A Database of Person Marking in South-Central Trans-Himalayan

1.1 Innovation through high resolution: A microtypological database

The PMST database is designed to facilitate research on the systematic comparison of person markers of different types across related languages, to understand how they function synchronically and change over time.

The overarching goal is to better understand the diversification of person markers through the high resolution afforded by the comparison of inherited and innovative constructions in related languages. We thus employ what can be referred to as a microtypological approach. We achieve high data quality through a broadly collaborative framework as we work with a large number of language experts, including both native speaker linguists and community-external experts. These collaborations enable us to collect data on variation, double-check data from published sources, and achieve a well-informed morphological segmentation of the collected forms. This is all the more crucial considering the nature of the data: SC languages exhibit complex indexation patterns, especially in transitive forms, outlined in 1.2. The database thus also makes a valuable documentary contribution, since comprehensive person marking data are missing for most SC languages in the existing literature.

1.2 Person marking in South-Central languages

The South-Central (a.k.a. “Kuki-Chin”) branch of the Trans-Himalayan (a.k.a. Sino-Tibetan) language family comprises around 50 languages, which are relatively closely related but exhibit great diversity in their person marking systems. Person is typically marked by free pronouns and verbal indexes for S (sole argument of an intransitive verb), A (actor argument of a transitive verb), and P (undergoer argument of a transitive verb). There are both preverbal markers and postverbal markers. The postverbal markers, which are inherited from Proto-Trans-Himalayan (DeLancey, 2010), are less widespread than the preverbal markers. In languages with both sets, they are often in complementary distribution: the postverbal set may be restricted to negative constructions, subordinated or intransitive clauses, or an informal register (DeLancey, 2023; Henderson, 1965); cf. (1) from Ranglong where 1sg is expressed by a preverbal marker in the affirmative and by a postverbal marker in the negative. In transitive paradigms, the two sets co-occur in some languages, indexing A and P arguments, respectively.

The systems also differ with respect to the presence or absence of markers for specific persons and person configurations. Indexation of speech-act participant (first and second person; SAP) P arguments draws on a great variety of constructions with various innovative markers. Consider Anal Naga in (2-a), where the configuration of 2sg A and 1sg P is indexed through a first person P prefix and a second person suffix not specified for syntactic role. Compare Monsang in (2-b), where the same constellation of arguments receives inverse and second person A marking; and Lamkang in (3), where the inverse variably combines with either second person A or first person P.

The South-Central languages thus provide an excellent testing ground for studying the dynamics of person marking as they showcase diverse indexation patterns while typically employing cognate or transparently innovated forms.

2.1 Current sample

South-Central Trans-Himalayan is comprised of approximately 50 languages spoken throughout the higher-elevation regions of the Indo-Bengali-Burmese borderlands. Our sample includes eight South-Central varieties, listed in Table 1. All but Hyow and Pangkhua belong to the tentative Northwestern subgroup (called “Old Kuki” in the older literature, e.g. Grierson (1904); Shafer (1952)), while Hyow and Pangkhua belong to the Peripheral (> Southern > Southeastern) and Central (> Core Central) branches respectively. All three major subgroups of Peterson’s classification (2017) are thus represented by at least one language in the sample. The geographical location of these languages is presented in Figure 1.

Figure 1

Different kinds of sources were available for different varieties, and coverage of person marking is not comparable across existing materials (grammars, sketches, topical articles, etc.). The majority of SC varieties are best characterized as highly understudied. This is beginning to change, but it remains true of Northwestern: until recently, little descriptive, let alone comparative, work on these languages existed in published form. The two non-Northwestern languages, Pangkhua and Hyow, have only recently come to be comparatively well described, with a grammar of the former in print (Akter, 2024) and one of the latter forthcoming (Zakaria, in press). This diversity of sources is illustrated by Ranglong and Monsang, for which the authors (Brown and Konnerth, respectively) drew on their own fieldwork and had forms checked by native speaker collaborators. For Lamkang, the forms were extracted from published sources in consultation with language experts (Chelliah et al., 2019), while for Hmar, unpublished materials provided by the main researchers on that variety constituted the primary source. When we encountered unclear or potentially missing forms or other apparent errors in the data, native speaker collaborators were consulted directly. Consultations were conducted by the authors and two student assistants via e-mail, WhatsApp, and video calls. Collaboration with language experts is thus a crucial element of our work, enhancing the reliability and consistency of the data (cf. Table 1).

The current sample includes only varieties for which data collection was completed at the time of writing, and so constitutes a convenience sample. We are currently in the process of collecting data from an additional 18 languages from different subgroups, such that future releases of the database will cover a larger and more balanced sample of SC varieties.

