To what extent do ASL morph types (typical alphabetic morphs and non-typical morphs) conform to the rule of three across sublexical domains (handshape, palm orientation, location, movement, and NMS)?

H₀: The proportion of rule of three violations within each morph type and sublexical domain does not exceed the prespecified 5% tolerance threshold (p ≤ .05).

H₁: The proportion of rule of three violations within at least one morph type and sublexical domain exceeds the prespecified 5% tolerance threshold (p > .05).

To test these hypotheses, the following methodology was employed.

The sampling frame consisted of publicly available ASL videos identified from the reference materials accompanying The Fingerspelling Code:Linguistics of the ASL Alphabet ([21]), together with supplementary ASLized academic discourse from [12]. These sources were selected because they contained naturally occurring examples of lexicalized fingerspelled forms representing the morph categories described by [21]. All videos were screened in their entirety for eligible morph productions prior to coding.

A token was defined as one complete production of a lexicalized fingerspelled form. Each occurrence was treated as an independent observation when it represented a separate production event. Tokens were included only when the entire production was visible and could be reliably coded across the sublexical domains of handshape, movement, palm orientation, location, and NMS. Tokens with incomplete productions or insufficient visual quality were excluded from analysis.

A total of 15 coders participated in the project, including 13 undergraduate students (12 hearing and one Deaf) and two graduate students (one hearing-mute and one late hard-of-hearing). Participants’ ASL proficiency ranged from advanced second-language signers to near-native signers. Before coding began, all coders received training on [21] morph framework, including the rule of three, morph classifications, and operational definitions for each coded variable.

The 15 students were divided into five coding teams and assigned one morph category: unimorph, bimorph, trimorph, quadmorph, or synomorph. The typical morph category was reviewed collectively by the entire coding team because it served as the citation form baseline for comparison. Each group agreed to select specific YouTube links from [21] reference list and [12] ASLized YouTube links using a purposive sampling approach. To facilitate tracking and coding, each video was labeled using the last four characters of its URL. This practice allowed us to resolve issues with coding and data interpretation.

The initial goal was to identify approximately 200 candidate signs for each morph category. This target was established to provide broad representation across the corpus while maximizing the likelihood of obtaining enough eligible morph tokens for subsequent statistical analyses. Because the study relied on naturally occurring ASL productions rather than experimentally elicited stimuli, the number of eligible signs varied across morph categories depending on their availability within the sampled video sources. Consequently, the final number of coded tokens differed across morph types after applying the study’s inclusion criteria.

[21] framework classifies ASL signs according to patterns of sublexical organization. Typical morph represents citation form for fingerspelling without any phonological processes, whereas unimorph, bimorph, trimorph, quadmorph, and synomorph reflect variation in the reduction, preservation, sequential organization, or simultaneous expression of sublexical units. These categories served as the coding framework for the present study. Table 3 summarizes the 15 video sources included in the corpus, along with their creators, publication years, clip codes, token counts, and source links. Additional details regarding the identified morphs and token frequencies extracted from each video source are provided in the Appendix.

Before data collection began, all coders received the same researcher-developed codebook based on [21] rule of three framework. The codebook provided standardized operational definitions for each morph category, sublexical domain (handshape, movement, location, palm orientation, and NMS), and phonological process (assimilation, movement epenthesis, reduction, and elision). Using a common codebook ensures that all coding teams applied consistent definitions and decision criteria throughout the project.

Table 3. YouTube Video sources included in the Corpus.

Note. The dataset consisted of 1579 coded tokens drawn from 15 video sources. Clip codes were used to identify the source video associated with each coded token. Video titles, creators, publication years, and source links are reported to support transparency and replicability of the sampling procedure.

Each coding team reviewed examples from the assigned video sources before beginning formal coding. During the coding process, team members discussed ambiguous productions within their groups and documented questions for discussion during research meetings. When uncertainty remained regarding morph classification or sublexical segmentation, the coding teams consulted with the Deaf professor before finalizing coding decisions. This iterative review process helped maintain consistency in the application of the coding protocol across all teams.

