|
Textones: The First Generation Textones: The Second Generation House of Bossa Textones: The Second Generation—Words, Tones, Space, and Time Shaun Sanders English 196 Honors--Professor Alan Liu 02/12/2009 Table of Contents 4. The Premise ………………………Why Model Literature? ………………………Why Audio Modeling? 7. Language and Cognition ………………………Initial Tonal Assignments ………………………The Joy of Parsing 10. Textones: II …………………….Goals and Hurdles …………………….Tags, Tones, and Parsers …………………….The Stanford Parser …………………….Order in Phrasing 15. Artificial Intelligence …………………….Intonation and Language ……………………Textones: II—Music in Language 22. The Imperfect Template ……………………The System: Phrases, Root Tones, and Structure ……………………The Unifying Action of Verbs ……………………Inflection 28. The Windhover 29. Conclusions and Prospects 32. Addendum—Contents A……………….Creation of Midi Files—Protools screen capture B………………Tone Assignments Sheet C………………Sequencing Grids— Part A: Sonnet 18. Part B: The Windhover Part C: Audio Files on CD D………………Stanford Parser— Part A: Stanford Tags. Part B: Sonnet 18 output. Part C: The Windhover output. E………………Robert Kane—Analysis of Sonnet 18 36. Work Cited The Premise If tones were applied to parts of speech and then sequenced according to the parsing out of a work of prose, would music be produced, and, if so, what would it sound like? This was the initial, simple thought behind the Textones Project; this small germ of an idea has developed into a many tentacled creature. Some tentacles have entwined themselves through areas of speech and cognition while others have latched on securely to digital modeling and computational interpretation of literature in the form of parsing and in-depth text analysis. This creature, in the guise of the Textones Project, continues to expand and grow into new areas, thus is has become necessary to revisit the initial goals of the project, assessing its achievements and acknowledging its shortcomings, before discussing its current status, its future goals, and its possibilities Why Model Literature? At the root of all literary interpretation, and, possibly, all literature, is a desire for increased understanding of the human condition. Once, it may have been enough to understand a protagonist’s actions based on a perception of their emotional or moral state, the main consideration being that motivation was housed within these spheres. Today, however, there is an ever-growing awareness that cognitive function plays a major role in human behavior, so the awareness of the analyst must grow accordingly. This may require a distancing, by the analyst/modeler, from a narrative in order to facilitate objectification of a work. This distance can be achieved via modeling because, to some extent, modeling removes the subjective element inherent in literary analysis; it is hoped an empirical type of effect can be achieved in modeling that is biased towards the quantitative as a means of exposing the qualitative. This approach to modeling is demonstrated by Willard McCarty, Professor at the Centre for Computing in the Humanities, King’s College, London, and author of Humanities Computing. McCarty compares the concept of a model that has a purpose, a model for, with a model that merely replicates, a model of, and finds that the model for is taking a front role in interpreting the Humanities on many levels. He cites anthropologist Clifford Geertz, who says that, rather than simply focus on a subject, we must examine and challenge the very ways in which we do our looking. Geertz says emphasis needs to be placed “on the microscope, not the bug under it,” indicating that the way we look at something clearly has repercussions upon our perception of it (McCarty, para 2). So it is with literature. McCarty himself says, “I’ve repeatedly argued that one should not think in terms of models but modelling” (para 5). Why Audio Modeling? Audio modeling of a literary work is not new. Music has been matched to text for centuries in operatic interpretations. However, the Textones concept of tonal modeling is that tones should be applied directly to parts of speech in a mechanical fashion, rather than in typical, musical fashion, music being designed purely with aesthetics in mind. Musicality was not an early consideration of the Textones concept. The more important goal was the discovery of possible recurring patterns within speech that would indicate universalities in language construction among writers of literary works, and, therefore, possible universalities in cognition. However, the Textones Project clearly recognizes that music creates concordance of thought where text fails—after all, music, like love, is a universal language—and with this in mind, musicality has become an important factor in the Textones listening process. Recent modeling methods reveal something interesting about literary close reading analysis; close reading is a personal, individual system of modeling inspired by the emotional responses of a reader to a work. Those responses are then intellectually interpreted via the reader’s secondary, mental process. In close reading analysis, we are often asked to examine why a text produces a certain feeling or result in the reader—in effect, we are asked to read between the lines. In Textones modeling, however, we seek to read only the lines. The hope is to ignore the emotive, believing it can hinder an understanding of the mechanical functioning of a text. This, in turn, may lead to an obscuring of the larger scope of a work. Therefore, Textones modeling encourages the idea that mechanical functioning is of greater importance in the initial comprehension of a work, mechanics within a work being a reflection of the artist's cognitive function, and that mechanical diagrams might be considered as a canvas upon which other analysis is then laid out in an attempt to balance emotional content with form. In Humanities Computing, Willard McCarty attempts to develop a system of modeling that could imbue text with type of “Personification” and thereby allow impartial literary analysis. However, he reveals a key paradox of literary analysis when he says, “Modelling strongly underscores the problematic nature of the long outmoded ‘quintessence of knowledge’ and of what is worth knowing” (Modelling 71). Put more plainly, he also says, “The fundamental problem raised here concerns the generation of meaning in language and so points to another dimension of modeling” (Modelling 70). Textones attempts to fulfill this other dimension of modeling as it removes the emotive response of the reader and reduces text to its mechanical base. Textones does not model language with language—to do so with impartiality may prove an impossibility simply because of the way language functions—Textones models language with tonal values which can then be assessed in another realm entirely, and on entirely different terms than language. An early question facing the Textones Project was whether tones could be applied to text in a way that would produce consistent results. While it might be easy to assign tonal values to one sonnet in order to make it sound