Semantic labeling with fulfillers in multilingual contexts

1 Introduction

Semantic labeling is the process of assigning structured meaning to linguistic units—it remains central to natural language understanding, especially in multilingual and typologically informed contexts. Established frameworks such as PropBank (Palmer et al., 2005), FrameNet (Baker et al., 1998), and Universal Conceptual Cognitive Annotation (UCCA; Abend and Rappoport, 2013) primarily define semantic roles using explicit lexical and syntactic cues. However, in many languages, meaning is frequently constructed from not just what is said but also what is left unsaid. In pro-drop languages (e.g., Hindi, Bangla, Persian, Polish, Czech, Portuguese) and topic-drop languages (e.g., Malayalam, Japanese, Korean), arguments such as the subject or topic are frequently omitted, yet their referents remain cognitively and communicatively active (David, 2015; Neeleman and Szendrői, 2007; Gordishevsky et al., 2025). For instance, consider the Malayalam sentence in (1) “kadayil poyi” (“[....] went to the store”), which lacks an overt subject. Native speakers infer the actor or subject argument based on the discourse context or the situational context. Similarly, in the Hindi sentence in (2) “darzaza khola gaya” (“The door was opened”), the agent is syntactically omitted but semantically recoverable based on verbal morphology (gender agreement). The stakes for this operationalization are not merely terminological. Section 4.4 shows concretely where a system relying on the argument structure alone diverges from a human reader's interpretation and how that divergence tracks a real distinction in the typological literature between grammatically licensed and culturally licensed omission. At the same time, there are expressions, particularly idiomatic expressions, which figurative meaning cannot be directly derived from their literal components. For example, the English phrase “kick the bucket” figuratively means “to die,” whereas its literal interpretation refers to the act of kicking a bucket. In such cases, the intended meaning arises not from the literal meaning of the words but from historical usage and cultural convention in English. To capture this often-overlooked layer of meaning, we incorporate the concept of fulfillers into semantic labeling. Originally proposed in earlier work (Pala et al., 2025) in the domain of cognitive semantics, fulfillers refer to absent elements that are pragmatically or culturally inferred and semantically consequential. This article extends that earlier proposal by developing a layered annotation framework and applying it to a multilingual corpus, thereby moving the fulfiller concept from a theoretical construct to an operational tool for cross-linguistic semantic analysis. Traditional annotation schemes typically ignore such implicit actors; a fulfiller-aware model annotates these inferred roles, enhancing the semantic representation's granularity. What this granularity is actually good for becomes clear once the categories are tested in the real sentences. Section 2 spells out what a standard one does not allow, and Section 4.4 connects the resulting pattern.

• (1) kaa-y-il po-y-i
• shop-EPENTH-LOC go-PST-3SG
• ‘(S/he) went to the shop.'

• (2) darwaaza khol-a gaya
• door.M.SG open-PFV.M.SG go.PFV.M.SG
• ‘The door was opened.'

1.1 Fulfiller vs. related constructs

To sharpen the distinction between fulfillers and neighboring concepts, we provide a detailed comparison. Unlike zero anaphora, null arguments, or FrameNet null instantiations, fulfillers are not restricted to syntactically licensed omissions; they further encompass inferences grounded in cultural scripts, affective states, and shared experiential knowledge. This broader scope is crucial for capturing meaning in discourse-oriented and culturally rich languages.

Table 1 distinguishes fulfillers from syntactically or lexically motivated constructs found in existing semantic frameworks.

Table 1. Comparison of fulfillers with related constructs.

1.2 Contributions

This article makes the following contributions:

• Introduces fulfillers as a typologically grounded concept for modeling implicit meaning in semantic annotation.
• Proposes a layered annotation framework (grammatical, discourse-pragmatic, cultural-conceptual) applied to a 194-sentence multilingual corpus.
• Integrates fulfillers into the geometric model of semantic weight distribution for both literal and figurative expressions.
• Demonstrates cross-linguistic relevance through detailed analyses of idioms and ellipsis in English, Hindi, Malayalam, Bangla, and Bagri (a western Indo-Aryan language spoken in Rajasthan, Haryana, and Punjab states of India), as well as through a set of everyday non-idiomatic sentences.

