Page  00000133 HOW COMPUTER MUSIC MODELING AND RETRIEVAL INTERFACE WITH MUSIC-AND-MEANING STUDIES: OVERVIEW OF PANELISTS' SUGGESTIONS FOR DISCUSSION TOPICS Cynthia M. Grund University of Southern Denmark at Odense Institute of Philosophy, Education and the Study of Religions cmgrund@ifpr. sdu. dk ABSTRACT Inspired by the interest generated as the result of a panel discussion dealing with cross-disciplinarity and computer music modeling and retrieval (CMMR) at CMMR 2005 - "Play!" - in Pisa, a panel discussion on current issues of a cross-disciplinary character has been organized for ICMCO7/CMMR 2007. Eight panelists will be dealing with the two questions: a) What are current issues within music-and-meaning studies, the examination of which mandates development of new techniques within computer music modeling and retrieval? and b) Do current techniques within computer music modeling and retrieval give rise to new questions within music-and-meaning studies? 1. INTRODUCTION Researchers within computer music modeling and retrieval (CMMR) are accustomed to receiving requests from industry with regard to technology needed in order to facilitate product development. The worth of the CMMR community in relation to the recording industry is indisputable, and CMMR laboratories are of course interested in developing products which will make money. As a representative of the impoverished humanities community, my somewhat terse observation is, that whereas humanists usually are aware of their own intellectual value while being hopelessly inept at evaluating their own potential economic value, CMMR researchers are keenly aware of their own economic value, but more often than not quite naive as to how the breathtaking work which is being done within this field can be contextualized within the humanistic tradition. For several years now, I have been a CMMR enthusiast of sorts, attending conferences and getting this message out in various fora; see, for example, [3], [4], [5], [6] and [7], where [7] especially is useful as an orientation with respect to music-and-meaning studies. It is thus an honor and a great pleasure to have been asked to organize and chair a panel at ICMCO7/CMMR 2007 whose task is precisely that of dealing with these issues. I have chosen to organize this panel discussion around the following two questions: a) What are current issues within music-and-meaning studies, the examination of which mandates development of new techniques within computer music modeling and retrieval? b) Do current techniques within computer music modeling and retrieval give rise to new questions within music-and-meaning studies? Music-and-meaning studies have been chosen as the privileged area of concentration within the humanities in order to give the discussion better focus, while at the same time providing touch points with many of the general concerns that are relevant on the side of the humanities. 2. THE PANEL Our panel is composed as follows: Barry Eaglestone (Sheffield University, UK), Cynthia M. Grund (University of Southern Denmark, Odense, Denmark), Ole Kiihl (Aalborg University at Esbjerg, Denmark), Nicola Orio (University of Padua, Italy), Tommaso Perego (Conservatorio Giuseppe Verdi, Milan, Italy), Victor Pushkar (Taras Shevchenko National University of Kyiv, Ukraine), Jeffrey Trevifio (Center for Research in Computing and the Arts, University of California at San Diego, USA) and Solvi Ystad (Centre National de la Recherche Scientifique - Laboratoire de M6canique et d'Acoustique, France). The topics that will be examined during the course of the panel discussion are the result of submissions made directly to me by panel members, or interest in these topics has been indicated by panel members to the ICMC07/CMMR 2007 organizing committee. An expanded version of this introductory paper is planned, one which reflects (and reflects upon) the contents of the discussion itself. In it there will be far more references to published material of relevance to the topic, many more than are included here, where citations have been kept to a minimum due to limitations of space. 3. PROPOSED DISCUSSION 3.1. What are current issues within music-and meaning studies, the examination of which mandates development of new techniques within computer music modeling and retrieval? 3.1.1. Content Jeffrey Trevifio points out that: "The study of music and meaning applies semiotics and linguistics to music with frequent recourse to the term 'content' " [9] and asks What do you think about experimental artistic practices that claim not to have any content, i.e. the "non-expressive" works of John Cage or Morton Feldman, in which the composers "let the sounds be just sounds?" Do these alternative traditions 133

