Over at Jim Carpenter's blog, he has a series building on why aesthetic language generation (ALG) is difficult. There are plenty of discussions on the nature of working with aesthetic text generation systems from a writer/reader perspective, but very little on the challenges and questions raised in the actual construction of these systems (outside of the purely technical discussions within fields like Computational Linguistics). Jim has constructed a large scale electronic text composition system entitled "Erica T. Carter" and knows first hand the issues involved. Anyone interested in this area should keep an eye on his posts.

Jim's latest post raises the issue of interdisciplinarity in the construction of ALG systems and the overwhelming knowledge one has to have in order to effectively work in the area of ALG systems building. I'd like to elaborate some more on this topic for I've been thinking about this particular issue as well over the last year while working on the development of another ALG system, the GTR Language Workbench.

If one were to adopt an interdisciplinary approach to ALG, what disciplines should it include ? The three which seem most relevant are Computer Poetry, Humanities Computing, and Computational Linguistics.

The term “Computer Poetry” as it is used here needs some clarification. Computer Poetry is defined as a form of poetry and poetic creation more specific than the commonly encountered terms “E-poetry” and “Digital Poetry”. Computer Poetry is specific to poetry and poetics created and informed by tools which analyze and model linguistic structure (semantic, syntactic, grammatical, rhetorical etc.) and transform and generate text based on these computerized linguistic models. Computer Poetry is substantially different in character from those “Digital/E Poetries” which are grounded in visualization and animation techniques provided by Internet related technologies (Java applets, Macromedia Flash/Shockwave etc.) and the new modes of narrativity and interactivity introduced by these technologies and hypertextual structures in general.

For poets who wish to use the computer in their compositional processes (other than the "traditional" role of the computer as word processor) knowledge is required of some type of computer programming language in order to manipulate and/or virtually represent linguistic structures. The Canadian experimental poet Christian Bök has stated:

Poets may have to become advanced typesetters and computer programmers - technicians, polyglot in a variety of machinic dialects: HTML and Quark, PERL and Flash. Poets may have to learn the exotic jargon of scientific discourses just to make use of a socially relevant lexicon, and now that cybernetics has effectively discredited the romantic paradigm of inspiration, poets may have to take refuge in a new set of aesthetic metaphors for the unconscious, adapting themselves to the mechanical procedures of automatic writing, aleatoric writing, and mannerist writing - poetry that no longer expresses our attitudes so much as it processes our databanks. (After Language Poetry)

Programming concepts and knowledge of various language computing vocabularies are required to understand and implement poetry generating systems. These concepts and vocabularies are not ones found in the average English composition or Creative Writing program today. This is changing slowly as those who are aware of the issues push and redefine the boundaries within their academic disciplines. A knowledge of software development concepts within the fields of Computing Science, and more specifically Computational Linguistics is necessary.

Computational Linguistics (CL) is the study of language structures from a computational perspective. It is closely related to the field of Natural Language Processing (NLP), but where CL is focused on theoretical and technical issues of language modeling, analysis and generation, NLP is focused more on the development of practical applications based on the research efforts of CL. Like the multitude of phenomena in language itself, CL is diverse in its research interests. Current research topics include: machine translation, auto summarization, knowledge representation, semantic disambiguation, semantic similarity measurements, natural language generation, parsing and modeling of syntactic, grammatical, temporal and rhetorical structures, and computer models of creativity and humour. The list is large and fragmenting every year into further sub disciplines.

This field provides those interested in ALG with the requisite tools for tackling some of the key technical challenges in parsing and generating text. What seem like simple problems for us, can be extremely challenging for a computer and it is those working in CL who help overcome these challenges. For example, the simple task of distinguishing sentences from a block of text is trivial for a human, but for a computer this task is not as easy as one would think due to the contextual ambiguity brought about by the multiple uses of the period '.' symbol as a means of abbreviation, ellipsis and sentence termination. Before a computer program can transform or generate a text, it must have some underlying representation/model of that text to work with. CL provides the computer poet with the technical tools to help realize their aesthetic goals.