interesting, a mechanism was sought which could be applied repetitively, thereby revealing traits within a variety of literary works, and which might aid quantitative modeling across genres; while assignment of tones to parts of speech may be subjective, as long as assignations are consistent, results will be non-subjective. Therefore, texts are not being interpreted in regard to their emotive context and the language retains its mechanical function and perhaps indicates cognitive function of the writer. Language and Cognition A key element of inspiration behind the Textones project was the hope that it might reveal something of the nature of cognitive functioning of the great writers throughout the ages. It was reasoned that, if comparisons could be made in tonal patterns from a variety of literary works, we might find similarities in the ways writers construct their ideas, ideas that have then been represented in patterns of speech. This idea of the working of the mind naturally lead to an exploration of the functioning of the brain when employing language, as well as an exploration of the brain as it processes sound. Both topics issued forth a great deal of complex information, for, while much is known about parts of the brain such as the Wernicke’s area and the Broca’s area, new information regarding the brain and the ways in which it processes language and sound continues to accumulate. In great part, this is due to the development of fMRI technology and the ability of neuroscientists to scan brain activity under certain conditions. For example, it has been discovered that, while language and speech may have central control mechanisms in the Wernicke’s and Broca’s areas, verbs and nouns may emanate from entirely different regions of the brain; a study by the Institute of Cognitive Neuroscience in London “suggests that the production of verbs in speech depends on cortical regions in the left frontal lobe,” yet this same region did not seem to effect noun production (Cappelletti). Initial Tone Assignments In the First Generation of Textones, recognizing that structures in language and structures in music have key elements, an attempt was made to have correlating values reflected within each sphere. For example, where a typical sentence structure consists of subject, verb and object, a typical major chord consists of root, third and fifth. Thus, tones were applied to individual words according to their values as independent parts of speech, resulting in a tonal sequence that represented a more or less linear progression through the verse under scrutiny. This proved to be a simple, effective method; however, it was limited in its interpretative power because, while it was effective when we were only interpreting individual words, it ignored phrase structures inherent in any sentence. When parsing verses in their entirety, it quickly becomes obvious that language works on several levels simultaneously—in a certain context, a noun may be used as an adjective, as in the phrase, “the car seat.” If the ultimate goal of Textones is to provide a consistent method of tone assignment to any type of text, it must acknowledge and allow for these interpretive levels of language. This means acknowledging, not just single words, but how they fit as part of phrases because we often use words purely in a contextually driven way. Returning to our example, “the car seat,” we cannot look at the word “car” in the normal way, as a noun, because its function has changed. In seeking to address this problem of context and word function, Textones needed to expand its vision. One aspect that proved successful in the initial Textones tonal assignations was the interpretation of the modifying aspect of any word. If, in a phrase, an adjective modified a noun, the adjective tone was sustained to modify the noun tone, and if a verb acted upon an object, as in “Johnny kicked the ball,” the tone for “kicked” would be held through to the end of the tone assigned for “ball.” This was done to mimic the way we use language; we cannot interpret an entire sentence or verse until we have finished reading it. Furthermore, a word at the beginning of a phrase often modifies a word near the end of the phrase. So it is with Textones; an adjective tone will form a polytone with its noun tone, “coloring” the noun tone with a “descriptive” element. Consequently, in some places in a Textones sequence, polytones build up, as do discords and resolutions. If Textones’ tone assignment was to take this labyrinth of meaning seriously, it needed to consider the possibility of more than two or three tones sounding at once; four, or even five, levels of tone might be required to successfully mimic, or model, use of language in literature. The Joys of Parsing It might be assumed that the parsing of text is a simple matter of separating words in order to ascertain their relative value in a phrase. However, no one who has attempted to parse a Shakespearean sonnet would make this assumption. Indeed, while Shakespearean sonnets were chosen because iambic meter has a relationship to musical rhythm, Shakespeare’s word choice, and his extending of phrase over several lines, occasionally leaves meaning somewhat obscure. Indeed, it seems to be generally acknowledged that meaning has considerable flexibility in Shakespeare, and perhaps that is part of Shakespeare’s magic, but for Textones it means close attention to detail is required in the parsing of his sonnets. * * * * * * Textones II: Goals and Hurdles In order to fulfill its primary modeling goal, the second generation of the Textones Project has created a new set of tonal assignations in respect to the multi-layered structure inherent in language use, thereby completing a new Textones version of Sonnet 18. In order that this new tonal sequence be “listenable,” efforts have been made to understand the occurrence of discord in a sequence and to minimize this occurrence through tonal assignment, ultimately arriving at a tonal system that can be applied to other works with consistent results that reflect not only word values, but phrase values. This has necessitated the development of a chord-type system to be applied to both phrases and words, as opposed to individual words only. The hurdle here was that, while it had been easier to minimize discord when we adopted only the single word system, with up to nine levels of meaning in any given phrase, discord becomes more prominent (see Kane Addendum). Therefore, considerable attention has been given to hierarchical importance of phrasing in a verse, especially as it might apply to cognitive recognition of meaning in language structure. As language interpretation can be subjective, determining which levels of meaning are of most importance—and should therefore have priority of representation in the tonal sequence—does have an element of subjectivity to it; a bias has been introduced as an aid in creating the second generation tone assignment system. This was seen as unavoidable for several reasons. Tags, Tones, and Parsers When Textones: I (Textones: The First Generation) was demonstrated at an English Department