Computing point-of-viewBy Hugo Liu

Thesis Proposal for the degree of Doctor of Philosophyat the Massachusetts Institute of Technology

November 2005

Professor Pattie MaesAssociate Professor of Media Arts and Sciences

Massachusetts Institute of Technology

Professor William J. MitchellHead, Program in Media Arts and Sciences

Alexander W. Dreyfoos, Jr. (1954) ProfessorProfessor of Architecture and Media Arts and Sciences

Massachusetts Institute of Technology

Professor Larifari AufhebungKing of Candy Land

Computing point-of-view Hugo Liu Media Arts and Sciences, MIThugo@media.mit.edu

November 2005

AbstractA point-of-view affords individuals the ability to judge and react broadly to people, things, and everyday happenstance. Your same sense-of-beauty is versatile enough to judge almost anything you put before it, be it a painting, a sunset, or a novel's ending. Yet point-of-view is ineffable and quite slippery to articulate formally through words—just as light has no resting mass, perhaps it could be said that viewpoint cannot be measured in stasis. Drawing from semiotic and epistemological theories, this proposal narrates a computational theory for representing, acquiring, and tinkering with point-of-view. I define viewpoint as a self's collected situations within latent semantic spaces such as culture, taste, identity, and aesthetics. The topology of these spaces are acquired through linguistic ethnography of online cultural corpora, and an individual's locations within these spaces is inferred through psychoanalytic machine readings of egocentric texts. Once acquired, viewpoints can gain embodiment as viewpoint artifacts, which allow the exploration of someone else through interactivity and play. The proposal will illustrate the theory by discussing interactive-viewpoint-artifacts built for five viewpoint realms—aesthetics, attitudes, cultural identity, taste-for-food, and humor. I describe core enabling technologies such as common sense reasoning and textual affect sensing, and propose a framework to evaluate the judiciousness of point-of-view representations and the value of viewpoint artifacts in affording people new ways for organizing, shaping, and searching human narrative content.

1 IntroductionSince the late 1950s, every few years, some researcher in Artificial Intelligence has exclaimed eureka, that they have almost engineered a human intelligence, or some basal capability of a person. But in 2005, four years after the computer H.A.L. should have played tricks with man in space, Artificial Intelligence feels still the same distance from this ever-present mirage of human-level intelligence.

So it seems there were several bad paradigms stalling progress on representing and computing people. First, too grandiose of claims were made about formal logic and purely symbolic representation—nicknamed Good Ole Fashioned AI by its detractors. Logic, with its immaculate

Key WordsPoint-of-view ModelsUser ModellingCommon SenseTextual Affect SensingAestheticsCulture

and universal calculus, treats minds like Rube-Goldberg machines, and idealizes thought process the way that Descartes did. Logic failed because thought is far too flexible, rich and opportunistic than can be contained by a mathematically rigid, symbolically sparse, and non-opportunistic representation like first-order predicate calculus. Second, much ado was made about purely connectionist representations like artificial neural networks. The idea was that a properly wired ‘baby machine’ could be deployed in the world and re-derive human mental capability applying only first-principles. Like logic, this touched another extreme of the representational spectrum, namely it was representationally agnostic. The approach has yet to demonstrate compelling emergent intelligence. Marvin Minsky (1990) reported on the stalemate—he suggested that the common error was that ‘neat’ representations are too rigid to capture the diversity of human intelligence. He proposed that the intelligence modeling enterprise should instead take a ‘scruffy approach’—combining ‘multiple representations’ (Minsky 1988). In advocating the overthrow of Cartesian hegemony, Minsky paralleled Gilles Deleuze and Felix Guattari’s defining work of our time “A Thousand Plateaus: Capitalism and Schizophrenia—“ (1987) which destroys Modernism’s immaculate linear account of life and thought.

