The AI experiment at The Traditional Tune Archive

Dear User,{{safesubst:#invoke:string|rep|
|1}} the experimental phase regarding the use of Artificial Intelligence applied to the Traditional Tune Archive, has ended. {{safesubst:#invoke:string|rep|
|3}}

Here is a very brief report on the experiment we conducted thanks to your valuable input and support.

Count of Threads

The results are based on the content of about 220 threads representing a participation of about 10 percent of Traditional Tune Archive subscribers over the course of two weeks.

Participation was, however, open to all (subscribers and non-subscribers), so this is not an encouraging result.

Analysis of the Questions

The content of the questions shows a strong user interest in:

Historical Context and Origins: Questions often asked about the origin and historical background of specific tunes, composers, or traditions. Example: “Who wrote the tune ‘The Shadow Bowers’?”
Musical Theory and Styles: Users inquired about specific scales, tunings, and regional styles, such as “What are some Lydian mode slip jigs?” or “What defines Cape Breton music?”
Cultural Significance and Variations: Queries about cultural connections and variations in traditional music. Example: “What makes Donegal fiddling unique?”
Practical Music Application: Many participants sought practical advice, such as tuning methods for specific instruments like the Irish bouzouki or performance tips for dance music.
Archival Usage: Users frequently asked about how to navigate the Traditional Tune Archive effectively, such as filtering tunes by metadata.

While most questions were specific and well-articulated, a subset was ambiguous, lacking sufficient detail for precise answers. For instance, “Tell me about Galop tunes” could have benefitted from specifying whether the interest was in history, structure, or performance.

Analysis of the Responses

The assistant’s responses were generally:

Thorough and Accurate: Many responses provided detailed historical and musical context, often referencing archival materials or well-documented sources. Example: The explanation of “Sir Roger de Coverley” included its historical, musical, and cultural significance.
Practical and Actionable: Responses often offered actionable advice or instructions. For example, a user asking about Irish bouzouki tuning received a clear explanation of popular tunings and their applications.
Scholarly Tone: The assistant maintained a scholarly tone suitable for the knowledgeable audience. However, some responses lacked accessibility for less experienced users.
Occasional Gaps: Some responses were incomplete or overly brief. For example, a query about “Galop des frontières” yielded an acknowledgment of missing information without directing the user toward alternative resources.

Quality of Questions and Answers

Questions:

Strengths: Well-targeted, domain-specific questions reflected the participants’ engagement and expertise in traditional music.
Weaknesses: A subset of queries were vague, indicating a need for better prompt guidance or user education on framing questions.

Answers:

Strengths: Many responses were well-researched and appropriately detailed. The incorporation of links to archival entries added value for users seeking deeper insights.
Weaknesses: Occasional inconsistency in response quality, with some answers providing less detail or leaving unanswered aspects of user queries.

Overall Analysis on Results

The experiment demonstrated a high level of user participation and engagement, characterized by:

Diverse Interests: Questions spanned a wide range of topics, showcasing user curiosity about traditional music’s historical, cultural, and technical dimensions.
Interactive Learning: Repeat questions and follow-ups suggest users were exploring topics iteratively, which is indicative of deeper engagement.
Potential for Improvement: While the assistant served as a valuable resource, areas for improvement include better support for vague queries and more consistent answer quality.

Meaningful Examples

High-Quality Interaction: A user inquiring about the “history of Eleanor Plunkett” received a detailed response that included the historical context, cultural connections, and references to specific collections.
Underdeveloped Response: A question about “Gånglåt” led to a response suggesting archival exploration but lacked an overview of what “Gånglåt” represents, which might have helped the user start their search more effectively.

Recommendations

Guidance for Users: Provide examples of well-structured questions to encourage specificity.
Improved Response Consistency: Implement quality checks or a fallback mechanism for ambiguous questions to ensure that responses consistently meet user expectations.
Interactive Features: Enable dynamic follow-up questions or clarifications to allow the assistant to refine its answers iteratively.

Conclusion

The experiment was successful in engaging a knowledgeable audience, with the assistant proving to be a valuable tool for exploring the rich domain of traditional music. By addressing the areas for improvement, future iterations could enhance both user satisfaction and the depth of insights provided.

We will now take some time to evaluate the results of the trial and decide whether (and how) to adopt Artificial Intellligence to provide further research support.