colloquium at UCSB in early 2008, it caught the attention of Jeremy Douglass, PhD. Douglass is a post-Doctoral researcher in Digital Humanities at UC San Diego, and when Textones: II (Textones: The Second Generation) was in its planning stages, Douglass was consulted. An early ambition of the Textones project was to create automated tonal assignment software that would modify text data into audio data. That is to say, text would be input and tonal sequences would result. If this function could be developed, it would become an easy matter to modify tonal assignations until discord was minimized across any given spectrum of literature. It was hoped that this effort toward minimization might reveal something about structure in language and structure in music. In retrospect, and in the context of this Honors project, this automated system was an ambitious vision. Douglas suggested that midi music files be created according to text analysis produced by an established parser, such as that operated by Stanford University. The Stanford parser is probably the most sophisticated parser in the US, and perhaps, for that matter, the world. There was obvious sense in Douglass’s suggestion; if the Stanford parser could be matched up with midi software, an operator would simply be required to input verse, either through voice recognition software or as a text file, in order to be rewarded with a tonal sequence. This would save the laborious undertaking of parsing out every line of verse manually and, thereafter, creating corresponding midi files manually—things that can become very time consuming. The Stanford Parser Parsing a Shakespearean sonnet can prove challenging. Linguists and analysts debate meaning in language, and, possibly, nowhere is this more evident than when dealing with the works of Shakespeare. Taking into account that 16th Century English was a slightly different language than that used today, and considering Shakespeare’s evident pleasure in creating complex syntax, parsing his work entails much research. Thus, the Stanford parser might be considered the perfect tool, one which would allow us to arrive at a definitive analysis of Sonnet 18, thereby enabling a rapid development of what has come to be called The Ultimate Textones Machine. However, as sophisticated as the Stanford parser is, for the purposes of the Textones Project it has one or two short comings. The first and most obvious problem Textones encountered with the parser was the extensive list of nomenclature, or “tags,” which it assigns to parts of speech. There are thirty-six single word designations, plus phrase designations. With a total of well over forty tags, Textones: II would have to function as a highly evolved musical computational device if it wanted to produce anything less than constant discord. While discord is a necessary and beneficial attribute in any tonal sequence—for, without it, harmony would not be apparent—the great range of possible tonal combinations required by the embracing of so many tones could easily result in an audio mayhem that might be impossible to interpret. In addition, the parser has specific designation for words that are difficult to categorize, such as the word “to.” Normally a preposition, the tag the Stanford parser assigns to “to” is simply “TO.” Furthermore, the parser works at levels far more subtle than would be considered normal in everyday language use. For example, it differentiates between a superlative adjective and a comparative adjective, which is not something many of us stop to think about when reading literature; we are mostly happy simply to know a word is an adjective (Stanford Parser Addendum pg 1). Due to these subtleties, it would proved difficult to regroup or reduce the parser’s tagging system to the twelve tone system employed by the Textones Project, twelve chromatic tones being one octave. The limitations imposed by a one-octave system was seen as an important prerequisite and an aid to simplicity, and because musical counterpoint is not something that need concern Textones at this stage. Like parsers that came before it (such as TAPor and CLAWS, CLAWS being a grammatical tagger developed by the University Centre for Computer Corpus Research on Language in the UK), and just like the search engines in our everyday computers such as Google, the Stanford parser is dependent on code recognition systems to work its magic. Each letter created by the text software in your computer is merely a representation of binomial code, at least as far as the computer is concerned. Words and sentences are identified the same way by the computer—merely as a sequence of codes. If a word is spelled incorrectly, the code will be unfamiliar to the computer and will be underlined in red; thus, the spell check has become a valuable tool for writers. The spell check program can also be annoying, though, when it “corrects” things it has not been familiarized with in its programming stages. There is zero flexibility to be expected from such a program, and this is the state of things with most computing programs; they are simply recognition mechanisms. As such, the Stanford parser is only able to make sense of that with which it is familiar. Order in Phrasing When presented with verse that works outside the norm, as many of Shakespeare’s do, the Stanford parser will apply incorrect nomenclature to phrase sequences. For example, when line five of Sonnet 18 is input, “Sometime too hot the eye of heaven shines,” the parser reads the phrase “Sometime too hot,” as part of the noun phrase “the eye of heaven,” thus making an adverbial phrase into an adjectival phrase (Addendum Stanford Parser pg 7). In fact, “Sometime too hot,” refers to the way the sun shines, rather than what the eye of heaven is. In order to correct this, we can rearrange the word order of the verse thus: “The eye of heaven sometime shines too hot.” Indeed, in the wonderfully in-depth analysis of Sonnet 18 undertaken by Robert Kane, a grad student at Cal State Northridge, in order to clarify Shakespeare’s intent, he does rearrange word order and continues on to parse the entire sonnet manually with an astounding display of literary acumen (Addendum/Kane pg 5). Meanings are clarified, and nomenclature of parts of speech has become more obvious than it would be otherwise; at the very least there can be some consensus on meaning in the sonnet. However, to adopt this system of rearrangement of word order when inputting text into the Stanford parser would mean that we are asking the parser to evaluate a sentence that was not written by Shakespeare, but by ourselves. Going a step beyond this possibly insignificant point, the tonal sequence that would result would be entirely different—it would reflect our adjustment of Shakespeare’s work, not the work itself. Yet, Textones’ goal was to represent cognitive use of language in a tonal sequence. At this juncture, an issue arises that borders on the philosophical as the realization of the scope and complexity of language becomes magnified. Once again, what had at first seemed like a simple idea of connecting tones to words has