andrew Hippisley, Mariam Tariq
University of Surrey

The experimental Node DataBlade (Brown 2001) is a software bundle that
extends the functionality of the object-relational database system Informix
by supporting the storage and manipulation of hierarchical data. We show
how this functionality provides a way of capturing specifically inheritance
relationships between linguistic data. One area where inheritance plays a
key role in the description of a set of linguistic expressions is in
derivational morphology. The partial derivational family of shkola 'school'
can be naturally represented hierarchically in the (partial) Shkola
hierarchy:

1. Shkola
1.1	shkol'nik 'pupil'
1.1.1	shkol'nicheskij 'schoolboyish'
1.1.1.1 shkol'nichesko 'in a schoolboyish manner'
1.2	shkol'nyj 'school' (adj)
1.3	shkolit' 'train'

The Node DataBlade stores node identifiers using the Dewey Decimal Scheme
and allows searches through the branches of the hierarchy using standard
sql. In our example the status of members of a derivational family can be
queried. We can create various queries to identify the relationship between
members of a derivational family, and queries to check the productivity of
a given family, i.e how many words does it derive. For example we can query
all the siblings of the Shkol'nik node to elicit all the co-derivatives of
shkol'nik, namely shkol'nyj and shkolit'; and all the children of the same
node to generate its derivative shkol'nicheskij. We can also query a
hierarchy for the number of nodes it has, giving a measure of its lexical
productivity.  The table compares the Shkola hierarchy and a separate Slovo
'word' hierarchy, where the 'Total' row gives the total number of family
members. Information about the distribution of the members is also
provided. For Shkola over 50% of members occur at level 3, i.e. are
derivatives of derivatives of the root word.
 

         No. of items at given level
Depth    Slovo           Shkola
Level 1  1               1
Level 2  6               4
Level 3  5               9
Level 4  1               1
Total    13              15

The hierarchical functionality of the Node DataBlade provides an elegant
way of capturing morphological relationships between words by encoding
derivational relations as hierarchical relations. To populate such a
database we could use the output of word formation rules, for example the
computable model proposed in Hippisley (2001). Amongst other things
questions about lexical productivity and frequency effects (Shreuder and
Baayen 1997) could be addressed, and etymological information could be
elicited: older words are more likely to have larger derivational families
(Dixon 1982).