Page  00000134 complicate, confirm, or suggest revision to the foundations or applications of our current MIR technologies? And the theoretical analogue: What of Susan Sontag's assertion that the art of the 20th century demands an "erotics," not a "hermeneutics," of art [9]? Trevii'o is not alone with regard to concerns regarding content. Nicola Orio remarks that we need to be more aware of the matter of relevance to an information need as opposed to the matter of relevance to a query: Talking about relevance, which is usually relevance to a query rather than to an information need, one thing that is needed is one step beyond the query-byhumming or query-by-examples paradigms. More related to my personal research is the role of "music lexical units" as content-descriptors of music documents. It is somehow amazing that almost everybody agrees on the presence of boundaries in the music flow, yet almost nobody - including automatic systems - agrees on where the boundaries actually are (apart from very simple cases). Highlighting the role of these units can help improving weighting schemes aimed at more effective retrieval, and understanding how musical meaning is structured. The improving effectiveness of retrieval methods that use very coarse models and (almost) raw data, poses interesting questions on the redundancy of the musical language and on the correlation of different dimensions, which could be answered by applying methods from information theory to music (which is somehow in contrast with the ambiguous role of "information" in music) [9]. 3.1.2. Gesture and movement Ole Kiihls answer to our first question is unequivocal: The obvious answer from my point of view is gesture modeling. Some work has been done here already..., but as far as I can see, theories on human cognition prompt us to go further... especially with the idea of gesture as a complete, self-consistent unit, a gestalt [9]. As part of her response to our first question, S6lvi Ystad reports that she is doing research on perception of movement in sounds, but in contexts in which sounds are regarded without having been contextualized into a musical context. She explains that this is work%.concerning movements and how to identify the parameters that are responsible for the impression of movement that a sound can give you. Several types of perceived movements have been identified (approach, pass, fall, rise), and we would again like to identify the signal parameters responsible for the sensation of these movements. This is important, for instance, when we want simulate sounds for virtual reality purposes where a sound should fit a visual scene, or for composers who would like to make sounds that evoke particular sensations [9]. Another research field in which Sdlvi works is the investigation of the interpretation and the importance of timbre. 3.1.3. Recognition and response In addition to matters of movement, S6lvi Ystad's research on sound and analysis of signals seeks to... identify the parts of the signals that are responsible for the meaning we extract when we listen to a sound. For instance, when someone is knocking on a wooden object, we recognize the sound of wood without watching the object and independently of its shape. What we try to do is to identify the parameters that are responsible for the recognition of the material [9]. Issues of response and recognition, although on a level in which sound is clearly contextualized as music, are also of paramount importance to Ole KMhl, who remarks: As far as I can see, information retrieval techniques may also be developed in the context of a revival of Meyer's theory of emotional response to music. According to this theory, human feeling can be described as linked to a homeostatic state of the body, and emotion arises in response to changes in such a homeostatic state in an emotion loop, which feeds back into the original motivational state. The homeostatic state consists of a number of physiological parameters that have specific values at specific times, and therefore they can be quantified and expressed mathematically. Changes in the parameters of the musical auditory stream will prompt corresponding changes in the homeostatic state, so we have a system with two related sets of variables [9]. 3.1.4. Time, experience - and ecology Jeffrey Trevii'o, who regards musical experience as being best conceived as part of a "creative ecology," observes: In terms of the ontologically relevant time horizon of listened musical experience, is it realistic or useful to consider a model of musical experience that is locally relational - consists of mental models that feed forward and back with environmental stimuli and take note of changes within a certain time horizon - but not broadly analytical (1i.e., aware of "structure," out of time)? How would such a model alter MIR technologies and their applications [9]? Trevii'o also feels that Composers who harness MIR [need] to start discussing the details of how contemporary music is experienced. The institutional coincidences of composition's place in the academy already afflict its study with a predilection for arcane patterns of numbers, deracinated from perceived experience; MIR can help composition pedagogy in a textoriented academic environment, because it has the potential to discuss experiences via detailed, quantitative data sets [9]. 134