2.2 Data collection

We had two goals in mind when delimiting which forms to collect for the database: first, to provide a comprehensive representation of person marking across paradigms for each language; and second, to collect the same types of forms in order to ensure inter-language comparability. To this end, we collected pronouns and intransitive and transitive verb forms with all known person distinctions for each language (cf. 1.2). As a useful compromise between comprehensiveness and comparability, we collected verb forms for four tense-polarity configurations (non-future/future × affirmative/negative). We collected additional subparadigms within these parameters only if the respective forms exhibit differences in terms of person marking, such as the generic non-future forms in Anal Naga. We tried to capture all subparadigms for which this is the case. We do not include reflexive scenarios of transitive verbs, as these are expressed with separate, intransitive constructions. Since the focus of PMST is on person forms, we did not attempt to systematically capture lexeme- and verb stem-based alternations, or inflectional classes that do not affect person markers.

We did include morphophonological variation that directly affects the forms of person markers, such as prefixes with copy vowels or homorganic nasal assimilation, and we included different verb stems in the source forms to capture these processes. We list the lexical verb stem only in the source form and represent it with a ‘Σ’ in the phonological and orthographic representations; but this abstraction notwithstanding, the forms we collected were full verbal forms as given in the sources or provided by collaborators.

2.3 Annotation and standardization of the data

In order to make the collected data comparable and reusable, we standardized and annotated forms based on the guidelines developed for the PMST database. These are explained in more detail in a manual for data curators, which can be found in the supplementary materials (at PMST-Meta/manual_curators.pdf, cf. section 5).

3.1 Dataset structure

In the following, we provide an overview of the implementation of the Paralex standard and present the organization of the datasets. The outline follows the Paralex documentation⁷ and focuses on implementations and details that go beyond or deviate from what is described and exemplified in the standard. Figure 3 shows the currently existing modules and their relationships within each dataset.

Figure 3

To facilitate comparative analyses across PMST sets, each file in each data set contains an additional column with the language identifier (language_id). This ensures that each row can be associated with the correct language when combining files. An R script for combining all the modules into one dataset is provided in the supplementary materials (cf. section 5). Each dataset contains the following modules:

Documentation: Each dataset is accompanied by a data sheet outlining how and by whom the data was gathered, annotated, and analyzed. It is meant as a guide for the dataset and an aid in the interpretation of the data. More general information can be found in the README file.

Metadata: The metadata is provided in json according to the Paralex standard and contains key specifications of the dataset and the files contained in it.

Forms: The forms file is the focal point of each dataset. Each row documents a single inflected form and includes: a unique identifier (form_id), the scenario (cell), the verb stem or a placeholder (lexeme), the source form exactly as found in the source materials (source_form), the form in orthography (orth_form), the form in IPA (phon_form), the segmented orthographic form (analysed_orth_form), the segmented IPA form (analysed_phon_form), various columns ending in _tag coding variation, and the source as a cite key (source) with page numbers (page) where applicable.

Morphs: This file (which is not part of the standard) lists all morphs that appear in the forms file in IPA (morph_id). For each morph, there is a list of cells (in_cells) and forms (in_forms) the morph appears in.

Tags: The tags file provides an overview of the tags used to index variants in the dataset. This file is derived from the tag columns in the forms file and can be linked to it via the tag identifiers. For each tag, it contains the following information: a unique identifier for each tag (tag_id), the type or category of the tag (tag_column_name), and a brief explanation of what the tag means (comment).

Cells: The cells file lists scenarios that appear in the forms file. It is derived from the cell column of the forms file and can be linked via that same column. It contains a unique identifier for each scenario (cell_id), the part of speech the form belongs to (POS), and a gloss for each scenario (label).

Features: The feature file presents the feature-value combinations found in the dataset. It is derived from the cells file and contains: a unique identifier for each value (value_id) and a label (label), the feature or category it belongs to (feature), and the value as it appears in the UniMorph schema (Sylak-Glassman, 2016).

Lexemes: The lexemes file lists the verb stems present in the source forms and their meanings. It contains: a unique identifier for each lexeme (lexeme_id) with which it can be linked to the forms file, the part of speech (POS), a list of all occurring forms of the lexeme (label), and the meaning (meaning).

Graphemes: The grapheme file provides a list of graphemes appearing in the source materials (grapheme_id). It is used together with the sounds file for generating the phon columns in the forms file.

Sounds: The sounds file contains each phoneme that appears in the forms file with phonological features. It is thus derivable from the phon columns in the forms file. For each sound, the following is given: a unique id for the sound corresponding to the IPA representation used in the forms file (sound_id), the corresponding IPA representation of the sound in CLTS and PHOIBLE (sound_clts, sound_phoible), the sound id as specified in CLTS (List et al., 2024) and PHOIBLE (Moran & McCloy, 2019), followed by the features as specified in PHOIBLE.

Sources: A BibTeX file contains the references of sources we consulted.