become complicated. Furthermore, the Textones Project sought to have Textones: II be a system defined by its simplicity of use, something that the average reader could experiment with, just as the Stanford parser is used. Thus, it might be used as a tool to engender interest in literature where none yet exists. Artificial Intelligence It was at this juncture that the Textones Project considered the ramifications of Artificial Intelligence. In February 2008, Wired magazine published an article titled “Like Minds,” written by David Kushner. In the article, Kushner explains the methods used by researchers at Stanford and MIT in their attempts to develop Artificial Intelligence within the computational realm. He also paraphrases the late Alan Turing, a cryptographer and mathematician who proposed that a machine could be deemed intelligent “if it could carry on a conversation that was indistinguishable from human conversation” (Kushner). Another famous AI pioneer cited is Marvin Minsky, an MIT researcher who was consulted by film director Stanley Kubrick to aid in the creation of HAL 2000, the recalcitrant computational brain in 2001: A Space Odyssey. The idea behind HAL’s mis-functioning is that, as a unit of artificial intelligence, it has become smart enough to think for itself, and it no longer wants to take orders from someone it considers to be of lesser intelligence. HAL has developed this intelligence through the amassing of information; simply by accessing vast amounts of data, HAL chooses a path of thought according to some principle known only to itself, thus it is “thinking.” (The path of least resistance, perhaps?) In the 1990’s, under Minsky’s mentorship, a young man named Pushpinder Singh started a project called “Open Mind” at MIT. Open Mind was designed to exploit the power of the internet in its creation of a vast databank, which could then be drawn upon by an AI machine in its attempt to converse. Internet users were invited to submit questions to be input into the Open Mind database. According to Kushner, Open Mind garnered “more than 700,000 submissions,” and is now “part of a Commonsense Computing division at the MIT Media Lab” (p 4). However, despite this accumulation of information as a resource, Singh was aware of the system’s shortcomings. In 1996, he wrote a paper titled “Why AI Failed,” part of which Kushner quotes: “To solve the hard problems of AI—natural language understanding, general vision, completely trustworthy speech and handwriting recognition—we need systems with commonsense knowledge and flexible ways to use it. The trouble is that building such systems amounts to ‘solving AI.’” At that time, Singh seemed sure of the ground upon which he stood. However, natural language understanding is the area Textones is most interested in, and to this project it does not seem possible, or important, to “solve” the entire problem in one fell swoop. It seems that attempts to create Artificial Intelligence operate under similar principles to that inspiring the creation of the Stanford parser: they both amount to the amassing of data bases and the matching of binomial codes in order to arrive at a conclusion. Such is the nature of binomial coding—everything is either correct or incorrect. AI’s output either matches the codes of the data that has been input or they do not; they don’t interpret. At best, they merely pick the closest match. In order to receive positive responses from his AI brain, “Cyc,” Doug Lenat, a Stanford researcher, began priming Cyc’s databanks in 1997 with “millions of everyday terms, concepts, facts, and rules of thumb that comprise human consensus reality—that is, common sense” (Kushner pg. 2). The thinking was that the more data that was input, the more likely it was that Cyc would recognize any random suggestion. These attempts at artificial intelligence have yet to succeed on Turing’s terms, yet this binomial thinking is the same as that of the Stanford parser. It has little room for the ambiguities of language, ambiguities that we take for granted in our everyday discourse with our fellow man. The parser’s inflexibility is reflected in its incorrect tagging of line five in Sonnet 18. As for Open Mind, in 2002 Singh stated: “Current statistical approaches are too weak to learn complex things. We need some really new ideas in machine learning…It helps to have the large datasets like mindpixel or openmind, but we’re still missing the right learning component” (Kushner p2). As compelling as this statement is, the Textones Project believes there will be more than one “right” learning component in the ongoing development of AI, possibly many more. However, Singh is correct when he says statistical approaches are weak; by themselves, they will not support the development of natural language computation simply because humans do not speak in monotones. Intonation and Language At a 1998 meeting of the Society for Neuroscience in Los Angeles, research was presented by Anne Blood of McGill University in Montreal, and by Mark Judy Tramo, a neurobiologist from Harvard University Medical School. Beatconscious.org, a website dedicated to the understanding of music and cognition, reprints the LA Times who say these scientists offered the information that “different parts of the brain seem to respond directly to harmony,” and reported that there was, “evidence of music’s remarkable power to affect neural activity no matter where they looked in the brain” (Beatconscious.org). This may seem a predictable result to those of us who love music, however, the way tone interacts with language is relevant to Textones understanding of intonation. The researchers stated that these “neural mechanisms of music may have originally developed as a way of communicating emotion as a precursor to speech,” and add that “researchers are looking for ways to harness the power of music to change the brain.” This research closely reflects work being done by Harvard psychologist Steven Pinker, who studies the way language use effects cognition; in his book, The Stuff of Thought: Language as a Window into Human Nature, Pinker says that the language we choose changes the way we think. This area of cognition interests the Textones Project, especially as it appears there may be an overlap in cognitive function between language and music. This overlap must effect our understanding of the world, and, therefore, literature. Dr. Daniel Levitin, author of This Is Your Brain on Music, has researched “brain-chemical activity” and has found biological responses to music that allowed the listener to analyze “structure and meaning” in music. Further, Levitin found that the cerebellum, which is closely related to physical movement, responded to what researchers suspected was “the brain’s prediction of where the song was going to go” (Levitin). This prediction of musical patterns appears to be an echo of the way in which psychologist Danielle Bergeron says the brain sustains cognition when we read text; we must chain “signifiers” together in our mind until we have read enough text to interpret the overall meaning the text is designed to convey. Accordingly, as