Machine learning is a rapidly growing research paradigm. Despite its foundationally inductive mathematical assumptions, machine learning is currently developing alongside traditionally deductive inferential statistics but largely orthogonally to inductive, qualitative, cultural, and intersectional research—to its detriment. I argue that we can better realize the full potential of machine learning by leveraging the epistemological alignment between machine learning and inductive research. I empirically demonstrate this alignment through a word embedding model of first-person narratives of the nineteenth-century U.S. South. Situating social categories in relation to social institutions via an inductive computational analysis, I find that the cultural and economic spheres discursively distinguished by race in these narratives, the domestic sphere distinguished by gender, and Black men were afforded more discursive authority compared to white women. Even in a corpus over-representing abolitionist sentiment, I find white identities were afforded a status via culture not allowed Black identities.

机器学习是一种快速发展的研究范式。尽管具有基本的归纳数学假设，但机器学习目前正在与传统的演绎推理统计一起发展，但在很大程度上与归纳、定性、文化和交叉研究正交——对其不利。我认为，通过利用机器学习和归纳研究之间的认识论一致性，我们可以更好地实现机器学习的全部潜力。我通过 19 世纪美国南部第一人称叙述的词嵌入模型经验地证明了这种对齐。通过归纳计算分析将社会类别与社会制度联系起来，我发现在这些叙述中，文化和经济领域以种族为话语区分，以性别区分的家庭领域，与白人女性相比，黑人男性获得了更多的话语权。即使在一个过度代表废奴主义情绪的语料库中，我发现白人身份通过不允许黑人身份的文化获得了地位。