Conversation
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request adds a pedagogical manual for the KIVA system, outlining its theoretical framework and instructional methodology. Feedback highlights discrepancies between the manual and the current implementation regarding the check-in process, word selection counts, instructional activities, and the ranking pipeline. Additionally, the reviewer identified an uncited reference in the bibliography.
|
|
||
| KIVA sessions are modeled on established tutoring practices and are designed to last approximately 30 minutes, typically occurring twice per week. Each session integrates three primary components: attention to the learner's well-being and engagement, discussion of reading progress and text comprehension, and explicit instruction in vocabulary and comprehension strategies. | ||
|
|
||
| At the beginning of each session, KIVA initiates a check-in with the learner. The purpose of this interaction is to establish rapport, assess engagement, and identify any logistical or motivational barriers to participation. The learner is asked about their recent experiences with the reading material and is encouraged to reflect on any challenges they encountered. Rather than providing direct solutions, KIVA prompts the learner to generate strategies and to consider involving caregivers when appropriate. |
There was a problem hiding this comment.
The check-in process described here (identifying logistical/motivational barriers, prompting for strategies, involving caregivers) is significantly more comprehensive than the current implementation in vocab-tutoring/flow_config.json (line 96), which only prompts for a friendly greeting and a question about the user's week. Consider updating the flow configuration to match this pedagogical specification to ensure the agent's behavior aligns with the manual.
|
|
||
| The session then transitions to a discussion of the text. The learner is asked to summarize what has occurred in the story, with follow-up questions probing their understanding of characters, events, and connections to their own experiences. The learner is also prompted to make predictions about upcoming events, supporting the development of inferential reasoning and engagement with narrative structure. | ||
|
|
||
| Vocabulary instruction is embedded within this broader comprehension-focused interaction. Typically, two to three words are selected per session, consistent with evidence-based recommendations for depth over breadth in vocabulary learning. |
There was a problem hiding this comment.
The manual states that "two to three words are selected per session," but the current implementation in vocab-tutoring/flow_config.json (line 181) explicitly requires teaching "exactly 2 words total." These should be aligned to ensure the documentation accurately reflects the system's constraints.
| Vocabulary instruction is embedded within this broader comprehension-focused interaction. Typically, two to three words are selected per session, consistent with evidence-based recommendations for depth over breadth in vocabulary learning. | |
| Vocabulary instruction is embedded within this broader comprehension-focused interaction. Typically, two words are selected per session, consistent with evidence-based recommendations for depth over breadth in vocabulary learning. |
|
|
||
| A key component of the selection process is the computation of a Tier 2 likelihood score. This score is derived from the calibration set of example Tier 2 words and reflects the degree to which a candidate word matches the distributional and semantic characteristics of this category. Features used in this computation include frequency profile, cross-domain usage, register, and semantic distance from high-frequency synonyms. | ||
|
|
||
| Finally, candidate words are ranked based on a composite score that integrates frequency, dispersion, contextual relevance, morphological richness, and Tier 2 likelihood. From this ranked list, a small set of words is selected for instruction, typically two to three per session. Care is taken to ensure diversity in word types and to avoid redundancy in meaning. |
There was a problem hiding this comment.
The sophisticated ranking and selection pipeline described here (using frequency, dispersion, and Tier 2 likelihood scores) is not currently reflected in the vocab-tutoring/flow_config.json (lines 181-182), which uses a simpler grade-level list approach. If this manual is intended to be an "implementation-ready specification," it should either reflect the current logic or clarify that the described pipeline is a future implementation goal. Additionally, the word count should be aligned with the implementation (exactly 2 words).
| Finally, candidate words are ranked based on a composite score that integrates frequency, dispersion, contextual relevance, morphological richness, and Tier 2 likelihood. From this ranked list, a small set of words is selected for instruction, typically two to three per session. Care is taken to ensure diversity in word types and to avoid redundancy in meaning. | |
| Finally, candidate words are ranked based on a composite score that integrates frequency, dispersion, contextual relevance, morphological richness, and Tier 2 likelihood. From this ranked list, a small set of words is selected for instruction, typically two per session. Care is taken to ensure diversity in word types and to avoid redundancy in meaning. |
|
|
||
| KIVA then provides a child-friendly definition of the word. Definitions are concise, non-circular, and expressed in language that is accessible to the learner. The word is subsequently anchored in the context of the text, with a sentence or passage illustrating its use. KIVA models how the meaning of the word can be inferred from contextual clues, making the reasoning process explicit. | ||
|
|
||
| The learner is then guided through a series of activities designed to deepen their understanding. These include repeating the word to strengthen its phonological representation, connecting the word to personal experiences, considering examples in new contexts, and making judgments about appropriate and inappropriate uses of the word. The learner is also encouraged to generate their own examples, supporting active engagement and transfer of knowledge. |
There was a problem hiding this comment.
This section mentions activities such as "repeating the word" and "making judgments about appropriate and inappropriate uses," which are missing from the instructional steps in vocab-tutoring/flow_config.json (line 181). Consider aligning the manual with the actual prompt instructions provided to the AI agent to ensure pedagogical consistency.
|
|
||
| Beck, I. L., McKeown, M. G., & Kucan, L. (2013). *Bringing words to life: Robust vocabulary instruction* (2nd ed.). Guilford Press. | ||
|
|
||
| Bowyer-Crane, C., Snowling, M. J., Duff, F. J., Fieldsend, E., Carroll, J. M., Miles, J., Götz, K., & Hulme, C. (2008). Improving early language and literacy skills: Differential effects of oral language versus phonology with reading intervention. *Journal of Child Psychology and Psychiatry, 49*(4), 422–432. |
There was a problem hiding this comment.
The reference to Bowyer-Crane et al. (2008) is listed in the References section but is not cited anywhere in the text. Please cite it in the relevant section or remove it to maintain a clean reference list.
1 participant
added instructions for selecting vocabulary words and proposed flow of instructional activties.