Students first reviewed their assigned morph signs and recorded observational notes regarding the sign’s structural characteristics. Each token was reviewed multiple times at normal speed and reduced playback speed before coding decisions were recorded. Each team documented the corresponding video reference and timestamp for every morph sign identified. Team leaders then presented the target signs using two 75-inch classroom monitors. Video clips were reviewed at both normal and reduced playback speeds to facilitate identification of subtle articulatory features.

A previously developed coding protocol used by the research team in a related study was adapted for the present investigation. The original codebook provided operational definitions and coding procedures for morph classification, sublexical domains, phonological processes, and anatomical and biomechanical features. These coding procedures were retained for the present study, with additional coding criteria incorporated to operationalize and empirically examine [21] rule of three. Coding was conducted across five sublexical domains: handshape, movement, palm orientation, location, and NMS.

Handshape units were counted when they represented distinct, stable articulatory targets within a fingerspelled form. Transitional finger positions occurring only during movement between target handshapes were not coded as independent handshape units. Movement was coded when it represented an independent articulatory target rather than transitional movement occurring solely to facilitate biomechanical movement between adjacent handshapes. Palm orientation was coded according to the stable orientation of the palm associated with each sublexical unit. Changes in orientation were recorded only when they represented distinct articulatory targets rather than continuous rotational movement. Location units were coded when they represented distinct target locations in signing space or on the body. In the absence of evidence supporting multiple independent location targets, continuous movement between locations was treated as a single movement rather than multiple location units. NMS included observable facial expressions, mouth movements, head movements, and other non-manual behaviors occurring simultaneously with fingerspelled production.

In addition to the five sublexical domains, coders recorded the presence or absence of four phonological processes: assimilation, movement epenthesis, reduction, and elision. These processes were identified using operational definitions derived from [21] and were coded independently of morph classification. Anatomical and biomechanical features—including wrist motion, thumb position, finger-joint configuration, and finger placement—were also coded for each token using the standardized coding protocol.

Special consideration was given to the fingerspelled forms J and Z because their continuous path movements can complicate sublexical segmentation. Initial coding followed the operational criteria specified in the codebook. Transitional movements required to produce these forms were not treated as independent sublexical units unless they met the established criteria for a distinct articulatory target. Alternative interpretations of J and Z were examined separately as exploratory analyses and were not used to modify the original coding decisions.

Table 4 presents the English equivalent words identified within each ASL morph category and the number of coded tokens included in each category. Specifically, the analyses evaluated whether morphs contained more than three sublexical units across handshape, movement, palm orientation, location, and NMS domains. The final dataset consisted of 1579 ASL morph tokens distributed across six morph categories. Table 4 summarizes the number of English equivalent words identified within each morph category and the final sample sizes used in the analyses. Consequently, the final sample consisted of unimorph (n = 163), bimorph (n = 200), trimorph (n = 181), quadmorph (n = 197), synomorph (n = 174), and typical morph (n = 664) as shown in Table 4.

Median analysis is a nonparametric approach based on the middle value of ordered observations ([5]). Median values were calculated for each morph type across five sublexical domains: handshapes, movements, palm orientations, locations, and NMS. Structural unit counts were coded on an ordinal scale reflecting the number of observed components within each domain (1 = 0 units, 2 = 1 unit, 3 = 2 units, 4 = 3 units, and 5 = 4 or more units). The data consisted of ordered categorical counts rather than continuous measurements; the median was used to summarize the typical structural complexity associated with each morph type.

Exact binomial tests were conducted to evaluate the extent to which observed [21] rule of three violation rates differed from the study’s established 5% tolerance threshold. The binomial test evaluates whether the observed proportion of one of two possible outcomes differs significantly from a hypothesized probability ([1]). For the present analysis, each production was classified dichotomously as either conforming to the rule of three constraint (three or fewer sublexical units) or violating the constraint (four or more sublexical units). Violations were treated as the event of interest, with the test proportion set at .05. Thus, the analyses evaluated observed violation rates relative to a hypothesized 5% tolerance threshold. Because statistical significance alone does not indicate the direction of the difference, observed violation percentages were examined alongside the exact binomial test results to determine whether violation rates were above or below the 5% threshold.

Table 4. English equivalent words across ASL morph categories.

Note. Values in parentheses represent the frequency of each English equivalent within its corresponding ASL morph category. Morph classifications were based on [21] ASL morph framework.