While some illusions have been overcome, Artificial Intelligence needed in the boom of expert systems and needs now again in the boom of knowledge-based approaches to sort out the importance of microscopic knowledge, given as expert rules, or “facts about the world—“ whatever that may mean. The shadow of Descartes haunts ‘facts’ as much as logic—for even if facts are received cum grano salis and their truth conditions are hedged, they still purport to evoked by people engaged in thinking. As a matter of reflexivity, much of our Open Mind Common Sense work at this lab (Singh, Barry & Liu 2004) is as vulnerable as Cyc (1995) to the stamp-collecting syndrome. Cyc’s 3 million assertions and Open Mind’s 800,000 sentence-based “facts” do not further them in a ‘horse-race’ toward human level knowledge. So long as representation is purely symbolic—as facts are—abilities granted to children like dexterously manipulating a ball (Singh 2003) or granted to adults like skill with people might occupy billions if not more sentences to describe judiciously. The warning to heed is that human intelligence is not about possessing rote knowledge. Having knowledge around does not ensure that it can be applied judiciously and opportunistically to form coherent thoughts and reactions.

Motivated by a search for coherent yet flexible representation and emulation of human intelligence, we identify point-of-view as a crucial metaphor for conceptualizing human intelligence. A layperson’s dissection of the “point-of-view” concept—two participants in an argument are debating the merits of an artwork and find that they disagree; one says to the other, “but from my

point-of-view, I see things differently.” Here point-of-view evokes an image of the two debaters standing at opposite ends of an opinion-space. In the middle is a large blob representing the true meaning of the artwork. The claim “from my point-of-view, I see things differently” reifies as one debater reporting that he can see a different side of the true meaning of the artwork than can the other debater, while allowing that she herself cannot grasp the whole meaning. So, having point-of-view relieves the anxiety of having true thoughts—instead, it privileges coherency and integrity over truth itself, for standing from the same vantage point, a debater will tend to report all sightings of meaning blobs with the same idiosyncratic tendencies, always seeing a certain side to things.

A point-of-view is easy. Every person is always operating under one or more points-of-view regardless of having reflexivity about it, because cognitive economy dictates that our knowledge and memories are always consolidated and systematized, with at least patchwork consistency. In Metaphors We Live By, George Lakoff and Mark Johnson (1980) report that language itself is organized and unified by culturally-specific metaphorical frameworks, which then shape the thoughts of cultural participants in the way that Lacan (1957) had presaged. For example, time is money, as in “I spent my day on you, I can’t believe I invested so much time in you, and you weren’t worth it.”

The grandeur of point-of-view’s economy is easily demonstrated. Look at this artwork, do you find it beautiful? Read this book ending, is it beautiful? Is this sunset beautiful? Or this government? Most likely, your sense-of-beauty viewpoint prepared you to judge all of these things, or at least attempt judgment. Point-of-view affords the immediacy of judgment over person, thing, idea, or situation placed within its realm. There is no need to move, to be agile, for judgment often happens like the natural reflex of a knee popping when stricken with a mallet. Whereas a facts-oriented view of thought requires conceptual knowledge, every person has abundant judgmental knowledge for virtue of possessing points-of-view like sense-of-beauty, sense-of-humor, sense-of-cultural-identity, a palette for food, and a personality. It is not necessary to store each judgment as a fact, for point-of-view’s lucidity readily produces judgments as it reacts to whatever fodder is put before it.

Economical, flexible, and broad in applicability, point-of-view is a powerful framework and mover for human judgmental thought, arguably exceeding conceptual and logical thought in breadth and utility. If point-of-view could be successfully modeled, acquired, and animated computationally for a few important human realms such as aesthetics, identity, and opinions, in toto, the computational system would be emulating a significant basal capability of human thinking.

To be clear, a computational model of an individual’s point-of-view would constitute a stereotype of that person that is not as agile, and that might make the same judgment

if asked ten times in a row. But I will argue in this proposal that this would still be an extremely useful stereotype. What if every person could access a computational stereotype representing 80% of their mentor’s judgmental capability—to bounce random things off their ‘virtual mentors’ without resource bounds? There would be real consequences for education if students could ‘tinker’ a la constructionist learning (Papert & Harel 1991) with the stereotyped opinions and perspectives of mentors, computationally producing ‘just-in-time’ and ‘just-in-context’ reactions to the student’s actions.

The goal of the research proposed here is to design, build, and validate systems for 1) modeling an individual’s point-of-view within various realms—such as aesthetics, attitudes, and identity; for 2) automatically acquiring an individual’s point-of-view model through machine readings of egocentric (self-revealing, self-describing) texts and 3) organizing the model into coherency; and for 4) animating point-of-view placed inside interactive artifacts such as virtual mentors by causing the artifact to judge and react to a very broad range of things placed before it, ‘just-in-time,’ and ‘just-in-context’.