Poetry as a topic of interest for CL researchers has largely been ignored since Margaret Masterman's work with Computerized Haiku in the late 1960's (Masterman is an important figure in the history of Computer Poetry, and I hope to have a separate post on her contributions in the near future). There has been some recent work done by Hisar Manurung and Pablo Gervás but the topic of poetry has essentially been ignored throughout the history of CL. The reasons for this are not too surprising considering the most important qualities to poetry (subjectivity and ambiguity), are qualities in direct opposition to the general scientific research methods adopted by most CL research. CL is a sub discipline of Computing Science, and as a scientific discipline it relies on objective observation, measurement and quantification of discrete and reproducible results. This methodology does not fit well with the subjectively driven, seemingly random and limitless aesthetic and creative exploration required by an artistic practice like poetry.

Some within CL do see the over bearing reliance on formalization and its detrimental effect. Graeme Hirst, a prominent researcher in the field of CL, has made the statement "All AI knows how to do is carry on as if Wittgenstein had never existed". His statement referencing the philosopher Ludwig Wittgenstein attempts to illustrate the lack of understanding regarding the ambiguous and shifting nature of language, an understanding explored extensively by philosophers like Wittgenstein. Hirst has observed the following:

"I think that AI in general is sometimes just a bit too impetuous in its desire to formalize things, and it tries to turn things into systems or logics without fully understanding them, as if simply by doing so they would thereby come to be understood. ... This seems to arise from a combination of over enthusiasm for Western scientific method and a misunderstanding of the nature of language that borders on fear. In this view, language is a messy and highly imperfect medium that is not to be trusted, but rather must either be sidestepped entirely or be beaten into submission by means of logic and formalism." (Context as Spurious Concept)

The work coming from the field of CL is an important component to the advancement of ALG, and one could argue the reverse as well if CL were only able to "lighten up" a bit on its allegiance to rigid formalisms.

The term "Humanities Computing" (HC) is fairly new within the general field of Humanities. One of the more prominent advocates of HC, Willard McCarty, describes it as such:

“Humanities computing is an academic field concerned with the application of computing tools to arts and humanities data or to their use in the creation of these data. It is methodological in nature and interdisciplinary in scope. It works at the intersection of computing with the arts and humanities, focusing both on the pragmatic issues of how computing assists scholarship and teaching in the disciplines and on the theoretical problems of shift in perspective brought about by computing. It seeks to define the common ground of techniques and approaches to data, and how scholarly processes may be understood and mechanised. It studies the sociology and epistemology of knowledge as these are affected by computing as well as the fundamental cognitive problem of how we know what we know. Its tools are derived from practical work in computer science, but like that work its application of them uses models of intelligence developed in cognitive science and philosophy of mind. It tests the utility of these models to illuminate particular objects of study by direct involvement in the fields of application. Its object of knowledge is all the source material of the arts and humanities viewed as data. Like comparative literature it takes its subject matter from other disciplines and is guided by their concerns, but it returns to them ever more challenging questions and new ways of thinking through old problems.” (What is humanities computing ?)

Where CL brings important technological advancements in the area of literary modeling, generation and analysis, HC brings the important balance of an epistemological and philosophical framework for questioning and understanding of literary computing within a larger social context. HC also brings with it the scholarly skills for analyzing style, content and theme. Humanities departments within North American academic institutions today are in the process of redefining themselves to overcome the "technological hump" which has resulted in declining enrollment from apathetic students who are technologically immersed and who see little relevance in anything without a technological foundation.

One example of the technological questioning taking place within HC is related to the topic of computerized modeling of language and knowledge. Computing Science readily creates models, but with little understanding of the nature of modeling itself. This deeper questioning of the very methods used to derive scientific results can only help enrich our understanding of those results and help evolve and improve the scientific methods themselves. HC can provide critical insight into the technology developed by CL as well as provide important skills related to aesthetic analysis and modeling in the areas of style, content, theme etc.

Much more could be said on each of these three fields and the importance they can provide to one another, but I will save them for future probable posts. To sum things up, collaboration is the key in making larger strides in the area of ALG systems development. The breadth of knowledge required to effectively navigate the issues inherent in this area are too great for one to manage alone. An open source development process and an open and interdisciplinary developmental dialog where knowledge can be pooled, dissected, augmented, accessed and enhanced is essential.