2 The fulfiller concept and its implications for semantic role labeling

The concept of the fulfiller represents an ongoing theoretical development aimed at expanding the explanatory power of semantic theory, particularly in the domain of semantic role labeling (SRL). Traditional frameworks such as PropBank and the Penn Treebank corpus primarily focus on overt arguments and explicitly stated semantic roles (e.g., Agent, Theme, and Experiencer). In contrast, the fulfiller concept emphasizes the implied or phenomenologically reconstructed experiential qualities that underlie comprehension. For example, consider the Malayalam expression in (1) “kaa-y-il po-y-i” (“[someone] went to the store”). Classical SRL systems mark an implicit subject and a goal/location argument but do not capture the rich, embodied experiential layer, such as culturally expected motivations or affective states that listeners often automatically reconstruct, as summarized in Table 2 for the Malayalam example above.. Idioms and culturally loaded expressions (e.g., “kick the bucket,” “spill the beans”) also present challenges. Existing frameworks often treat them as fixed expressions without a deeper experiential modeling. The fulfiller framework proposes to formalize how cultural, pragmatic, and phenomenological inferences complete the meaning space, even in the absence of explicit linguistic elements.

Table 2. Malayalam “kaa-y-il po-y-i” ([someone] went to the store)—classical semantic role labeling (SRL) vs. fulfiller framework.

As the table shows, a classical annotation leaves the agent slot empty, even though a listener never actually experiences that gap; a habitual or expected agent, along with the reason for this, is supplied automatically. Annotating this distinction systematically, rather than leaving it to prose description, is what makes the categories usable beyond a single example, and tagging the fulfillers consistently across the corpus makes it possible to measure how much of a language's meaning-making relies on cultural vs. grammatical vs. lexical information to compare the reliance across languages and helps test whether a fulfiller-aware model can flag expressions that a standard SRL-trained system would leave underspecified. Section 4.3 works through this same example in detail, and the Tables 3–5 in Section 4.4 extend this comparison across the pro-drop and topic-drop examples.

Table 3. Malayalam example: “kadayil poyi” ([someone] went to the store)—classical semantic role labeling vs. fulfiller framework.

Table 4. Bagri example: “Byaa ma gai” ([someone] went to the marriage)—classical semantic role labeling vs. fulfiller framework.

Table 5. Hindi example: “darwaja khol gaya”—classical semantic role labeling vs. fulfiller framework.

Once the fulfillers are tagged rather than left as an undifferentiated gap, the annotation supports several concrete uses in research practice. Here, someone can ask, for any single sentence, what is filling the gap—grammar, discourse, or culture—and check it against the data rather than relying on intuition. The discourse-recovery systems already try to do this, so they can be used to train or evaluate those systems directly. The cultural and experiential tags point to a different set of tasks; the idiom detection and non-literal translation need a knowledge base rather than a better parser. Because the tags, here, stay consistent across languages, they also support direct comparison; in our data, pro-drop and topic-drop constructions recover their missing arguments almost entirely through discourse-pragmatic fulfillers. In contrast, idiomatic expressions in the same languages rely instead on cultural and experiential ones, and Section 4.4 comes back and connects to broader work on language typology. This distinction's cost is that the two different problems get treated as one. An argument that verb morphology already recovers is a parsing problem. An argument that only a shared cultural script recovers is a knowledge problem, and no amount of better parsing fixes the second kind, and Section 4.5 works through a real instance of this from our own corpus, where a fulfiller-blind system would not even register that anything had gone wrong.

2.1 Relation to PropBank and Penn Treebank

PropBank focuses on verb-argument structures with explicit argument labels, and Penn Treebank semantic role annotations are similarly tied to explicit syntactic constituents. Both frameworks have limited support for implied experiential or cultural inferences. The fulfiller proposes an additional annotation layer that marks implied experiential structures, including affective and cultural background inferences it; also includes cognitive intentionality, such as inferred goals or affective states, and treats idioms and figurative expressions as sites of experiential collapse rather than purely lexical units. This is not a purely theoretical gap. The downstream tasks that rely on PropBank- or Penn Treebank-style annotation inherit the gap directly: a culturally or discursively recoverable agent that is never marked as an agent, so a downdtream model cannot learn to expect it. Section 4.4 discusses what this looks like in practice.