I plan to address these four steps as follows. 1) To develop representations of viewpoints across the realms of concern, I will draw heavily from well-established semiotic and epistemological theories of said realms from the psychology and literary theory literatures. For example, Carl Jung’s Modes of Perception (Think, Intuit, Sense, and Feel) (1921) form the dimensions of my proposed aesthetic viewpoint space, as I pose aesthetics as the perceptual manner and priority with which an individual approaches some topic—a realist sees a sunset, but a romantic might prefer to feel the sunset. The realist is thus located at the position, 100% Sense, 20% Think, 20% Intuit, 20% Feel, for example. 2) To automatically acquire an individual’s point-of-view, I propose to apply natural language processing tools such as my widely used MontyLingua package (Liu 2002), in conjunction with my common sense reasoning package ConceptNet (Liu & Singh 2004b), and my textual affect sensing system known as Emotus Ponens (Liu, Lieberman & Selker 2003). In particular, I anticipate that reading emotion out of text will be vital to modeling viewpoint because human judgment often reifies in narratives through emotional appraisal or mannerisms around a topic’s discussion. 3) To make point-of-view models somewhat coherent, I will apply analogy-based reasoning (Gentner 1983; Fauconnier & Turner 2002). For example, knowing that a person loves trees, by analogical-extension, they might also love rocks (note that this is different from a layperson intention for the word ‘analogy’); however, pitfalls must be avoided—for example, a dog lover may hate cats, even though dogs and cats are both pets. 4) Finally, to animate point-of-view, I proceed along the methodological lines of Just-in-Time-Information-Retrieval (JITIR) (Rhodes & Maes 2000) which prescribes that interface

agents—in my case a virtual mentor reacting to things that you are writing or doing using its viewpoint—continuously mine present user context and utterances, searching for opportunities to retrieve and present relevant information – in my case, a viewpoint-produced judgment about whatever the user is doing—on the chance that it can lend insight, inspire, or teach the user.

While the acquired models will not be absolutely complete or always correspond to true viewpoint, and while none of the produced reactions will be as spontaneous or as flexible as those of the actual person, I believe that even a first-order approximation of model acquisition and animation can produce incisive models of individual perspective, that upon animation will afford novel and effective new ways to search, gain insight into, be inspired by, and connect with someone else and their collected narrative content. I have italicized three words in the previous sentence because these words constitute the tripartite agenda of our Ambient Intelligence Group. I believe that our group has the most to gain from such a thesis, as the methodological conclusions of this research would directly inform much of the impact we seek for our technologies to have on people.

Finally, this thesis is as diverse and as simple as I believe Media Laboratory research should be—diverse in the methods and theories it draws from, but simple in that it is attacking a basic problem of relevance to people—so basic that it’s goal could be explained to anyone on the street. This thesis draws from Sociology, Literary Theory and Psychology for its computational framing of point-of-view, from Computational Linguistics and Artificial Intelligence for reasoning about text, and from Interaction Design for designing point-of-view artifacts. I have developed but not assembled nor integrated some implementations for this thesis, and already it forms the basis for an AAAI workshop on computational aesthetics, which I will co-chair upon the proposed completion of this thesis. To do justice to an idea as complex and with as long as a history as ‘point-of-view,’ it will be important to clothe the thesis in all of the relevant literatures and to spend as much time on a computational theory of point-of-view, as on technical details of implementation. Otherwise, this work would lose a golden opportunity to be absorbed by an AI community that is interested in how machines can appraise beauty and emotion, and by a humanities and cultural studies community that would be very interested in the computation of its long-standing but thought incomputable theories of identity and aesthetics. The rest of this proposal will reflect my emphasis on the importance of grounding this thesis in the literatures, and on the importance of distilling reusable methodology and a robust theoretical framework. I will, of course, motivate all theory with many implemented demonstrations and task-based evaluations.

2 Proposed ResearchIn the following subsections I propose a theoretical framework for representing and computing point-of-view, and detail how the framework and its associated methodology will be supported by point-of-view systems I have been researching for several realms including aesthetic-space, opinion/attitude-space, cultural identity-space, tastebud-space, humor-space, and commonsense-space. I have implemented many of these systems, and completed various task-based evaluations. I propose to reframe these disparate systems so that they support and illuminate a central theoretical framework. In what follows I also nominate core technologies, knowledge representations, and techniques needed to assemble point-of-view systems. Finally I outline an evaluation strategy.