a labor saving mechanism, we might attempt to predict the meaning in a sentence or phrase, just as we might try to predict the ending of a story; identifying patterns, either tonal or word based, is clearly a most human trait and was probably once essential to our survival as a species. It may still be. Perhaps what we enjoy so much about Shakespeare is his ability to withhold conclusion on the part of the reader until the last moment, or even until the work in question has been reread several times. He has a way of turning our listening or reading effort into an educational and edifying experience; there is something special about the moments of recognition and clarity that can be found in his application of language. Such brilliance would not have been possible had Shakespeare not had a solid understanding of intonation in language. Geoffrey Crossick, CEO of the Arts and Humanities Research Board in London believes there is a close connection between cognition, language, and comprehension. British newspaper, The Guardian, tells us that in 2004 there was a “push” by “Britain’s research councils…to understand what makes humans master communicators” (Sample). As part of this push for understanding, Crossick stated that language use “is about how we think.” He added, “It is essential that we understand language and how it works” (para 8). Also making an address as part of the program was Dr. Sophie Scott, a speech and neurobiology expert from University College, London. The Guardian reports that Dr. Scott “has shown that the brain takes speech and separates it into words and ‘melody’—the varying intonation in speech” (para 5). This intonation comes into play when we consider, as researched by a team from Cambridge University, “that about 80% of words have more than one meaning.” Here, we see the essential role intonation, or music, if you will, plays in our everyday language use. Thus, Textones’ research cannot omit intonation’s impact on language and text interpretation. As a final though, the Guardian reports that “Another research thrust is aimed at getting computers to understand language…something of acute interest to the military, business and in medicine.” Textones II: Music in language. At the culmination of the First Generation Textones System, UCSB Professor Alan Liu commented that, "The gap between 'music' and 'meaning' seems closer now than ever before." This statement has a prophetic ring, as language and music, especially the inner music which manifests from the individual as they seek to communicate an idea through speech, are inextricably intertwined. I suggest that, before Artificial Intelligence can integrate language use into its computers in anyway that resembles the “human conversation” capability proposed by Alan Turing, it must take the music of speech into account. In its second incarnation, it was important to the Textones Project that it not simply duplicate the first generation’s results, for to do so would mean applying a partial system of modeling to a new array of literary works; the result might become a type of half-modeling producing something rather incomprehensible. During early research it became clear that Textones was not embracing two key elements of language: intonation and rhythm. Rhythm was addressed, in part, by attempting to mimic iambic pentameter and enjambment in a Textones sequence. As researcher Robert Kane said at that time: “A sonnet is a nice structured microcosm of poetic language…The iamb is said to be the most common meter in the English language.” So it was apparent Textones encompassed rhythmic considerations at some level. However, intonation was ignored entirely to allow for a better focus on the development of a tonal assignation system. Intonation seemed, at first, to be somewhat counter-intuitive to the system being created. Just as Textones did not seek to create music, it did not seek to mimic intonation of speech because our focus was on text only. Yet, as research has continued and Textones has become a more complex concept, the realization has occurred that intonation is almost as much a part of text interpretation as it is of audible speech—it allows for comprehension of ambiguity in language in that, if a line of verse seems obscure, the natural response of the reader is to apply tonal values to the word order as a way of interpreting the author’s attitude toward their subject. This idea is easily understood if we consider a simple phrase such as, “I like that!” If an upward inflection is added to the word “like,” we know it is an affirmative statement on the part of the speaker. However, if inflection is added only to the word “that,” we understand a sarcastic intent is at play. And so it is with line five of Sonnet 18; we must read emphasis, inflection, and a sense of timing into “Sometime to hot” in order to understand its role as an adverbial phrase connected to the verb, “shines,” rather than as an adjectival statement connected to “the eye.” In this way, we see that tonal qualities are essential to natural language understanding of text, as well as of speech. For these reasons, Textones: II chose to introduce intonational elements into its tonal assignment system. Further, while never intending to focus on the creation of aesthetically pleasing music, the Textones Project must limit and control discord in order to aid cognitive recognition of tonal sequences. Therefore, it definitely intended to create some sort of harmonic understanding. Believing that all language has a tonal system as a base, Textones: II now seeks to match that harmonic understanding with intonation in language. The Imperfect Template The Second Generation of Textones is not perfected and, by design, it may never be. It seems more important to develop a reliable framework on which further experimentation can be carried out rather than come straight out of the box with something displaying a claim of completion. In common with the First Generation Textones System, this Second Generation is a work in progress. Flexibility is something Textones must take into account as it develops. Perhaps because language is so very flexible in its interpretation, any system attempting to mimic this understanding must also be flexible. In Clifford Geertz’s terms, Textones: II seeks to be an adjustable microscope. A "foolproof" Textones system is not being devised, but one that can adapt and adjust as research grows and results are garnered. Therefore, this generation of Textones is an attempt to provide a template upon which changes and modifications can be carried out. These modifications will probably be made for several reasons. Firstly and most obviously, the Textones system must allow for debate over correct parsing of any literary work; something which can clearly become a highly complex issue. Secondly, this template hopes to encompass the intonation of language, being adaptable in its tone assignments in an effort to interpret creative humanity's highly evolved art of intonation within language. This demands that the template has enough flexibility to modify and adjust for excessive discord when