2.2 Annotation guidelines outline

2.2.1 Objective

To annotate fulfiller elements—that is, experiential, cultural, and pragmatic inferences that are implied but not overtly expressed—as an additional semantic layer beyond standard SRL.

2.2.2 Annotation units

• Sentences or discourse segments
• Tokens or phrases where experiential completion is inferred

2.2.3 Decision tree for annotators

1. Is there an explicit argument missing?

• Yes → Proceed to next step.

• No → No fulfiller annotation needed.

2. Does the interpretation rely on cultural, affective, or pragmatic knowledge?

• Yes → Identify type(s) of inference:

– Cultural inference: social scripts or cultural frames.

– Affective inference: implied emotional or bodily states.

– Experiential inference: imagined perceptual or bodily enactments.

3. Mark the corresponding “fulfiller” categories:

• FULFILLER-CULTURAL

• FULFILLER-AFFECTIVE

• FULFILLER-EXPERIENTIAL

4. Annotate supporting evidence or rationale (e.g., context, cultural references).

2.2.4 Annotation principles

• Focus on the inferences necessary for complete situated understanding.
• Annotate only those fulfillers that substantially enrich or shift interpretation.
• When in doubt, consult cultural or pragmatic references (e.g., idiom dictionaries).

The annotation scheme is very vulnerable to over-elaboration, including plausible but unwanted inferences, and this risk takes a different form depending on the fulfiller's source. For fulfillers tied to conventional scripts—the cultural and affective content behind the idioms and everyday expressions—a fulfiller is only taken if most speakers of the relevant speaking community arrive at the meaning. For example, in “She went to the store,” for instance, details such as the mode of transport or any particular store is left out because it is only one guess among several equal ones, with nothing in the sentence to favor it over the other alternatives. The annotation draws only on the situational or discourse context actually present in the source sentence, not on an invented scenario that merely feels plausible. This annotation scheme is built on capturing the general, default properties of the utterance, not the individual circumstance of a particular speaker or referent. The underlying discipline here is to annotate only what is warranted by convergence across speakers or the context actually given, never what an individual speaker's particular situation is or what simply seems possible to the annotator.

2.2.5 Example corpus snippets with fulfiller annotations

To demonstrate that fulfillers are not restricted to idiomatic expressions, we present both a non-idiomatic everyday sentence and an idiom.

2.2.6 Sentence (everyday expression)

She went to the store.

PropBank/Penn Treebank Roles:

• ARG0: She
• V: went
• ARGM-LOC: to the store

Fulfiller Annotations:

• FULFILLER-EXPERIENTIAL: Imagined bodily movement, navigating aisles.
• FULFILLER-CULTURAL: Script of purchasing or social interaction.
• FULFILLER-AFFECTIVE: Possible emotional state (relief, duty, enjoyment).

2.2.7 Sentence (idiomatic expression)

After he kicked the bucket, they argued about the inheritance.

PropBank Roles:

• ARG0: he
• V: kicked
• ARG1: the bucket

Fulfiller Annotations:

• FULFILLER-CULTURAL: “kick the bucket” as idiomatic death reference.
• FULFILLER-EXPERIENTIAL: Visualizing funeral or death context.
• FULFILLER-AFFECTIVE: Implied grief or conflict-related tension.

2.3 Outlook for the annotation layer

The fulfiller annotation layer provides a systematic approach to capturing culturally and experientially implied meanings beyond explicit arguments. While promising, rigorous empirical validation and comprehensive annotation guidelines are essential for operationalizing this framework and distinguishing it from existing standards.

3 Quality control and agreement framework

Ensuring annotation consistency is crucial for any semantic annotation framework's validity and reproducibility. For the proposed fulfiller framework, traditional inter-annotator agreement (IAA) metrics are valuable but may not fully capture the subjectivity inherent in cultural and experiential inferences. Thus, a hybrid approach is recommended.

3.1 Initial phase: adjudication and guideline refinement

Annotators first work independently on a shared subset of data; disagreements are explicitly discussed in joint adjudication sessions, and guidelines are iteratively refined to address ambiguities and improve clarity.