2.1 Theoretical Framework

Overview. I define viewpoint as a self's collected situations within latent semantic spaces such as culture, taste, identity, and aesthetics. The definition reflects a school of psychology called Situationalism ( ) or Social Constructionism ( ), emerging out of Jacques Lacan’s notion that the ego is always defined in the other (1957). The topology of these spaces are acquired through linguistic ethnography of online cultural corpora, and an individual's locations within these spaces is inferred through psychoanalytic readings of egocentric texts (self-revealing, self-describing), for example, a diary, a research paper, a social network profile. Once acquired, viewpoint models gain embodiment as viewpoint artifacts, which allow for self-

Figures 1a-d. Viewpoint models. (clockwise from upper-left) a) viewpoints are computed as situations within latent semantic spaces; b) a realist’s perspective in a 5-dimensional realm defining perceptual aesthetics—a topic, such as “sunset” depicted here, casts semantic shadows on each dimension, as shown; c) represents opinion-space as a sheet of topics which overlays two different perspective sheets, whose alignments are shown; d) depicts cultural identity space as a fabric of cultural interests; an individual’s ‘pattern of liking’ constitutes an ethos imprinted in the fabric.

reflection or the exploration of someone else through interactivity and play.

Entertaining the idea of point-of-view in the abstract (Fig. 1a) is uncontroversial, but the real theoretical challenge posed in this thesis is to reify the abstract notion into workable computational systems. A computational theory of point-of-view will describe the dimensionalities and properties of some major viewpoint realms (Figs. 1b-1d are examples soon to be explained); will specify how the topologies of such spaces can be acquired and how individuals can be modeled within such spaces; and will demonstrate how space+location can together be used to predict an individual’s reaction to any person, thing, idea, or situation put before it—I will refer to this something collectively as “fodder.”

Knowledge representation for viewpoint spaces. Figs. 1b-1d illustrate three varieties of knowledge representation used in this thesis research to model latent semantic spaces. But why three and not one? Because sometimes the dimensionality of a space is known (Fig. 1b) while other times it is not (Figs. 1c-d). The goal of locating an individual’s viewpoint within a space is to reduce the task of predicting reactions to fodder to simple Cartesian distance measurements. Ideally, a dimensional space such as Fig. 1b can be identified as appropriate. In dimensional spaces, information is most organized and unified, and the notion of distance is most straightforward. Dimensions of space could of course be inferred statistically through approaches such as Latent Semantic Analysis (Deerwester et al. 1990), Support Vector Machines (Joachims 1998), Multi-

Figure 2. Semantic diversity matrix. Point-of-view spaces can be conceived in terms of their consistency and connectedness—for each case, an appropriate knowledge representation is specified. The top row is semiotic/symbolic in quality; the bottom row is ethnographic/connectionist in quality.

Dimensional Scaling (Kruskal & Wish 1978), Principle Components Analysis, and the like, but in these cases, the quality and human-readability of the dimensions cannot be assured—for example, in the document classification problem, LSI can appropriate one dimension for each word or punctuation mark. The proposed thesis only appropriates dimensional spaces when the dimensions are semiotic in nature—that is to say, they are named, canonical, and well studied in the Psychology and Cognitive Science literatures. Fig. 1b depicts the example of a space for perceptual aesthetics—its dimensions are taken from Carl Jung’s modes of perception (1921), which is well studied in psychology and precursor to the popular Myers-Briggs Type Indicator (MBTI) model of personality (Briggs & Myers 1976).

When semiotic dimensionality is not known, it is still possible to aim for a fully connected representation such as the semantic fabric in Fig. 1c. If that is unavailable, a semantic sheet representation is appropriate (Fig. 1d). Inspired by Marvin Minsky’s “causal diversity matrix” organizing reasoning methods by the number of causes and effects (Minsky 1992), in Figure 2 I pose a “semantic diversity matrix” which organizes knowledge representation according to the connectedness and consistency of the semantic spaces they best represent. The reader should note that this several species are omitted in the graph. For example, if a third dimension were introduced for semioticity, we could distinguish “dimensional spaces” as being either a semiotic /structuralist space like Jung’s modes of perception, or as being a data-emergent “quality space” for concept formation in AI ‘baby machines’ (Gärdenfors & Holmqvist 1994; Johnannesson 1996; Gärdenfors 2000).