Page  00000135 3.1.5. Cognitive styles and cognitive typologies Barry Eaglestone points out that one of the key dimensions of cognitive style identified in the psychological literature is that which is defined in terms of the global/analytic distinction. Eaglestone has done research which suggests that cognitive-style traits among electroacoustic composers map well onto this axis. He theorizes that this suggests that there exist characteristic cognitive styles within the community of electroacoustic composers and that these cognitive styles "correlate with particular approaches to composition and also to levels of satisfaction with composition software." [9] Victor Pushkar regards cognitive styles - and more generally cognitive typologies - as important ingredients in an answer to our first question. Pushkar calls for the application of psychological typology.. to man-machine interaction within the process of music composition and performance. By psychological typology we mean - Jungian psychological type concept and related practical techniques; - representative systems: - cognitive styles. The research would lead us to - creating computer emulated intuition (large and distant target), - creating systems individual user optimized for individual users (small and practical target) [9]. based on sharing than the traditional, linguistic one with a sender and a receiver. The question of ecological perception is not new to the discussion of music and meaning, but CMMIR [computer modeling and music information retrieval, CMG] seems to provide ways of developing this issue, bringing it higher on the agenda. Both of these issues, the gesture and the emotion loop, are discussed at length in my book Musical Semantics [8], [9]. 3.2.3. The role of psychological knowledge within cross-disciplinary research on music and music-related technologies With respect to our second question, Victor Pushkar is one again of the opinion that current techniques within computer modeling and retrieval would benefit from more attention to psychological typology: Computer science is too serious matter to leave it only to programmers. If we want true interdisciplinary research on computer music, we should employ the widest range of human science, including philosophy, psychology, anthropology, sociology and art criticism. If we desire practical results, is the comparative research of human and artificial intelligence practical enough? Computer-based systems attempt to emulate intuition for some operations, but those attempts still yield far from satisfactory results. Modeling human intuition using artificial logic might possibly lead us to an in principle new concept of artificial intelligence. The practical experiments and applications should fit into some strategy. Besides explaining what we do it is useful to understand why we do it [9]. 3.2.4. Intentionality, culture, society - and memory In a vein which provides a bridge between our two questions, I (Cynthia M. Grund) feel that issues raised by the outsourcing of tasks to machine "listening" that traditionally have been carried out by human listening continue to require examination, and I keep an eye on the ways in which MIR continues to develop a huge repertoire of tools for crunching through vast amounts of acoustic data. This is as thought provoking, as it is impressive, however, if we begin to think about the consequences: Does the ability to access music information in this fashion somehow impoverish the quality of our connections with music and the connections music facilitates between an individual and her fellow human beings by - additionally - eroding our abilities to concentrate, get information from our environment into our memories and contextualize it, so we can remember it [4]? More generally, I continue to be intrigued by issues regarding the interrelations between CMMR, music, language and memory - on individual as well as cultural levels - and look forward to hearing how the panel will address these. 3.2. Do current techniques within modeling and retrieval give rise to new within music-and-meaning studies? computer questions 3.2.1. The meaning of "information" Nicola Orio's answer to our second question: Maybe the most important aspect is how the central concept of "information need" (introduced in textual information retrieval) can be translated into the music domain. The term "information" itself does not seem suitable for the music domain, and should be refined (or re-defined) by studies on music and meaning. I'm not a fan of the phrase "music information retrieval,"... I prefer the simpler "music retrieval". A formal definition of what the "music (information) needs" are would help also in rethinking the concept of "relevance" of music documents, which needs to be addressed from different angles: from the information retrieval point of view, because it plays a main role in any kind of evaluation, and from the musicological point of view as well [9]. 3.2.2. Ecology - once more Ole Kiihl reponds to the second question by telling about his work with a model for ecological perception where the subject feeds back to the auditory stream, reacting to it and changing it at the same time. This stream can be shared by several subjects/computers, and may lead to a more basic communication model 135