In the dataset, the quadmorph category contained 197 tokens, including 32 apparent violations of the rule of three constraint in the handshape domain. These violations were primarily associated with the sequences A-T-I-C (n = 15) and L-O-G-Y (n = 17) which involve four sequential handshape configurations. However, closer examination suggested that phonological processes may compress these articulations into fewer functional sublexical units.

The fingerspelled form, PRAGMATIC, was produced in three segments: PRAG-M-ATIC, with the initial sequence PRAG forming a quadmorph, followed by a brief pause and an isolated M handshape, and then another brief pause before the production of the second quadmorph, -ATIC. Within the final sequence A-T-I-C, the segments share overlapping articulatory features (see Table 11). The A-handshape is produced as a closed fist with all fingers and the thumb flexed. During the transition to the T-handshape, the thumb shifts beneath the index finger while the remaining fingers retain their closed configuration, resulting in minimal structural change. The production of I-handshape involves extension of the little finger while the rest of the fingers maintains the prior configuration, creating an assimilated, synomorph-like transition between T- and I-handshapes. Finally, the C-handshape emerges through extension and curvature of the index and middle fingers and thumb, while aspects of the preceding I-handshape are retained.

Table 11. Sublexical checklist for the final sequence ATIC.

Note. This table presents the sublexical feature checklist for the fingerspelled final sequence of A-T-I-C, showing the structural units associated with each letter across the domains of handshape, palm orientation, location, and movement. Each domain contains four structural units, exceeding the three-unit limit proposed by the rule of three constraint and therefore representing an apparent structural violation. Progressive assimilation is observed across adjacent segments in the sequence, indicating phonological influence between neighboring letters. No instances of epenthesis, metathesis, reduction, or elision were identified in this example.

Although this sequence could be interpreted as four distinct handshapes, the articulations shared a common base configuration and exhibited progressive assimilation across transitions. As a result, the apparent four-unit structure reflects transitional modification rather than four independent sublexical units corresponding to each segment (A-T-I-C). Unlike a typical morph sequence, which would be articulated as four discrete alphabetic handshapes (A-T-I-C) separated by identifiable segment boundaries, the final sequence instead exhibits continuous articulatory restructuring and progressive assimilation across adjacent handshapes. Under a strict counting approach, A-T-I-C contains four handshapes and therefore constitutes an apparent violation of the rule of three constraint. However, from an articulatory perspective, the sequence may be more appropriately interpreted as a coordinated lexicalized pattern in which adjacent handshapes share a common articulatory base and are linked through progressive assimilation rather than functioning as four fully independent sublexical targets. The sublexical organization of the ATIC sequence is summarized in Table 11.

A similar pattern was observed in the fingerspelled form, AXIOLOGY, particularly, the final sequence LOGY. Because the video presentation was intended for Deaf audiences who might have been unfamiliar with academic terminology, the Deaf native signer articulated the sequence clearly to maintain lexical transparency. In this production, the O handshape differed slightly from its typical morph form. Specifically, the middle, ring, and little fingers were not fully engaged, producing a reduced configuration that facilitated smoother transitions between L and G.

Both L and G required the index finger and thumb to be extended while the remaining fingers remained flexed, although the spatial relationship between the index finger and thumb differed slightly. In the G handshape, the thumb moved closer to the index finger due to flexion at the index finger’s MCP joint, and the articulation often incorporated wrist rotation to adjust palm orientation. The modification of the O handshape, therefore, appeared to support smoother articulatory transitions between L,G, and Y, illustrating the role of reduction and transitional movement in shaping quadmorph production. The structural organization of the LOGY sequence is summarized in Table 12.

At first glance, the ATIC and LOGY sequences did not appear highly complex in ways that would require extensive phonological process assistance. These patterns reflect coarticulation, in which adjacent handshapes influence one another through continuous, overlapping articulatory transitions. As a result, what appears as four sequential handshape units (A-T-I-C or L-O-G-Y) may function phonologically as a smaller set of perceptual units due to shared articulatory bases and assimilatory movement.