In research already completed, two viewpoint realms were modeled using semiotic dimensional spaces—Jung’s modes of perception was the basis for a perceptual aesthetic space (Fig. 1b), and the well studied “Big Five” model of personality (John 1990) was the basis for—aptly—the personality viewpoint space (Liu & Mueller, forthcoming). For cultural identity space—i.e. the space of things that people like such as music, books, sports, and subcultures—a semantic fabric representation (Fig. 1d) was chosen since the space was rather inconsistent, and since there was an opportunity to mine a fully connected structure out of a large corpus of social network profiles (Liu & Maes 2005a; Liu, Maes & Davenport 2006). Opinion-space—i.e. all possible systems of attitudes toward arbitrary topics about the world, about politics, or about academic subjects—is believed to be much more unorganized and opportunistic due to the myriad of causes and conditions which can shape opinion like social influences and experiences; therefore a “semantic sheets” representation is chosen ( ). “Semantic sheets” are also used to create a model of humor-space ( ). ConceptNet’s semantic network model of common sense reasoning (Liu & Singh 2004b) and Synesthetic Recipe’s annotation model of tastebuds (Liu, Hockenberry & Selker 2005) will also be discussed as viewpoint spaces of low

consistency and moderate connectedness, for the sake of theoretical completeness.

Organizing Principles of Viewpoint. Consistency gives shape to viewpoint space. Without consistencies, applying viewpoint models to predict reactions to fodder would have to resort to memory-based and case-based reasoning—such that if a fodder is not explicitly specified in the model, no reaction could be given. Dimensional Spaces are fully consistent and organized because a total ordering exists for each dimension. Organizing principles in Semantic Sheets and Semantic Fabrics are more opportunistic and only patchwork consistency exists. For Semantic Fabrics in the example of cultural identity space, I nominate three topological organizing features – hub-and-spokes, n-cliques, and neighborhoods, discussed elsewhere (Liu, Maes & Davenport 2006) in preliminary form. For Semantic Sheets, three organizing features are nominated—1) Minsky’s imprimer theory (Minsky, forthcoming) informs how one individual’s system of attitudes/opinions is partially structured by the systems of their parents and mentors; 2) folksonomies of topics imply underlying consistency of attitudes (e.g. “macramé” is a subtopic partial structured by the topic “arts & crafts”); and 3) analogical reasoning (Gentner 1983; Fauconnier & Turner 2002) conceptual blending, structure-mapping) can be applied just-in-time to predict reactions to unknown fodder (e.g. attitude toward “rocks” can be predicted by attitude toward “trees” by their conceptual resemblance).

Techniques from truth maintenance systems (Doyle 1980) are applied to maintain patchwork consistency, though contradictions do occur and these are presented as “soft-constraints.”

Simulating Viewpoints. Statically viewpoints are space+location, but to fully appreciate and understand a viewpoint, it must be animated and allowed to react to a broad many things. Analogy (Gentner 1983; Fauconnier & Turner 2002) and context-biased spreading activation (Collins & Loftus 1975; Liu 2003) are chief techniques for anticipating how the implications of a viewpoint inferred from across many contexts can then be applied to create a reaction in a new context. Although with viewpoint models we go beyond the “rote” memory-based application of old ideas to new fodder, viewpoint simulation is still not capable of applying viewpoint models in any particularly clever way to new situations. Humans are capable of evolving their viewpoint nimbly as new fodder presents opportunities for belief revision, but machines are not capable of simulating this complex self-dialectic (Bakhtin 1935). A goal for the thesis is to discuss how the simulation of viewpoint could become dialectical, how an artificial viewpoint could contradict and overcome itself cleverly—what Hegel calls Aufhebung (1807). Viewpoint models and simulation carry specific implications for dialectics—a central problem in

critical theory. If Aufhebung could be simulated, it would represent a major breakthrough for the computation of inspiration.