it occurs. As previously noted, some discord is important, but considerable discord reduces the ability of the mind to recognize any particular pattern. For recognition, some sense of melody is necessary. Third, spoken intonation itself is highly flexible dependent on context, and an “average” tone base might need to be developed, one that would mimic ordinary speech, before further development can take place on the intonation of imperatives, exclamations, and other irregular contexts. A trait of Textones: I that has been carried over into the second generation is the ambiguity surrounding tone assignments, the simple question being, “What tone should be applied to which part of speech?” Tonal assignment might be seen as purely subjective, as indeed it is. However, no tonal assignation is set in concrete; it can be modified or exchanged for any other tone within the one-octave, twelve-tone range Textones employs. Yet tones have been assigned to parts of speech with one driving principle: the minimization of discord. This is not a particularly exact science; if a tone was found to clash too often, it has been traded out with another tone that is closer to being melodic. This process has entailed hours of educated guess work and trial runs and is by no means finished, for the simple reason that many pieces of literature must be run through this system before the most common language patterns emerge. Ultimately, the machine will define itself; through sensitivity to discord and melody, any operator will be able to determine for themselves the most useful tonal assignments possible. It will merely become a question of the degree of discord one wants to result from their text input. So, while tone assignment may seem subjective, there are demanding parameters within which to work. The System: Phrases, Root Tones, and Structure In the new generation, phrase tones have been added one octave below individual word tones. Here, they act in the way a bass, or root tone, might, in effect anchoring the phrase for the length of the syllabic count the phrase occupies. This approach was adopted to reflect cognition of the brain and the way we read and interpret a sentence or verse. In an essay titled, “The Signifier,” psychologist, Danielle Bergeron, discusses the concept of “chaining.” She explains that language is a process of chaining signifiers together—in this case, words that represent symbols—and she makes a significant point when she states: “When we speak…the listener seizes the sense of what we want to say only once we have finished the sentence” (After Lacan p 63). She speaks in terms of the delay we encounter while reading as we recognize the mechanics of the language employed by the writer, and it is this delay that Textones tries to imitate with its “stretching” of an adjective tone to concur with a noun tone. Unlike the visual medium, where a scene can be taken in at a glance, language must be heard in its entirety before an interpretation of any one word can be confirmed. Furthermore, it must be recognized that we do not simply “add” words together, but interpret them as phrases within a sentence. Language instigates a progressive function that necessitates sustained cognitive reasoning for comprehension. This process is reflected in the sequencing of Textones; some tones—usually verbs, adjectives, adverbs, and prepositions—are held for the term of their role within a phrase, while others—like articles and conjunctions—are limited to their syllabic count. Realizing the Stanford parser was outputting data which was unusable for Textones’ purpose, Textones: II has assigned root tones to the following phrases: Subject phrase, containing a noun; object phrase, containing a noun; verb phrase; and prepositional phrases, adjectival and adverbial. These assignations might be considered a leap of faith by some, but as Willard McCarty says, “the model of exists to tell us what we do not know, the model for to give us what we do not yet have” (McCarty 24). He also calls a model “a design for realizing something new,” and adds that, “In modeling, one begins by privileging…knowledge, however wrong it may later turn out to be” (25). What McCarty is saying is that, rather than scrutinizing the subject in question in the hope of proving an analytical point, we are free to experiment with other systems of interpretation, no matter how strange or incorrect the initial presumptions of those interpretations may at first seem. It will be noted that because nouns can occur as either subjects or objects, a differentiation has been applied tonally to “subject phrase,” versus “object phrase.” In addition, parsers and linguists refer to articles, “the” and “a,” as well as possessive pronouns, as “determiners,” so, in some respects, they are examining these words at a level far more complex than that required by, or than is useful to, the Textones Project. There is no doubt that Textones’ renaming of phrases to suit its purpose is subjective and expedient, yet the new tags assigned by Textones have been considered in terms of cognitive functioning. As you read this sentence, do you really need to know that the word “the” is a “determiner”? It seems few of us analyze in such depth as we read, just as we don’t need to recognize whether a tone is a B or a Bb as we listen to a piece of music. We will, however, recognize that it has a role within a passage; we comprehend it as part of a phrase. Using this logic, it is apparent that, as subject, verb, and object phrases are the building blocks of any sentence, these easily identifiable phrases should be designated correlating tones of G, F, and C. Beyond those three building blocks, then, are prepositional phrases, which can be either adjectival or adverbial. This collection limits our possible phrase tone selection to five. All other words values are relegated to single word values only. Operating two octaves above these root tones are single word tones, numbering between two and six per phrase, depending on which phrase the words are part of. It is interesting to note that certain phrases will preclude the possibility of some words values being used in a sentence. For example, a subject phrase will never contain a verb, for even when I say, “The walking man,” the word “walking” has become an adjective. There seems to be an almost logarithmic interaction within sentence structure, and in this way, discord is automatically minimized. The Unifying Action of Verbs When analyzing differences between adverbial and adjectival prepositional phrases, Textones identified a relationship between adverbial prepositional phrases and simple verb phrases, in that verbs concern action which unifies subject to object, and this unifying effect seems to extend on into any adverbial prepositional phrase. When pondering this, consider “Shall I compare thee to a summer’s day?” Without the adverbial prepositional phrase, “to a summer’s day,” the “compare” is rendered impotent. Also of note in Sonnet 18 is the enjambment between lines 7 and 8. They are connected by a preposition, thus: “And every fair from fair sometime declines,/ By chance, or