Lexical data collection in language documentation often contains transcription errors and borrowings that can mislead linguistic analysis. We present unsupervised methods to identify phonotactic inconsistencies in wordlists, applying them to a multilingual dataset of Kokborok varieties with Bangla. Using phoneme-level and syllable-level n-gram language models, our approach identifies potential transcription errors and borrowings. We evaluate our methods using hand annotated gold standard and rank the phonotactic outliers using precision and recall at K metric. The ranking approach provides field linguists with a method to flag entries requiring verification, supporting data quality improvement in low-resourced language documentation.

Field linguistics increasingly relies on computational tools to organize, analyze, and preserve linguistic data, yet the classificatory assumptions embedded in these tools are rarely examined. A pervasive assumption is that languages can be treated as discrete, genealogically defined units, with relatedness modeled as tree-structured descent. We argue that this assumption misrepresents linguistic evidence in contact-heavy regions and risks distorting the computational mediation of field linguistic data. Focusing on South Asia, we show that widely assumed boundaries—such as the Indo-Aryan–Dravidian divide—collapse in long-standing contact zones characterized by convergence, dialect continua, and institutional multilingualism. Through historically grounded case studies including Kannada–Telugu and Tamil–Malayalam, we demonstrate how convergence, script-mediated distance, and post-hoc standardization reshape how field data is segmented, compared, and interpreted when organized through genealogical labels. We argue that contact-aware, relational models of linguistic relatedness are necessary if NLP tools are to support, rather than distort, the documentation and analysis of linguistic diversity.

Interlinear glossed text (IGT) creation remains a major bottleneck in linguistic documentation and fieldwork, particularly for low-resource morphologically rich languages. We present a hybrid automatic glossing pipeline that combines neural sequence labeling with large language model (LLM) post-correction, evaluated on Jungar Tuvan, a low-resource Turkic language. Through systematic ablation studies, we show that retrieval-augmented prompting provides substantial gains over random example selection. We further find that morpheme dictionaries paradoxically hurt performance compared to providing no dictionary at all in most cases, and that performance scales approximately logarithmically with the number of few-shot examples. Most significantly, our two-stage pipeline combining a BiLSTM-CRF model with LLM post-correction yields substantial gains for most models, achieving meaningful reductions in annotation workload. Drawing on these findings, we establish concrete design principles for integrating structured prediction models with LLM reasoning in morphologically complex fieldwork contexts. These principles demonstrate that hybrid architectures offer a promising direction for computationally light solutions to automatic linguistic annotation in endangered language documentation.

Automatic speech recognition (ASR) is a crucial tool for linguists aiming to perform a variety of language documentation tasks. However, modern ASR systems rely on data-hungry transformer architectures, rendering them generally unusable for underresourced languages. We fine-tune a wav2vec 2.0 ASR model on Yanyhangu, an Indigenous Australian language, comparing the effects of phonemic and orthographic tokenization strategies on performance. In parallel, we explore ASR’s viability as a tool in a language documentation pipeline. We find that a linguistically informed phonemic tokenization system substantially improves word error rate (WER) and character error rate (CER) compared to a baseline orthographic tokenization scheme. Finally, we show that hand-correcting the output of an ASR model is much faster than hand-transcribing audio from scratch, demonstrating that ASR can provide significant assistance for underresourced language documentation.

We propose a method for efficient field data collection of speech resource data which leverages short-form verbal arts, namely riddles and proverbs, which permit a predictable transcript to be assigned to naturalistic but conventionalized utterances. As a proof of concept, we describe a 5.25 hour corpus of proverbs and riddles collected for Kom, a low-resource language of Cameroon, and conduct ASR modeling experiments on the corpus. Results suggest that the method yields high quality speech data, albeit with relatively low lexical diversity. We highlight the alignment of the collected data with community priorities for cultural education and preservation in the Cameroonian context.

While qualitative descriptions (in the form of reference grammars) and benchmarks for low-resource languages are becoming increasingly widespread, computational linguists do not often use quantitative methods to describe a new lect rather than a new model. This paper intends to close this lacuna. The case study is a Lemko text transcribed at the beginning of the twentieth century. Using morphosyntactic tagging and topic modelling, the study demonstrates areal influences and archaic features of the lect. Fine-grained evaluation significantly assists in identifying subtle patterns that are not readily apparent through traditional metrics such as accuracy score. The results highlight the necessity of a more detailed analysis of model performance, which may yield more linguistically significant results than a purely manual check. This information is present in the resulting dataset, which can be used for further investigation into the structural features of the Lemko lect.

We conduct a preliminary study of the order of subject (S), object (O), and verb (V) in Tatyshly Udmurt (Finno-Ugric) on the basis of approximately 900 clauses from oral folklore and non-folklore narratives (including contemporary texts and texts recorded earlier) using a gradient approach. We show that the most frequent word orders are SOV, SV, and OV. In full clauses (with both S and O), in folklore texts SOV order (≈ 70%) is followed by OSV order (≈ 15%). In contemporary non-folklore texts, however, SOV order competes with SVO order (50% vs 30%), which may be explained by the influence of Russian. We note that full clauses may differ from clauses with only S or with only O: in contemporary folklore texts VS order is much more frequent in S-only clauses (≈ 23%) than in full ones (≈ 4%), and in contemporary non-folklore texts VO order is more frequent in full clauses (≈ 35%) than in O-only ones (≈ 12%). Moreover, we show that word order can depend on the type of clause. For example, in existential clauses the order is almost always SV, while clauses with verbs of speech often have VS order.