To animate computed viewpoint models, viewpoint artifacts are created—such as the Identity Mirror (Liu, Maes & Davenport 2006; Liu & Davenport 2005), the Aesthetiscope (Liu & Maes 2005b; Liu & Maes 2006), virtual mentors in What Would They Think? (Liu & Maes 2004), and avatars in Synesthetic Recipes (Liu, Hockenberry & Selker 2005). Viewpoint artifacts reify space+location models by having them constantly react just-in-time and just-in-context to a broad range of fodder put forth to them implicitly or explicitly by a user, and by visualizing these reactions through visual metaphors. Furthermore, each viewpoint artifacts allows for tinkering, play, and explanation, e.g. virtual mentors can “justify” their reactions with quotes, and identity can be negotiated in the Identity Mirror by a “dancing” interaction in front of the mirror. The importance of tinkering is likely due to the fact that a reaction’s motivation cannot be easily grasped without exploring the immediate context and conditions surrounding the reaction.

2.2 Core Enabling TechnologiesThree core technologies that drive the acquisition of viewpoint models from machine readings of text are natural language processing, common sense reasoning, and textual affect sensing. Machine learning techniques and hand engineering of many support semantic knowledge bases are also important, but they are not discussed here.

Natural language processing. Because some viewpoint spaces are acquired by ‘linguistic ethnography’ over cultural corpora available online (such as a corpus of social network profiles, or a corpus of conservative versus liberal news texts), and because all of an individual’s locations within spaces are acquired by psychoanalytic readings of egocentric (self-revealing, self-describing) texts, natural language processing (NLP) is central to this research. Relevant NLP tasks include discourse segmentation, tokenization, named-entity recognition, spelling correction (Levenshtein 1965), part-of-speech tagging ( ), deixis resolution ( ), verb and noun chunking ( ), prepositional linking ( ), gisting syntactic, semantic, and thematic role frames ( ), natural language generation ( ), topic spotting ( ), summarization ( ), and statistical language modeling ( ).

For the bulk of these tasks, I have developed a natural language understanding platform for Python, called MontyLingua (Liu 2002)—now widely used since my releasing it to the Computational Linguistics and AI communities.

Commonsense reasoning. Commonsense reasoning is a core component of machine readers that will read texts to acquire viewpoint spaces and locations. The essential insight that distinguishes machine reading—or Story

Understanding / Narrative Comprehension as it is also called—from mere deep text parsing is that more than what a text explicates, it also implies and insinuates through subtext, and it requires contingent knowledge in the form of backtexts to decipher the full meaning of an utterance. As a community, Computational Linguistics has focused on Syntax via Grammars and Formal Lexical Semantics via dictionaries and WordNet (Miller et al. 1990), surely due to the deep impression left by Chomsky ( ) on linguistics. The rest is forced into relatively under-explored buckets called “semantics”, “pragmatics” and “discourse theory.” To read subtexts and with backtexts, the Artificial Intelligence community has applied approaches such as Schankian scripts and plans (Schank & Abelson 1977), and more recently, large scale databases of world knowledge (Lenat 1995; Mueller 2000; Singh et al. 2002). The proposed thesis uses the latter approach as it gives broader semantic coverage—a feature necessary to the interpretation of domain-independent texts.

Cyc (Lenat 1995), ThoughtTreasure (Mueller 2000), and Open Mind Common Sense (Singh et al. 2002) are three approaches to large-scale common sense knowledge acquisition and reasoning. Cyc and ThoughtTreasure have logical representations and are more suitable for rigorous deep reasoning about situations, while Open Mind Common Sense and its ConceptNet (Liu & Singh 2004b) has a natural language representation, and thus excels at contextual reasoning over natural language texts (Liu & Singh 2004a). ConceptNet is semantic network of common sense facts, with built in methods for contextual expansion and analogy.

Examples of use in this thesis are as follows. The conceptual analogy faculty of ConceptNet is used to apply viewpoint models to predict reactions to unknown concepts in WWTT (Liu & Maes 2004) by situating the unknown fodder into the space of known concepts, also called conceptual alignment in the Cognitive Science literature (Goldstone & Rogosky, 2002). In the aesthetic viewpoint space, ConceptNet’s getContext() feature is used to brainstorm the rational entailments of a text, in order to generate the “shadows” that a fodder casts onto the “Think” axis. Finally, ConceptNet is a principle component of another core technology—textual affect sensing.