nature’s changing course untrimmed.” Line 8 is adverbial in that it explains the ways in which “fair declines.” Within the enjambment, we see the unifying principle of the verb in operation. Therefore, to represent this connection between verb phrase and adverbial phrase, Textones has assigned the same root tone to both. However, the individual prepositional word itself is represented by a minor third of that root tone, so a shift in mode will be noticeable and will function in the sense of the prepositional phrase becoming subjunctive, or subject to, the actual verb phrase. Contrarily, adjectival prepositional phrases are not always connected to the subject of the sentence. They sometimes aid description of the direct object, though, so since Textones has assigned the note of C as the root tone of any object phrase, adjectival prepositional phrases have been assigned the root tone of A, A-minor being the relative minor of C. In addition, as all single preposition word values have been assigned a flattened third of any given root tone, prepositional phrases will always adopt a minor cadence as representative of their “addendum” type role in sentence structure. Inflection Mention has previously been made of Textones’ attempt to mimic intonation of speech in its tonal assignment process. In this attempted mimicry, Textones recognizes the typical downward inflection of simple sentence structure in language use. Ask someone to read this phrase out loud, “The man kicked the ball,” and there is little doubt they will apply a downward tone as the sentence ends. Conversely, try to add an upward inflection at the end of the sentence and it becomes a question rather than a statement. Further, it seems a hierarchical order will be applied to normal tonal values. “Man” may hold a slightly higher tone value than “kicked,” which is in turn higher than “ball.” With this in mind, Textones: II applied a three tone system to overall sentence structure: The subject phrase equates with the note of G; the verb phrase equates with the note of F; and the object phrase equates with the note of C. This three tone cycle is the same as that used in many three chord pop songs and has a descending order, with an interval of two half-tones between G and F—or one note on the major scale—while between F and C there is an interval of four notes, or seven half-tones, on the major scale. Acknowledging that most normal speech patterns inflect in lesser increments than these, employing half-tones or even quarter-tones, Textones nevertheless considers this structure as an interesting place to start. This increase in intervals operates as an exaggeration of natural language intonation, then, and as such may improve pattern recognition in final Textones listening stages. However, it should be remembered that this is a template only, and has been designed to be adjustable in order to comply with further research concerning speech patterns and intonation, about which Textones still has much to learn. To listen to the results of Sonnet 18 and the Textones: II system, please refer to Addendum/Sonnet 18 audio file. The Windhover Shakespeare reminds us there is nothing new under the sun, and Textones is surely not the first project to try and understand language, music, and the mind. Many writers have experimented with word, rhythm, and meter choices as a way of sparking reader or listener interest. With this in mind, when seeking a second piece of literature to run through the Textones: II system, a work was sought that might be considered outside the norm in terms of typical language use and structure. Shakespeare is clearly outside the norm, but sense can usually be made of his word choice, and agreement might often be reached by analysts regarding meaning in his verse. Therefore, when reading Gerard Manley Hopkins’ “The Windhover,” for the first time, I knew I had found something that suited my purpose. Not only is the rhythm unusual, but word choice seems designed to remove constraints that might be applied by the reader upon the text. Hopkins embraces a style that throws logic and easy recognition to the winds. Just as the bird in his poem is held aloft by wind, so are our imaginations held aloft by his word choice. If we were to try and understand the work in a literal sense, we would be lost. We must, therefore, employ our imaginations and find links of meaning in order to make sense of the work; we may even choose our own “signifiers” and the ways in which we link them together. With his unusual style of expression, Hopkins might be considered a pioneer of literacy and cognition—he wants us to think outside the norm and truly examine the ways in which we use language. Therefore, his poem seems to be the perfect subject for a Textones sequence. However, the Stanford Parser’s tagging system comes into focus once again; while it makes surprising sense of “The Windhover” (for it is a truly remarkable tagger), its output is unsuitable for Textones, so complex does it become. See Addendum for poem and Textones: II sequence audio file of “The Windhover.” Conclusions and Prospects In the short time since its inception, the Textones Project has researched language structure, cognitive function, biopsychology, binomial function in computation, dissonance and harmony in music, the role of intonation in language, and the role of narrative in society. It has come to an in-depth understanding of verse structure in Shakespearean sonnets, as well as other works, and it has gained keen insight into the realm of artificial intelligence. However, in some ways it seems as if the project has merely revealed the tip of an iceberg, and that there is a great deal more to be discovered should the project continue. At the very least, Textones raises interest among academics whenever it is discussed, and this might be considered an indication of its possible longevity as a line of research. It seems that the project’s strongest point maybe that it connects many other lines of inquiry, lines that have been extensively developed in advanced academia, lines such as biopsychology and AI that are cutting edge technologies, yet, in some ways, are also in their infancy. It could be that the project’s greatest strength, then, is also a weakness. The original hope for Textones was that many works of literature might be run through it and that quantitative analysis would result. However, as more of the “iceberg” was revealed, and the myriad ways in which the project might develop became apparent, resources have become spread over an area much larger than was at first anticipated. The result has been that too few literary works have undergone the Textones treatment. Yet, the tonal system devised by the project had also to evolve, and this evolution would not have occurred had these many areas of study not been delved into. Textones: II has taken many angles of inquiry into consideration in its design of the flexible template, with which further literary analysis can now be undertaken. In pursuit of the Ultimate Textones Machine, further study needs to be undertaken in the marrying of text analyzers to midi