This stage prioritizes establishing a shared interpretive framework over numerical agreement scores.

3.2 Later phase: interannotator agreement metrics

Once guidelines are stabilized, formal IAA metrics can be introduced:

• Krippendorff's alpha (α): Handles multiple annotators, missing data, and nominal/categorical labels.
• Cohen's kappa (κ): Suitable for pairwise agreement analysis.
• Fleiss' kappa: For multiple annotators labeling the same data.

IAA scores serve as a quantitative measure of consistency and help identify guideline areas that need further clarification. The annotation was carried out by two trained annotators under the supervision of a semanticist. They first independently annotated a 20-sentence subset. Disagreements were resolved through discussion, yielding an agreement rate of approximately 85% on the presence or absence of a fulfiller. Formal IAA metrics (Krippendorff's α, Cohen's κ) were not computed for the full corpus because the primary goal was guideline refinement. A full-scale IAA evaluation is planned for a subsequent phase.

3.3 Alternative and complementary approaches

3.3.1 Probabilistic annotation

Annotators can provide probability distributions over possible fulfiller labels, capturing degrees of certainty or interpretive ambiguity. Divergences can then be measured using cross-entropy or Kullback-Leibler (KL) divergence.

3.3.2 Confidence scores

Annotators explicitly rate their confidence (e.g., on a 1–5 scale) for each label. Low-confidence cases may be flagged for additional review or excluded from agreement calculations.

3.3.3 Task-based evaluation

The practical utility of fulfiller annotations can also be evaluated via improvements in downstream tasks (e.g., interpretability of idioms, semantic role classification performance). This complements agreement-based measures by focusing on functional impact.

3.4 Reporting recommendations

We recommend that future annotation studies using this framework report both numerical agreement scores and qualitative adjudication summaries, describe common sources of disagreement and the strategies used to resolve them, and document confidence distributions where applicable.

3.5 Outlook for quality control

A flexible combination of adjudication, iterative refinement, probabilistic evaluation, and traditional IAA metrics will support a robust and transparent development of the fulfiller annotation framework, accommodating its inherently subjective and context-sensitive nature.

4 Semantic weight distribution: a cross-linguistic analysis

To better understand how meaning is constructed through both explicit and implicit linguistic features, we adopt the semantic weight distribution model proposed by Pala et al. (2025). This model is grounded in a geometric representation of experience and incorporates multiple dimensions of meaning: spatial (θ_*S*), temporal (θ_*T*), experiential facets (θ_*EF*, θ_*ES*, θ_*EM*, θ_*EQ*), cultural salience (θ_*C*), and fulfillers (ϕ), which represent contextually inferred elements (Pala, 2023).

Rather than treating fulfillers as standalone semantic markers, we embed them within a broader framework where they interact with other parameters to create composite meaning structures. This allows us to model both literal and figurative expressions across languages in a structured, typologically aware way.

4.1 Corpus and analysis methodology

The corpus sources are the : English-British National Corpus (Consortium, 2007), supplemented with web-collected idioms; the Hindi—Gyan Nidhi parallel corpus and native speaker examples; and Bangla, Bagri, and Malayalam—examples constructed by native linguist annotators based on naturalistic usage (Table 6). All data comply with ethical guidelines, and the annotated data set will be made publicly available (Table 7).

Table 6. Corpus composition and fulfiller counts.

Table 7. Summary of languages used.

The total number of distinct tags in the tagset is 135 four-letter codes.

Our multilingual corpus spans more than 150 annotated sentences drawn from English, Hindi, Malayalam, Bangla, and Bagri (Consortium, 2007, 2008; Pilo Paul, 1902; David, 2015), and each sentence was annotated using the semantic weight parameters, with particular focus on how meaning is distributed across both overt and covert semantic cues. For figurative and idiomatic constructions (Cook et al., 2008), we included ethnographic annotations to capture culturally shared understandings critical to interpretation. The corpus was asymmetric, having English and Hindi sampled at a larger scale to support the idiom and pro-drop contrast, while Bangla, Bagri, and Malayalam were included as targeted typological contrast points, such as topic-drop/pro-drop varieties, while also planning to develop the corpus further.