4.1 Distributional profiles of morphs per language

The PMST sets provide a starting point for an empirically based description of person forms, especially in the absence of annotated corpora for SC languages. We demonstrate this here by illustrating the distribution of morphs in one language using a distributional profile, a visualization of the overall distribution of morphs derived from the dataset. It summarizes patterns into a format that allows for cross-checking of database entries and provides a point of departure for comparison of morphs and their distributions within and across languages, as well as for bottom-up morphemic analysis. Figure 4 shows a distributional plot for the Ranglong dataset, which should be read top to bottom for each morph. Each bar represents a morph appearing either before or after the verb stem, or, in the case of pronouns, appearing as a free form. Each panel shows the distribution of the morph across grammatical categories such as tense-aspect and polarity, person, and number.

Figure 4

We illustrate how the distributional profile can be used to pinpoint the properties of individual morphs by comparing three of them, keej-, -me, and -u, across the three panels. From the top panel, we can infer that keej- and -me are restricted to a single paradigm each (future-affirmative and nonfuture-negative, respectively), while -u occurs across all four tense-polarity configurations. From the middle panel, we see that keej- and -me occur only where S/A is 1st person and P is either absent or 2nd/3rd person, meaning both index 1.S/A. -u occurs in every scenario except 1st person intransitive, and so must be a non-person, non-tense-polarity marker. Finally, from the bottom panel, we see that keej- occurs only where the S/A is singular, while -me occurs only where the S/A is plural. -u occurs in all scenarios where at least one argument is plural. Putting it all together, we can state that keej- marks 1sg.sa.fut.aff, -me marks 1pl.sa.nfut.neg, and -u is a plural marker that can pluralize any argument but a 1st person S/A in any tense-polarity configuration. We are thus able to arrive at three important facts about Ranglong verbs: 1) argument role and tense-polarity configuration are cumulatively expressed via person indexes; 2) either or both arguments of verbs with additive plural marking may be semantically plural, and the distinction is context-dependent; and 3) there is an asymmetry in number marking such that 1st person S/As are pluralized using portmanteau indexes, whereas all other arguments are pluralized by a number-neutral index plus a general plural marker.

This information can also be used to identify allomorphs and propose a morphemic analysis. For example, the postverbal morphs iŋ and uŋ are segmentally similar and appear in the same contexts. It is thus very likely that they are allomorphs of one morpheme. Of course, the plot is not meant to replace careful descriptive work, but to complement it and provide a data-driven way for fine-tuning the analysis. Perhaps more importantly, it facilitates comparison of cognate morphemes, whose functions and distributions often vary widely between languages.

4.2 Comparing the length of forms across scenarios

Studies on person marking patterns often assume a fundamental split between SAPs and 3^rd person forms (cf. Benveniste (1946); Siewierska (2010)), which is expected to be amplified in languages characterized as exhibiting ‘hierarchical’ indexation patterns (like SC languages) where SAPs are systematically favored morphologically over 3^rd person forms. There are many ways in which this split could manifest, and so it is not immediately clear how to test such claims. One way of operationalizing this is to compare the length of SAP vs. third person forms measured in number of phonemes, the assumption being that SAP forms would be systematically longer than third person forms. This is based on current proposals regarding person forms in Proto-South-Central, where third person forms are reconstructed as shorter or zero (cf. DeLancey (2015)).

Figure 5 summarizes length measured as number of phonemes across scenarios in the current sample languages in transitive affirmative scenarios. Despite being generally described as ‘hierarchical’, the patterns in individual languages differ markedly. For example, forms for local scenarios (SAP→SAP) are not consistently and substantially longer than all other forms, i.e. mixed (SAP→3 and 3→SAP) and non-local (3→3). Although we observe a decline in length from local to mixed to non-local for most languages, there are various individual patterns, and the differences in overall length are smaller than expected. Only Ranglong and Pangkhua exhibit the expected cline. Even within local scenarios, various patterns emerge: In Monsang, Hmar, and Hyow 1→2 forms are longer, while in Chiru and Pangkhua 2→1 forms are longer. In other cases, the forms show no substantive length differences (cf. Ranglong, Anal Naga, and Lamkang). The inclusive forms,⁸ which structurally represent a separate person category in SC rather than a type of first person plural (see also (Ozerov, 2019, 28)), largely pattern with other forms of the same scenario type. An exception to this tendency is found in Lamkang, where the 3→incl form is the longest, and in Hyow, where 3→incl patterns lengthwise with 3→3.

Figure 5

This comparison shows that a simple SAP vs. 3 split is not evident in most languages when considering the length of forms. This invites further, more detailed research on the topic, also because even languages that are assumed to be closely related (such as Anal Naga and Lamkang, cf. (Ozerov, 2019, 26)) exhibit major differences in their patterning.