Sex assigned at birth is the classification recorded for a newborn, usually as male or female, using observed anatomy and sometimes chromosomes or hormones. The classification involves interpretation and judgment, and biology does not always fit a simple binary. section 5.2.2
Sex is distinct from gender. Gender includes socially constructed roles, behaviors, expressions, and identities; gender identity can differ from assigned sex, extend beyond the binary, and change over time. Pronouns can express and affirm that identity. section 5.2.3
Treat sex labels carefully when using them in data and language systems:
- Recognize that medical data may include categories beyond female and male, such as unknown or something not listed (specify). section 5.2.2
- Avoid treating a name as a definitive sex marker: names can correlate with recorded labels, but a particular name may have multiple labels. section 5.2.2 section 5.2.4
- Separate assigned-sex prediction from gender inference, because systems that connect names, pronouns, and identity can reproduce gender bias and transphobia. section 5.2.3
A probabilistic name signal
A birth-certificate dataset can train a character-based perceptron to predict the recorded label associated with a first name. The result is a probability or confidence signal, not a definitive classification of a person. For example, the sampled records include both female and male entries for Timothy. section 5.2.2 section 5.2.4
>>> np.random.seed(451) >>> URL = 'https://gitlab.com/tangibleai/nlpia2/'\ ... '-/raw/main/src/nlpia2/data/baby-names-us-10k.csv.gz' >>> df = pd.read_csv(URL) #1 >>> df = df.sample(10_000) #2 >>> df.shape (10000, 6)
Inspect a sampled US birth-certificate dataset containing names, recorded sex labels, years, regions, counts, and frequencies before modeling. section 5.2.4