4.2 Case studies of idiomatic expression

English—“Kick the bucket”: This idiom figuratively denotes death. Semantic contribution stems from the dynamic interplay between the qualitative traits of kick (suddenness, finality) and bucket (container, ritual object) and the fulfiller (ϕ), which encodes a culturally remembered practice of kicking a holy water bucket at funerals.

Malayalam (3)—“kollam kandavanu illam venda”: Translated roughly as “one who has seen Kollam needs no home” this expression reflects the idea that a transformative experience renders prior needs irrelevant. The phrase draws heavily on cultural salience (θ_*C*), as Kollam is historically understood as a wealthy, vibrant trading port. The fulfiller here encodes that sociocultural background.

• (3) kollam kand-avanu illam venda
• Kollam see.PFV-REL.WHO home need.NEG
• ‘One who has seen Kollam needs no home.'

Bangla (4)—“baro mashe tero parbon”: This expression illustrates cultural rhythm and social tempo. Semantic weight is distributed between temporal density (θ_*T*) and experiential flux (θ_*EF*), with the fulfiller representing shared knowledge of Bengali cultural exuberance.

• (4) baro mashe tero parbon
• twelve month.LOC thirteen festival
• ‘thirteen festivals in twelve months'

Hindi (5)—“Lakshman rekha langhana”: This idiom figuratively denotes the crossing of a moral or social boundary. The semantic weight is distributed between the functional quality and the abstract quality of overstepping, where the fulfiller (ϕ) represents a culturally shaped experience rooted in the mythological history of Ramayana.

• (5) Lakshman-rekha langh-ana
• Lakshman-line cross-INF
• ‘to cross the Lakshman line'

Bagri (6)—“Ganv-ganv gogo, ganv-ganv khejdi”: This idiom figuratively denotes omnipresence and regional identity. The semantic weight here is distributed between the spatial distribution across localities and the cultural function of the deity and nature (θEF). The fulfiller encodes a culturally shared experience between Rajasthani's and links the geographical landscape of the village to a spiritual and environmental reality.

• (6) ganv-ganv gogo ganv-ganv khejdi
• village-REDUP deity village-REDUP khejri.tree
• ‘In every village, the deity; in every village, the khejri tree.'

4.3 Case studies from pro-drop/topic-drop languages

Malayalam phrase (1)—“kada-y-il po-y-i” ([someone] went to the store. The phrase's interpretation arises from the functional properties of the noun “kada” (shop) as a place where buying and selling occur. The locative marker “-il” provides spatial information, indicating that the location of the event is the shop. The verb “po” (go) contributes qualitative properties of motion, introducing the notion of movement toward a location. The suffix “-i” encodes temporal/aspectual information, marking the event as completed in the past. Finally, the phrase's fulfiller is giving information about the actor, who is not overtly expressed in the phrase and is recovered from situational context and the additional information, such as why they went, what they went for, and so on.

Bagri phrase (7)—“Bayaa ma gai”([she] went to the marriage). The interpretation of this phrase arises from the functional properties of its lexical and grammatical elements. The noun “bayaa” here refers to a social event or the location of ceremonial activities, such as weddings. And post-position “ma” functions as a locative marker, providing spatial information and indicating that the event is directed toward marriage location. The verb “gai” is composed of two elements, where “ga” (go/went) indicates the notion of movement toward a destination, and the suffix “-i” indicates the aspectual and agreement information, marking the action completed. The subject is not explicitly expressed here, but it can be inferred through the agreement marker on the verb and the situational context.

• (7) bayaa ma ga-i
• wedding LOC go.PFV-3SG.F
• ‘(She) went to the wedding.'

Hindi phrase (2)—“darwaaja khola gaya” ([someone] opened the door). The interpretation of this phrase is derived from the functional properties of the noun “darwaaja” (door) as a physical object that serves as an entry or exit point between places or spaces. The verb “khol” (open) here contributes the qualitative property of an action that causes a change in the state; the auxiliary “gaya” functions as a part of a compound verb construction, where it adds a completive nuance that indicates the action has been carried out successfully; and the suffix “-ya” in “gaya” encodes temporal and aspectual information. Where the fulfiller provides information about the actor who performs the action but the actor is not overtly expressed in the phrase, it is recovered from the situational context and additional background information about the actor.

