Remind Me To Check The Stove Before I Leave The House: Authoring Personalized Context-Aware Smart Home Reminders Using Everyday Language DOI: XXXXXXX.XXXXXXX CCS: Human-centered computing Ubiquitous and mobile computing systems and tools

Reina Szeyi Chan, Sujendra Jayant Gharat, Maya Lampi, Yueran Jia, Avi K Srinivasan, Xiang Zhi Tan · 2026

Read Remind Me To Check The Stove Before I Leave The House: Authoring Personalized Context-Aware Smart Home Reminders Using Everyday Language DOI: XXXXXXX.XXXXXXX CCS: Human-centered computing Ubiquitous and mobile computing systems and tools by Reina Szeyi Chan, Sujendra Jayant Gharat, Maya Lampi, Yueran Jia, Avi K Srinivasan, Xiang Zhi Tan (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-23085.txt.

Introduction

Chan and five coauthors build a pipeline for making tailored, context-aware smart home alerts in plain language. Fixed schedules, location triggers, and simple rules cannot exploit rich smart home sensing. Users also struggle to specify complex alert logic. Their pipeline turns plain requests into set forms and working logic with time, activity, sensor, and state conditions. Two studies with 40 and 10 users show chat guidance helps turn diverse, ambiguous wording into flexible alerts.

Catalog record

CREATORReina Szeyi Chan, Sujendra Jayant Gharat, Maya Lampi, Yueran Jia, Avi K Srinivasan, Xiang Zhi Tan
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE94154
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.23085

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Remind Me To Check The Stove Before I Leave The House: Authoring Personalized Context-Aware Smart Home Reminders Using Everyday Language DOI: XXXXXXX.XXXXXXX CCS: Human-centered computing Ubiquitous and mobile computing systems and tools Reina Szeyi Chan, Sujendra Jayant Gharat, Maya Lampi, Yueran Jia, Avi K Srinivasan, Xiang Zhi Tan Abstract . Reminder systems commonly rely on fixed schedules, location triggers, or simple rules, limiting their ability to leverage the rich sensing capabilities of modern smart homes. A key challenge lies in enabling users to specify context-aware reminders without requiring complex configurations. We present a system pipeline that supports reminder authoring through natural language and conversational interaction. The pipeline translates user requests into structured representations and executable logic, incorporating time-based, activity-based, sensor-based, and state-based conditions. We conducted two studies to examine how users express reminder intent and how conversational support influences the authoring process. In Study 1 (N=40), we analyzed 233 user-authored reminders and identified challenges in expressing reminders with diverse and complex logic. Based on these findings, we refined the system and evaluated it in Study 2 (N=10), demonstrating improved handling of time-based, activity-based, sensor-based, and state-based conditions. Our results highlight the diversity and ambiguity of user expressions and show that conversational guidance can help structure these expressions into flexible, context-aware reminders. 1. Introduction People often need reminders not simply at a fixed time, but at the right moment in the flow of everyday life. A person may want to be reminded to take supplements after dinner, wash their hands when they arrive home from work, or take food out of the microwave if they forget it. These reminders are easy to express in everyday language, yet difficult to create concisely with existing reminder systems. Today, most reminder tools still rely on fixed schedules, singular triggers, or location-based cues ( 35 ; 21 ; 16 ) and are one-off commands – “remind me to call my son at 7pm” or “show my grocery list when I am at the grocery store”. However, many everyday activities do not follow a fixed schedule. Because time-based reminders must be tied to a specific clock time, they require assumptions about when the relevant activity will occur ( 30 ) . As a result, they may fail to capture the complexity and variability of real-world routines. Smart home technologies provide an opportunity to create reminders that are more closely aligned with everyday activity, rather than relying on users or systems to estimate the right time. For example, voice assistants like Alexa can notify users when someone is at the door, and phones can remind us to drink water when we wake up in the morning. These devices increasingly assist with routine tasks by sensing, interpreting, and responding to changes in our environment ( 65 ) . Beyond simple event detection, sensing capabilities can enable human activity recognition (HAR) ( 61 ) , offering new opportunities to condition on higher-level activities. With little or no training data, HAR can now detect complex resident activities ( 1 ; 64 ) . Combined with ambient sensor data, these capabilities enable a broader range of context-aware reminder scenarios that go beyond singular events. While these systems capture a wealth of contextual information about daily activities, their potential to support more meaningful, personalized experiences, such as intelligent reminder systems, remains underexplored. The challenge is that useful everyday reminders often depend on context that is richer than a single clock time, location, or sensor event. The most common approach is trigger-action programming (TAP) (e.g., “IF [trigger], THEN [action]) ( 62 ) , where a single event (“front door open”) or a specific time can trigger the reminder. While effective for simple tasks, this model breaks down for reminders that depend on more complex or delayed user behaviors ( 11 ; 34 ) . For instance, the reminder – “ remind me to take the food out of the microwave if I forget. ” – cannot be directly inferred from a single sensor reading. Instead, it requires the system to detect and interpret a sequence of sensors over a period of time. First, it needs to detect that the microwave door has been opened, indicating food was placed inside. Next, it detects a rise and fall in the microwave’s power consumption, indicating the microwave was used. Finally, the microwave door needs to remain closed for a fixed duration. Only after that can the system infer that the person has forgotten to take out the food. One straightforward approach is to extend existing TAP and rule-based methods by adding additional fields for recognizable activities, options for delays, a process to detect a sequence of sensor readings, and implementation of multi-condition triggers. However, this approach can quickly lead to large and complex configuration files that are difficult for average users to create and manage ( 38 ; 18 ; 39 ) . In addition, sensor availability and deployment vary widely across households, making it challenging to design one-size-fits-all rule sets that generalize well in real-world settings ( 61 ; 4 ) . As a result, system capabilities may not be transparent to users, and user intent may be constrained or misinterpreted by the system. In this work, we propose a conversational reminder authoring pipeline that guides users in creating complex, context-aware reminders that can be triggered by various sensor readings. The core of the system is an alternative representation of reminder execution logic, where each trigger is defined as a function that takes in the time, sensor, and HAR information as input. However, requiring users to directly construct such function remains challenging, as end-user programming approaches (e.g., IFTTT-style systems) still demand users to translate their intent into formal structures ( 62 ; 14 ) . Rather than requiring users to construct these functions directly, we introduce a conversational pipeline that allows users to describe reminders using everyday language. Prior work has shown that natural language provides an intuitive way for users to express reminder intent in terms of routines and situations ( 31 ; 12 ; 14 ) . Building on this, our pipeline translates these conversations into structured function representations. The system uses a large language model (LLM)-powered chat assistant to guide users through the authoring process, helping them refine underspecified requests and align their intent with the capabilities of the smart home environment. While this approach aims to make reminder authoring more natural and accessible, designing such a system is challenging. Users may express the same reminder intent in many different ways, and the system must handle this variability while supporting flexible interaction. Moving beyond simple, single-trigger reminders introduces additional challenges, and how users author more complex reminders in a conversational setting remains underexplored. To address this, we also investigate how users express reminder intent in natural language, how these expressions align with system capabilities, and how conversational systems can support the creation of context-aware reminders. To this end, we conducted two studies across six everyday scenarios. In Study 1, we collected 233 reminders from 40 participants and analyzed reminder authoring interactions to understand how users describe reminders and to identify gaps between user intent and system capabilities. These findings allowed us to refine the authoring pipeline and better align it with how

Read the complete text