The ATIC sequence appeared to be produced in a manner closer to a typical morph structure because the handshapes shared similar articulatory bases. In contrast, the LOGY sequence differed because the typical morph configuration for O required different articulatory properties than those required for L and G. This structural mismatch likely contributed to the greater visibility of quadmorph features in the LOGY sequence compared to ATIC.

This analysis raises an important question: do the sequences of ATIC, LOGY, OLZ, and Z truly violate the rule of three theory? From a strict interpretation of [21] framework, these cases could be considered violations because they appear to exceed the three-unit structural threshold. However, [21] claim had not previously been empirically tested. The findings, therefore, suggest the need for additional studies examining similar morph sequences to determine whether the observed patterns consistently support the rule of three theory or whether revisions to the framework may be warranted.

Table 12. Sublexical checklist for the final Sequence LOGY.

Note. This table presents the sublexical checklist for the fingerspelled form LOGY, showing the structural units associated with each letter (L-O-G-Y) across the domains of handshape, palm orientation, location, and movement. Each domain contains four structural units, exceeding the three-unit limit proposed by the rule of three constraint and therefore representing an apparent structural violation. Progressive assimilation occurs between adjacent segments, and wrist rotation is observed in the movement domain, particularly in the transition between O-handshape and G-handshape. Reduction is also present in the handshape of O, which differs from the canonical fingerspelled form. No instances of epenthesis, metathesis, or elision were observed in this example.

[21] discussed the historical development of the ASL alphabet using the Yebra Alphabet illusion, noting that the sign for Z was originally articulated with the pinky finger before evolving to its modern form produced with the index finger. Historical records indicate that manual alphabets emerged as early as the seventh century in monastic communities for communication among monks and with ill individuals ([21]). These systems later became associated with Deaf education during the sixteenth and seventeenth centuries and eventually spread across Europe before reaching North America ([15]; [21]). [21] therefore argued that the Z form predates the standardized ASL alphabet. Canonical Z forms challenged the rule of three and suggest that the lexicalized ZZ form emerged as a strategy for reducing structural complexity ([21]). More recently, [18] identified the bent-V realization of ZZ as a conventional representation of the double-Z sequence in contemporary ASL fingerspelling. Together, these observations suggest that Z-related forms may be especially susceptible to restructuring processes that preserve lexical information while reducing articulatory complexity. The concentration of apparent violations in Z-containing tokens observed in the present study is consistent with [21] observation that Z occupies a unique position within the ASL fingerspelling system.

Although this historical account provides useful context, it does not necessarily explain the location of violations observed in the present dataset. [21] acknowledged that the Z example in his work had not yet been systematically examined. The current study analyzed 1579 ASL morph tokens, of which 69 were initially identified as exceeding the rule of three constraints under the study’s strict coding criteria. These results broadly support [21] claim that the rule of three functions as a structural tendency in ASL morph organization. However, violations were concentrated in a small number of lexical items, including tokens containing the Z sign.

Both trimorph and typical morph tokens exhibited location violations associated with Z. This pattern likely reflects the articulatory structure of the sign itself. The Z sign traces the zigzag shape of the printed letter, producing a path of movement that traverses multiple spatial positions ([15]). When analyzed strictly under the rule of three framework, this trajectory can appear to generate four distinct locations while maintaining only three movements.

However, this interpretation may overestimate the number of location units involved. From a phonological standpoint, Z’s “four locations” are not discrete lexical targets but inflection points along a single continuous path movement. In signed language phonology, movement is a primary sublexical, and intermediate spatial points along a trajectory do not count as independent sublexical units. The relevant perceptual unit is the overall path, not each articulatory coordinate the hand passes through. This framing helps explain why Z appears to violate the rule of three even though its structure reflects a single phonological movement rather than multiple location features.

Within the ASL alphabet, J and Z are the only letters that require continuous path movement ([19]; [20]). The movement and spatial position are closely intertwined in these signs; the segmentation of locations may be analytically ambiguous. If the initial, intermediate, and final spatial points are interpreted as part of a single movement trajectory rather than independent locations, the apparent violation may be reduced or eliminated. Under this interpretation, the Z sign may not represent a true violation of the rule of three, but rather a case where movement and location interact in ways that complicate sublexical segmentation (see Table 13). The apparent violations in the Z sign suggest that the rule of three may prioritize perceptual units over articulatory points; while the hand passes through four spatial coordinates, the movement is perceived as a single, continuous phonological event.