4.4 Cross-linguistic findings

Across the examined languages, fulfillers operate not in isolation but in interaction with a variety of semantic weights. Their presence improves the interpretability of idioms and ellipsis-prone sentences in computational models and provides a rich resource for typological generalizations. Importantly, these elements offer systematic patterns of meaning construction that surface-oriented frameworks fail to capture.

A clear pattern emerges: In pro-drop and topic-drop languages, discourse-pragmatic fulfillers (agent/goal recovery based on context) dominate, while in idiomatic expressions, cultural and experiential fulfillers are more prominent. For instance, the Malayalam and Hindi pro-drop examples consistently rely on inferred agents built from situational scripts, whereas idioms like “Ganga Nahana” draw heavily on cultural salience. These differences highlight the need for a flexible annotation scheme that can accommodate both inferential types. These cross-linguistic differences align with typological observations on pro-drop and argument realization (König, 2025), and parallel patterns have been noted in cross-linguistic speech-marker analyses (Santi et al., 2024). The typological literature on pro-drop generally treats referential density, how much of an argument a language leaves for context to supply, as a single property of the language, carried mainly by verb morphology and discourse tracking (Neeleman and Szendrői, 2007; Gordishevsky et al., 2025). Our everyday, non-idiomatic examples fit the picture closely. The Malayalam, Hindi, and Bagri cases in Section 4.3 recover the missing subjects almost entirely through grammar and situational context, with culture playing little or no part at all. The idioms complicate this picture in a way the existing typology does not anticipate. Hindi switches to an entirely different resource once an idiom is involved: For example, “lakshman rekha langhna” is not recovered through the verb agreement or discourse at all but through a shared cultural and historical script unconnected to how Hindi ordinarily handles a dropped subject. A typology built only around the grammatical axis gets Hindi's pro-drop behavior right and has nothing to say about this idiom. Read together as two separable dimensions, one grammatical and one cultural, that happen to coexist within a single language rather than reduce to a single parameter (König, 2025).

This analysis demonstrates that integrating fulfillers into a broader semantic framework allows for nuanced and inclusive modeling of linguistic meaning-especially in discourse-driven and culturally embedded contexts. This split has a direct effect on how much a language relies on the sentence itself vs. the discourse or the cultural surround to complete its meaning, the distinction that underlies the broader referential-density literature on pro-drop and topic-drop systems. Our data here suggest a finer-grained version of that picture: The same language can behave as strongly context-dependent in everyday or non-idiomatic clauses, where the missing material is recovered from the verb morphology or the situational context, while the idiomatic layer draws on entirely different resources, which are shared cultural or historical knowledge that no amount of discourse tracking will supply.

4.5 Discussion

This study highlights the critical, yet often overlooked, role of fulfillers in semantic interpretation, particularly in languages characterized by discourse orientation, pro-drop syntax, or culturally situated expression. Our findings show that fulfillers are not marginal or occasional anomalies but systematic, inferrable elements that carry meaningful semantic load.

Importantly, we do not position fulfillers as an isolated category. Rather, they function as one component in a broader semantic architecture that includes syntactic, pragmatic, and culturally grounded inference mechanisms. Their semantic contribution is relational: Fulfillers acquire meaning through interplay with surface forms, discourse expectations, and shared world knowledge. The language diversity in our corpus, encompassing languages from the Indo-Aryan and Dravidian families, allows us to observe that the triggers for fulfiller inference vary—grammatical licensing in pro-drop environments vs. cultural script activation in idioms, while the fundamental cognitive mechanism of integrating absent information appears to hold in both types of sample, suggesting a universal pattern.

By embedding fulfillers within a geometric model of experience proposed in prior work (Pala et al., 2025), we demonstrate their compatibility with other cognitive-semantic constructs, such as experiential parameters and cultural salience. This layered approach allows us to move beyond surface-level syntactic parsing toward a more inclusive semantic representation model that captures both expressed and inferred elements rather than replacing classical SRL categories.

