Identity Resolution from Names and Lineage: Normalization and Record Linkage

Abstract

Repeated names and their many spellings make distinguishing between individuals a fundamental challenge in genealogical systems. This paper reviews the record-linkage literature, from the statistical foundation laid by Fellegi and Sunter [1], through phonetic matching and string-similarity metrics [2][3], to the particular challenges of Arabic names [4], and on to modern entity resolution [6]. We then describe, at a high level, the Nasayeb platform's methodology for building an "identity fingerprint" without disclosing its internals.

1. The Problem

In any large family, one question recurs: is this "Muhammad ibn Ahmad" the same person as the one in another branch, or a different person who happens to share the name? The wrong answer means either two records for one person (false inflation and contradiction), or the merging of two different people (a loss of the truth). The difficulty is compounded by two properties of names: the many ways they can be spelled, and how heavily they repeat within a single family.

2. The Name Challenge

A single name may be written in multiple forms depending on the spelling of hamzas, the alif, the taa marbuta and others (Fuʾad/Fuwad, Aḥmad/Ahmad). Names also repeat frequently in honour of ancestors, so within one branch you find several people with the same name and the same father's name. The name alone, then, or even the name together with the father's name, is not enough for reliable disambiguation. This is a well-known problem in linking historical and demographic records generally [5].

3. Literature Review

In 1969, Fellegi and Sunter established the classical probabilistic framework for record linkage, based on weighing the probabilities of a match versus a non-match across multiple fields [1]. Before and after that, phonetic matching evolved: the Soundex algorithm and then NYSIIS grouped names that sound alike despite differences in spelling [2]. String-similarity metrics such as Levenshtein and Jaro–Winkler added a quantitative measure of how close two text strings are [3].

Arabic names, for their part, pose particular challenges in normalization and transliteration, addressed by natural-language-processing research with techniques for unifying spelling variants and handling prefixes and affixes [4]. More broadly, recent surveys have expanded on "entity resolution," which generalizes the problem to linking records referring to the same entity across multiple sources [6], with comprehensive practical frameworks in the data-matching literature [7].

A name points to many; an identity points to one.

4. From Name to Structure

The recurring lesson in this literature is that a name on its own is a weak signal; accuracy comes from combining multiple signals, and from linking an entity to its surrounding structure rather than treating it in isolation. In the context of lineage, the surrounding structure is the kinship network itself: descent, marriage, and position within the chart. This ties the problem of identity closely to the problem of representing kinship.

5. The Nasayeb Methodology (Advanced Level)

The Nasayeb platform builds an "identity fingerprint" that combines the name, the chain of descent, and additional signals such as a spouse's name, after a linguistic normalization that unifies spelling variants. The convergence of these signals makes an accidental match between two different people rare, so duplicate records are detected and unified with high accuracy. The precise parameters, matching weights and thresholds are part of our proprietary work, which we do not disclose.

The contribution here is twofold: adapting the matching literature to the specifics of names and lineage across multiple cultures, and tying identity to the kinship structure itself instead of processing names in isolation. The result is a clean family record in which each person is kept once, one you can trust and rely on.

6. Conclusion

Distinguishing between individuals in families with high onomastic similarity is an old problem with a rich literature, from probabilistic matching to entity resolution. The practical solution is not a single signal, but an identity fingerprint that combines name, lineage and surrounding context, built on a linguistic normalization that respects the particulars of the language. In this way the family remains an accurate record, not a heap of contradictory copies.

References

1. Fellegi, I. P., & Sunter, A. B. "A Theory for Record Linkage." Journal of the American Statistical Association, 1969.
2. Russell, R. C. / Odell, M. Soundex (US Patents, 1918/1922); Taft, R. NYSIIS, 1970.
3. Winkler, W. E. "String Comparator Metrics and Enhanced Decision Rules in the Fellegi–Sunter Model." US Census Bureau, 1990.
4. Marton, Y., & Zitouni, I. "Transliteration Normalization for Information Extraction and Machine Translation." Journal of King Saud University — Computer and Information Sciences, 2014.
5. Newcombe, H. B., et al. "Automatic Linkage of Vital Records." Science, 1959.
6. Getoor, L., & Machanavajjhala, A. "Entity Resolution: Theory, Practice & Open Challenges." Proc. VLDB, 2012.
7. Christen, P. Data Matching: Concepts and Techniques for Record Linkage. Springer, 2012.