Table 13. Sublexical checklist for the final sequence OLZ (Variant B).

Note. In Variant B represents a phonologically modified production of the lexicalized fingerspelled sign OLZ, in which sublexical units are merged or reduced through phonological processes. Values indicate the number of structural units identified within each sublexical domain for each letter (O-L-Z). Progressive and regressive assimilation occur across the sequence ([16]), affecting the distribution of structural units. In the palm orientation domain, the L segment exhibits a directional twist influenced by both the preceding O segment (progressive assimilation) and the following Z segment (regressive assimilation), despite being counted as a single structural unit. The Z segment exhibits a zigzag movement trajectory, contributing to the movement count and multiple location changes. Additionally, reduction is observed in the initial O handshape, which differs from its typical morph form due to assimilation influences from the following segment. Merging refers to the phonological compression of multiple sublexical units into fewer structural realizations as a result of processes such as assimilation and coarticulation. These merging processes are evident in the domains of location and movement, resulting in fewer distinct sublexical units compared with Variant A. Totals reflect the cumulative number of sublexical units across the sequence. This variant demonstrates how phonological processes compress structural complexity and reduce or eliminate instances that would otherwise exceed the rule of three constraints. NMS, epenthesis, metathesis, and elision were not observed in Variant B and are therefore marked as absent.

The sequences of ATIC, LOGY, OLZ, and Z ([4]; [7]) illustrate how phonological processes can produce apparent departures from the rule of three constraints. Under a strict interpretation of [21] framework, these tokens may be classified as violations. However, closer examination suggests that many of these cases arise from assimilation ([2]; [8]; [9]; [16]), reduction, and movement transitions ([3]; [10]; [19]) that compress multiple articulatory elements into a single continuous production ([15]; [18]; [20]). In this sense, the apparent violations may reflect either analytic segmentation of decisions or genuine structural exceptions, both of which require further investigation. [21] proposal had not previously been tested empirically; the present study provides an initial evaluation of the theory. Future research should examine additional lexical items and datasets to determine whether similar patterns emerge across broader samples of ASL morph structures. Such work will help clarify whether the rule of three operates as a strict constraint or as a strong structural tendency within ASL phonology.

Taken together, the apparent violations observed in Z,OLZ,ATIC, and LOGY suggest that departures from the rule of three constraint are not random but are systematically linked to phonological processes operating during articulation ([21]). In each case, processes such as assimilation, reduction, and transitional movement influence how sublexical units are realized in surface production. These processes may compress or redistribute structural features across movement trajectories and handshape transitions, making the segmentation of independent units analytically complex. As a result, some tokens that appear to exceed the rule of three threshold may instead reflect continuous articulatory patterns rather than discrete structural violations. Taken together, these findings support the view that the rule of three functions as a strong organizing constraint within ASL morph structure rather than an absolute restriction, with apparent expectation often arising from phonological processes, articulatory pressures, and lexicalized production patterns.

To explore this rule of three violations and assumptions, the research team has created a research agenda to further these studies. The extended studies of this research include a chi-square analytic approach to further examine the strength of association between phonological processes, spread of fingers, thumb movement, wrist movement, palm orientation, and Hold-Movement segmentation features. Furthermore, another round of research involving this team focuses on a multi logistic regression (MLR) analytical approach to predict the relationship between fingers and wrist movement as biomechanics and [21] morph types.

While this study provides key information for future studies of ASL fingerspelling processes, there are limitations to the current study that must be addressed. The team that completed this study consisted of 13 undergraduate students and therefore is vulnerable to human errors in coding and interpretation. To control this limitation, decisions were made in groups and were checked by a Deaf professor as well as discussed with the team of doctoral students. Although multiple systems were set in place to ensure accuracy, human error must be accounted for as a limitation of this study.

A second limitation to this study addresses the availability of data from [21] references in his book. While the research team utilized links from the references page for their study, some of these links were broken or not accessible when the current study was conducted. The team utilized links that were available and collected data from the same sources as [21] but were limited to the available videos. Due to the accessibility of some of these videos being blocked, there may be data points that were not collected that could impact the study’s findings.