This distinction also aligns with computational applications, and it differs depending on the type of fulfiller involved. For pro-drop/topic-drop constructions, where the fulfiller is grammar-guided rather than open-ended, this means that the annotated corpus could feed into or evaluate the existing coreference-resolution and argument-recovery systems, using the default agents recorded as a point of comparison. Cultural and idiomatic fulfillers are trickier because the same idiom does not necessarily carry the same fulfiller across different languages, or even across varieties of the same language. For example, the Bagri language idiom “ganv-ganv gogo ganv-ganv khejri,” which translates into English “village-village Gogo, village-village Khejri tree,” means almost nothing to an English speaker. A literal rendering keeps the words but loses the idiom simply because nothing in the literal translation system's training data represents the cultural layer. This is a general problem for low-resource language pairs, where parallel data covering idiomatic usage is less to begin with. A fulfiller-annotated resource would not fix this on its own, but it could work as a flag, marking expressions where a literal translation is likely to lose cultural and experiential content, so a system knows to look for a non-literal equivalent. A fluency check or word-overlap metric would score the literal Bagri-to-English rendering as a success: Every word is present, the grammar is intact, and almost none of the intended meaning survives the trip. The output does not look broken, which is precisely why it would go unflagged in most existing systems. Marking the fulfiller at the annotation stage turns this from a failure that only a bilingual reader happens to catch into one a system can be tested against and, eventually, trained to avoid rather than one it ships without ever registering that anything went wrong.

However, certain limitations must be acknowledged. While our multilingual corpus provides significant insight, it remains regionally focused, with a strong emphasis on Indo-Aryan and Dravidian languages. This restricts the generalizability of our findings. Additionally, our treatment of idiomatic meaning, while enriched by the fulfiller concept, would benefit from a deeper integration of historical and metaphorical factors beyond cultural inference alone.

5 Conclusion

This article introduced fulfillers as a novel dimension in semantic labeling: elements that are absent from the linguistic surface but semantically active due to grammatical, discourse, or cultural inference. By integrating fulfillers into a typologically informed annotation framework and embedding them within a multidimensional semantic model, our analysis identified a class of different meaning components, such as cultural, affective, and experiential inferences, that classical SRL categories do not encode.

Our analysis across English, Hindi, Malayalam, Bangla, and Bagri demonstrates that fulfillers are crucial for resolving semantic gaps, especially in ellipsis-heavy or culturally nuanced constructions. We argue that the inclusion of fulfillers has the potential to improve semantic interpretability in downstream Natural Language Processing (NLP) tasks, such as machine translation, coreference resolution, and question answering, although empirical validation of this claim remains for future work.

More broadly, this work contributes to an evolving understanding of meaning as shaped by not only explicit structure but also what is contextually expected and culturally shared. Fulfillers help bridge this gap by capturing implicit roles that conventional annotation schemes overlook.

5.1 Future work

Future research should test the applicability of fulfillers across a wider typological range, specifically languages: Finnish, because Finnish offers a non-Indo-European, non-pro-drop language with rich case morphology, and Turkish, because it offers an agglutinative pro-drop language with a distinct argument-licensing profile. This broader empirical base will be critical to assessing the universality and variability of inferred semantic roles.

In addition, we suggest deepening the theoretical treatment of idiomatic expressions by incorporating metaphor theory, historical semantic change, and cognitive blending. Doing so will allow for a more holistic understanding of how implicit meaning emerges across languages and cultures.

Ultimately, we argue that acknowledging absence as a structured, inferable part of language opens new pathways in linguistic typology and NLP. Fulfillers, while just one component, help foreground this missing dimension in our models of human meaning-making.

A practical priority is the simplification of the annotation process. The current decision-tree approach, while systematic, remains time-consuming and requires substantial cultural expertise. Streamlining the annotation workflow (e.g., through semiautomatic suggestions or pre-annotated cultural knowledge bases) is an essential step toward integrating fulfillers into large-scale computational pipelines. Pilot experiments could embed fulfiller features into neural SRL models to evaluate their impact on downstream performance.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Meenakshi Sharma Yadav, King Khalid University, Saudi Arabia

Reviewed by: Alexandra Krauska, University of Massachusetts Amherst, United States

Mohamed Taiebine, UEMF, Morocco