Relational Chatbot Design Grammar

• RCDG is a framework that formalizes chatbot interaction through grammar-based patterns to support culturally grounded maternal health care.
• It encodes mediated decision-making, recognition of silence, episodic use, and infrastructural fragility as foundational design commitments.
• The framework provides actionable heuristics for adapting chatbots to collective decision-making and resilient care in contexts like peri-urban Lahore.

The Relational Chatbot Design Grammar (RCDG) is a framework introduced by Hameed et al. to reorient the design of maternal health chatbots toward collective, culturally grounded care in low-resource settings. RCDG formalizes chat interaction patterns according to four design commitments that respond to the socio-technical complexities of peri-urban Lahore, Pakistan, where phone access is shared, decision-making is collective, and infrastructure is fragile. Rather than treating mediated consent, episodic usage, silence, and systemic fragility as obstacles, RCDG encodes them as baseline conditions, seeking to align chatbot behavior with the relational realities of its users (Hameed et al., 31 Oct 2025).

1. Motivation and Context for RCDG

In peri-urban Lahore and similar low-resource, patriarchal contexts, the individualistic design assumptions underpinning conventional maternal health chatbots—private device use, direct user inquiry, reliable connectivity, and individualized care—routinely fail. Empirical findings from a WhatsApp-based deployment with 48 pregnant women revealed salient barriers: shared phones impeding privacy, family members mediating consent and authority, intermittent device access, taboo around explicit questioning, and infrastructural unreliability. Conventional chatbot designs, predicated on autonomous and literate engagement, thus yield low sustained usage and reproduce forms of inequity and exclusion. RCDG is motivated by the need to systematically reframe and remediate these assumptions at the level of chatbot behavioral logic, treating collective mediation and infrastructural precarity as foundational (rather than exceptional) use cases (Hameed et al., 31 Oct 2025).

2. Core Commitments of the Relational Chatbot Design Grammar

RCDG consists of four key commitments, each articulated as a high-level design pattern and illustrated via pseudo-BNF grammar fragments. These commitments condition how the chatbot parses, responds, and structures dialogic sequences within the collectivist, fragile care ecologies observed in the empirical study.

2.1 Mediated Decision-Making

Authority and consent in maternal health are subject to ongoing negotiation among kin, clinicians, and the pregnant woman. RCDG requires chatbots to support layered consent, giving the woman agency over information sharing while recognizing the distributed nature of decision-making. Grammar fragments formalize this pattern as:

$$\begin{array}{l} \langle\text{MediatedSegment}\rangle ::= \langle\text{LayeredConsentPrompt}\rangle \;|\; \langle\text{EscalationNotice}\rangle \ \langle\text{LayeredConsentPrompt}\rangle ::= \text{\bf “Would you like to share this advice with [RELATIVE]?''} \ \langle\text{EscalationNotice}\rangle ::= \text{\bf “Your doctor will see: ‘}\langle\text{UserUtterance}\rangle\textbf{’. Proceed?”} \end{array}$$

2.2 Recognition of Silence as Engagement

In settings where vocalizing concerns is fraught or discouraged, RCDG encodes silence, truncated input, or hesitation as engagement, not disengagement. The grammar includes:

$$\begin{array}{l} \langle\text{SilenceSegment}\rangle ::= \langle\text{NarrationReframe}\rangle \;|\; \langle\text{MinimalReplyInvite}\rangle \ \langle\text{NarrationReframe}\rangle ::= \text{\bf “I hear that you’re describing ‘}\langle\text{UserNarration}\rangle\textbf{.’ Would you like tips or a summary?”} \ \langle\text{MinimalReplyInvite}\rangle ::= \text{\bf “It’s okay to be brief. Tap 1 if you’d like check-in reminders.”} \end{array}$$

2.3 Support for Episodic Use

Given that women may access the chatbot sporadically—such as during rare phone access or clinic visits—chatbot statefulness is essential. RCDG patterns preserve conversational history and treat return interactions as resumptions, not restarts:

$$\begin{array}{l} \langle\text{EpisodicSegment}\rangle ::= \langle\text{ReentryGreeting}\rangle \;|\; \langle\text{Moment\_of\_NeedAdvice}\rangle \ \langle\text{ReentryGreeting}\rangle ::= \text{\bf “Welcome back! Last time you mentioned ‘}\langle\text{LastConcern}\rangle\textbf{.’ Where shall we pick up?”} \ \langle\text{Moment\_of\_NeedAdvice}\rangle ::= \text{\bf “Need quick reassurance on [TOPIC]? Reply with that word, and I’ll send you essentials.”} \end{array}$$