Page  00000136 It turns out that not unrelated topics are also of interest to Tommaso Perego, as his musical studies have brought him to research "what the relation between human will and culture - in order to create, interpret and receive music - looks like [9]." Perego would like to discuss results from brain research as they relate to communication and expression, and plans also to bring his experience as a composer to the table [9]. Jeffrey Trevii'o offers the following thoughtprovoking observation about the relevance - in an MIRcontext - of the individual-social distinction to the sound-music distinction: MIR technologies [are] increasingly allied with fMRI and other time-dependent methods that reveal hidden details of an individual's listening experience. If music is truly defined by its social aspect, we must stop retrieving information from sound and start retrieving information from music, a human's interaction with sound [9]. 3.2.5. Meaning - and ethics Closely related to the concerns expressed in 3.2.1 are others voiced by Jeffrey Trevii'o: What are our responses to what Nicholas Cook (director, Royal Hollaway Center for the History and Analysis of Recorded Music) calls, "the Luddite Objection"[2]: that MIR only shows you something that you already know, because you experience it, or that it shows you something that isn't experienced, so it doesn't matter? How does inquiry into Music and Meaning help us understand this objection [9]? On at related note, Trevii'o remarks: Nick Collins published a scathing rant in Array, the Journal of the International Computer Music Association... entitled, "Composing to Subvert Content Retrieval Engines" [1]. Given the current state of MIR technologies and applications, why might composers specifically view music information retrieval technologies as hostile to their practice? Are the goals of MIR detrimental to human creativity [9]? Trevii'o calls for humanist MIR researchers who can argue persuasively about how we might create situations in which MIR can potentially foster and inform creativity, instead of harming it, as is frequently assumed of MIR technologies by both scholars and the public. We have an ethical obligation to locate morality in the person using the technology, not in the technology itself [9]. 4. PRELIMINARY PAUSE; IT'S TOO EARLY FOR CONCLUSIONS It is exciting to see the extent to which the CMMR community is now beginning to reflect over what I, coming from philosophy, have considered to be issues of philosophical relevance within the context of CMMR; see, once again, [3], [4], [5], [6] and [7]. Let's hope that our panel discussion at ICMCO7/CMMR 2007 will give rise to a provocative, lively and ongoing debate concerning these matters. 5. REFERENCES [1] Collins, Nick. "Composing to Subvert Content Retrieval Engines," Array, winter 2006, pp. 37 -41. (~ Nick Collins, Barcelona, 31 May 2005.) [2] Cook, Nicholas. Remarks during the presentation "The Mazurkas Project: Creating a Toolkit for Performance Analysis" at the symposium Reactions to the Record: Perspectives on Historic Performance; Stanford University, April 19 - 21, 2007. Transcribed by Jeffrey Trevii'o. [3] Grund, Cynthia M. "A Philosophical Wish List for Research in Music Information Retrieval." (Poster) ISMIR 2006: Proceedings of the Seventh International Conference on Music Information Retrieval. Roger Dannenberg, Kjell Lemstr6m and Adam Tindale, eds. Victoria: University of Victoria, 2006, pp. 383-384. [4] Grund, Cynthia M. (2006) (version which was expanded upon after panel discussion) "Interdisciplinarity and Computer Music Modeling and Information Retrieval: When Will the Humanities Get into the Act." Computer Music Modeling and Retrieval, LNCS 3902. Richard Kronland-Martinet, Thierry Voinier and Solvi Ystad, eds. BerlinHeidelberg-New York: Springer Verlag, pp. 265-273 [5] Grund, Cynthia M. (2005) "Music Information Retrieval, Memory and Culture: Some Philosophical Remarks." Proceedings of the Sixth International Conference on Music Information Retrieval. Joshua D. Reiss and Geraint A. Wiggins, eds. London: Queen Mary, University of London, pp. 8-12.) [6] Grund, Cynthia M. (2005) "Double Jeopardy: The Interdisciplinary Study of Music and Meaning:" "Viewpoint" in Danish Yearbook of Musicology, Vol. 32, 2004 pp. 9-14. [7] JMM. The Journal of Music and Meaning, [8] Kiihl, Ole. Musical Semantics. Peter Lang: Bern, Berlin, Bruxelles, Frankfurt am Main, New York, Oxford, Wien, 2007. [9] Remark which has been submitted directly to me in correspondence, or which has appeared in remarks addressed by the panelist to the ICMCO7/CMMR 2007 organizing committee. My thanks to everyone on the panel for providing input to this overview. 136