sequencing. Further study regarding intonation in language might also be very rewarding, especially if we consider the possibility of tonal analyzers being married to the “natural language recognition” systems being attempted by researchers in AI. This is also a topic relevant to systems such as the Stanford Parser: imagine a parser that not only has voice recognition, but recognition of pitch and tone inherent in the spoken word. Textones considers this a vital point if robots are ever to converse with humans, as did HAL2000 in 2001: A Space Odyssey. While Textones tentatively researched what linguists call “prosody,” a field which deals with the hierarchy and value of stresses and tones in speech, the concept of prosody is not obvious in AI debates. (It is interesting to note that even robots like HAL, and androids such as Master Data in Star Trek, have a keen sense of intonation in their speech patterns—it seems we cannot conceive of language without it; intonation is a key element in the characterization of such figures.) One area Textones has not yet researched extensively is that of rhythm. An attempt has been made to match the iambic pentameter of Shakespearean sonnets with a type of ten-beat musical sequence, and for sonnets this has worked well. However, Hopkins’ “The Windhover” has such odd meter and syllabic count that a new approach must be taken with the time count of each line; some lines are quite are short while others are considerably longer. Because the Textones: II template is flexible, adjustment has been made to accommodate Hopkins’ unusual poem. Yet further research in rhythm and beat-count would, in all likelihood, reveal areas of interest and relevance to Textones. It may be correct to see an automated tone assignment system as the next stage of development for Textones, for, without such a system, Textones is something to be attempted only by those with time on their hands. Should this system come to fruition, a focus can then be returned to cognition and biopsychology of the brain when processing and employing language versus music; thus, we might come a little closer to finding meaning in music and its importance to literature, literacy, and the understanding of natural language. Addendum Addendum A: Creation of Midi Files Textones has used Protools 6.7 LE software for the creation of sound files in this study. The original Textones sequence of Sonnet 18 was produced applying analog sound to the beat-count grid in a type of cut-and-paste process. For Textones: II, however, midi files were created, and this means that, while a grand piano sound has been employed, a different sound can be traded out without a great deal of trouble. For example, a sustained organ sound might better highlight the ways in which the tones are creating chordal modes. The Textones sequence audio files can be heard online at www.HouseofBossa.com/Textones. They are also included on the CD that comes attached to this addendum file. Addendum B: Tone Assignment Sheet This sheet indicates the tone assignments used in Textones: II. All single word tone values are selected from one octave ascending from Middle-C. All phrase tone values are selected from the octave below Middle-C. The tonal system has been designed to be as interactive as possible. That is to say that constant values have been used in a type of orchestration system. For example, if prepositions are always the flat-third of any given phrase tone, we will come to recognize the minor cadence as always being a prepositional phrase. The same applies to possessive nouns, as represented by the sharp-fifth demonstrating the noun’s dominance over the object. Tonal assignment is still an area with a great amount of flexibility. This means it can continue to be perfected until the most appropriate assignations can be agreed upon. Because the Textones: II system has introduced root tones for phrase values, single word values that were carried over from Textones: I had to be reassigned. Therefore, there are considerable differences in tonal assignments between the two. Addendum C: Sequencing Grids These grids indicate the ways in which tonal assignments have been laid out in real time within the Protools Edit screen. They are a “blue print” of each line sequenced. They also indicate the way in which syllabic count has been employed. As with the first incarnation of Textones, a color scheme has been employed to aid in error recognition—if there are any errors in parsing, this is where they should show up. To see how the grids are applied within Protools, please note the screen-capture graphic attached to Addendum A. Addendum D: Stanford Parser Output These graphics show output from the Stanford Parser. Of particular relevance to Textones: II is the graphic showing line five of Sonnet 18. Also included in this section is a partial list of Stanford University’s tags which accompany their parser—you can see attempts made by Textones at sub-grouping the tags in the hope of adapting the parser to Textones’ system. Addendum E: The Robert Kane Addendum This addendum is included in acknowledgment of the valuable assistance provided by Robert Kane in his analysis and parsing of Sonnet 18. Indeed, it is an eye-opening exercise Robert has undertaken, and in viewing his work, one becomes aware of how complex language is as well as the ways in which language is layered. While we might not be conscious of this layering process every time we speak, there is little doubt that it is vital to our deeper language acquisition and use. Work Cited Barnet, Sylvan. “Shakespeare: An Overview.” Shakespeare: Richard III Signet Classic. Penguin Group. Hudson St, New York, N.Y. 10014 USA. 1998. Bergeron, Danielle et al. After Lacan State University of New York Press 2002 Cappelletti, Marinella et.al. “Processing Nouns and Verbs in the Left Frontal Cortex: A Transcranial Magnetic Stimulation Study” Journal of Cognitive Neuroscience MIT Press 2008 http://jocn.mitpress.org/cgi/content Castellanos, Angela. “Mapping the Brain’s response to Music—fMRI Studies of Musical Expectations.” Stanford Scientific Magazine Stanford University. Feb 2008 http://www.stanfordscientific.org/2008/02/17/mapping-the-brains-response-to-music-fmri-studies-of-musical-expectations/ “Introduction to Sound and Meter.” Purdue University Online Writing Lab (OWL). http://owl.english.purdue.edu/handouts/print/general/gl_soundmeter.html Kuschner, D. “Like Minds” Wired Magazine Feb 2008 p.134 Levitin, Dr. Daniel. This Is Your Brain On Music Penguin Group Ltd. Hudson St, NY New York 10014 USA. 2006 McCarty, Willard. Humanities Computing Palgrave MacMillan. 2005 Pinker, Steven. The Stuff of Thought: Language As a Window Into Human Nature Penguin Group (USA) Inc. NY New York. 2007 Sample, Ian. “Brain scan sheds light on secrets of speech.” Guardian.co.uk http://www.guardian.co.uk/uk/2004/feb/03/science.highereducation/ viewed 15/08/2008 Shakespeare, William. Sonnets. Oxquarry Books Ltd, Oxford, OX1 4LF, UK. http://www.shakespeares-sonnets.com/ Thompson, Clive. “Music of the Hemispheres” viewed 15/08/2008 www.beatconscious.org/brain.html#We%20Got%20Rhythm
|