2.4 Treating Fragility as Baseline

RCDG encodes infrastructural breakdowns—e.g., device sharing, connectivity loss, disasters—as the norm. The grammar introduces multi-channel fallback and proactive checks:

$$\begin{array}{l} \langle\text{FragileSegment}\rangle ::= \langle\text{ChannelFallback}\rangle \;|\; \langle\text{DisruptionCheckIn}\rangle \ \langle\text{ChannelFallback}\rangle ::= \text{\bf “It looks like WhatsApp failed. May I send an SMS?”} \ \langle\text{DisruptionCheckIn}\rangle ::= \text{\bf “I heard there’s flooding. Are you safe? Reply SAFE or HELP.”} \end{array}$$

3. Grammar Structure and Mapping to Interaction Patterns

RCDG is structured as a two-level design grammar. The upper level comprises the four commitment nonterminals:

• $\langle\text{MediatedSegment}\rangle$
• $\langle\text{SilenceSegment}\rangle$
• $\langle\text{EpisodicSegment}\rangle$
• $\langle\text{FragileSegment}\rangle$

Under each, a set of design patterns or production rules—such as LayeredConsentPrompt, NarrationReframe, ChannelFallback—instantiate the high-level commitment for a given context or user input. The terminal symbols correspond to actual surface forms in chatbot dialogues: the specific prompts, questions, fallback notifications, or escalation messages. No holistic BNF or production rule enumeration is provided, but the illustrative fragments clarify the logic for each category (Hameed et al., 31 Oct 2025).

4. Illustrative Dialogue Snippets

The implementation of RCDG commitments in chatbot dialogue is illustrated by a set of annotated interaction snippets:

Scenario	User/Chatbot Exchange	RCDG Segment
Mediated Decision-Making	User: “I’ve been feeling dizzy.”<br>Chatbot: “I heard ‘feeling dizzy.’ Would you like to share this note with your mother-in-law before I suggest next steps?”	〈MediatedSegment〉 → 〈LayeredConsentPrompt〉
Recognition of Silence	User: [no reply for 30 s]<br>Chatbot: “I notice you haven’t typed anything. That’s okay. If you’d like a gentle breathing tip, tap 1; if you’d just prefer a moment of quiet, tap 2.”	〈SilenceSegment〉 → 〈MinimalReplyInvite〉
Episodic Use	[One month later]<br>Chatbot: “Welcome back! Last time we talked about iron supplements. Would you like a quick update or is there something new on your mind?”	〈EpisodicSegment〉 → 〈ReentryGreeting〉
Fragility as Baseline	[WhatsApp fails]<br>Chatbot: “It seems my last message didn’t go through on WhatsApp. May I send it via SMS instead?”	〈FragileSegment〉 → 〈ChannelFallback〉

These mappings demonstrate the grammar’s grounding in actual conversational routines, where chatbot “turns” are governed by recognition of social, temporal, and infrastructural context.

5. Empirical Evidence and Evaluation

The commitments of RCDG were surfaced through the deployment and observation of a maternal health chatbot among 48 pregnant women in Lahore. Out of these, only 9 engaged with the chatbot even once, and only 5 used it more than once. The observed engagement patterns—mediated access, persistent silence, episodic contact, and frequent infrastructural breakdown—substantiate RCDG’s premise that these are not aberrations but regularities determined by social context and material conditions. The study reframes low engagement, silence, and non-use not as failures, but as culturally and infrastructurally produced patterns of care and communication.

Notably, no RCDG-informed chatbot prototype has yet been built or subjected to quantitative outcome evaluation. The framework is thus proposed as a set of design heuristics rather than as an empirically validated intervention. No new engagement metrics, retention statistics, or clinical outcomes are reported for RCDG-specific implementations. A plausible implication is that further work will be required to measure any improvements in trust, equity, or health outcomes produced by adherence to RCDG (Hameed et al., 31 Oct 2025).

6. Significance and Prospects for Application

RCDG represents a conceptual advance in sociotechnical chatbot design, shifting emphasis from individualism and information delivery to collectively mediated, context-sensitive dialogue routines. Its grammar-based articulation encourages decomposition of surface conversational forms into modular, socially motivated interaction patterns. In this sense, RCDG can serve as a generative framework for designers tasked with adapting LLM-based health chatbots to fragile, shared, and collectivist care settings.

A future research direction involves the construction of end-to-end RCDG-compliant chatbot prototypes and their evaluation via standard and decolonial engagement measures. The necessity of formalizing interaction grammars that recognize silence, mediation, and fragility as primary, rather than residual, design factors is underscored by the empirical insights from Hameed et al.’s study. Continued development of RCDG may also yield analytic and practical tools for other domains characterized by communal decision-making and infrastructural instability.

---

All claims, definitions, and dialogue patterns above are grounded in the presentation and analysis in Hameed et al. (Hameed et al., 31 Oct 2025).