Textual affect sensing. Judgment is the behavioral and measurable expression of viewpoint, and the primary quality of judgment is affect. In fact, Ortony, Clore and Collins (1988) concisiated the definition of “emotion” to mean the expression of an affect about a person, thing, or event. Emotion and judgment thus can be represented basically as the bound pair (thing, affect). In some of the viewpoint systems to be presented in this thesis, affect manifests as choice implicature. For example, in the cultural identity space acquired through linguistic ethnography over social network profiles, individual choose to display certain items into their profile of “my favorite things,” and that choice can

be viewed as a judgment act (Austin 1962; Habermas 1981) which says that things listed in the profile are more pleasurable and arousing and dominated over than things not listed in the profile.

Other times though, affect must be inferred from unstructured natural language texts—for example, the machine should learn from the utterance “my mother is a loving and generous woman” that the speaker judges his mother positively. To complete this task, a topic spotter looks for the topics present in sentences, paragraphs, and documents, while a textual affect sensor appraises the affective qualities of each segment of text. Binding those two outputs to each other as (topic, affect) pairs, and using classical reinforcement learning (Kaelbling, Littman & Moore 1996) to generalize stable (topic, affect) pairs from training data, we have the beginnings of a model of a person’s system of attitudes/opinions.

To accomplish comprehensive textual affect sensing, I sense separately surface and deep affect. Surface, or rhetorical affect, can be measured as word-choice; I sense it by combining the Sentiment headwords of Roget’s Thesaurus (1911), a corpus of psychologically normalized affect words called ANEW (Bradley & Lang 1999), and an affective lexical inventory produced by Ortony, Clore and Foss (1987).

Deep affect is the pathos permeating from the contingent imagined consequences of an utterance and can be communicated without mood keywords at the surface. For example, the utterance “I was fired, my wife left me, and she took the kids and the house” uses no surface keywords to nonetheless convey a negative affect quite powerfully. Deep affect sensing is attempted using Emotus Ponens (Liu, Lieberman & Selker 2003), a textual affect sensor built using the Open Mind Common Sense corpus (Singh et al. 2002). The basic idea is when the affect of a concept is unknown, it can be approximated by the affect in its surrounding conceptual neighborhood. For example, supposing that the concept “get fired” is not annotated with affect, ConceptNet (Liu & Singh 2004b) has semantic links which connects “get fired” to other nodes which are annotated with affect such as “recession” (probable cause), “stupid person” (probable cause), “no money” (probable consequence), “hungry” (probable consequence). Thus the affect of “get fired” can be guessed by its context.

2.3 Viewpoint Artifacts and InteractionsDescribing viewpoint spaces and their organizational dynamics is one pillar of the present research. Acquiring the topology of spaces and the location of individuals from psychoanalytic readings of text is a second pillar. The third pillar then, is to draw from Interaction Design ( ) principles to construct interactive viewpoint artifacts and animate their reactions to fodder. These artifacts allow viewpoints to be explored and tinkered with, thus they hold great promise for a great variety of applications such as technological support

for self-reflection, perspectival tools for learning from others, interfaces for visualizing and searching human narrative content, psychographic visualizations for marketing and ethnography, and so on.

The proposed thesis will narrate several viewpoint artifacts already built, and then distill from those a set of core methodologies and considerations for embodying viewpoint. An inventory of research results obtained thus far is given below.

2.3.1 Major Examples

(Opinion Space) What Would They Think? (Liu & Maes 2004) is a system for modeling personal attitudes and the space of opinions at large using the Semantic Sheet representation shown in Fig. 1. A user can build a new “persona” by supplying an icon and pointing the system to some egocentric texts that are self-revealing and self-describing—i.e. position papers, instant messenging logs, emails, weblogs. The system reads and infers from the text a system of attitudes for that persona. Personae are embodied into virtual mentors (Fig. 3a) who continually observe the user’s browsing and writing activities, offering up just-in-time and just-in-context feedback to the user’s “fodder” through visual metaphors. To find out why a mentor reacted in a particular way, mentors can be double-clicked to pop up an explanation window—this window displays a list of quotes snipped from the mentor’s

“memory” of egocentric texts, rank-ordered by how well they justify the reaction that was given. For example, virtual mentor Roz Picard reacts negatively to the utterance “Robots will have consciousness” which is defended with quotes like “Several of my colleagues believe it’s just a matter of time and computational power before machines will attain consciousness, but I see no science nuggets which support such a belief.” Fig. 3b depicts the modeling of two cultures qua personae. In WWTT, cultures can be treated commensurately with individuals. The proposed thesis will pre-generate a fabric of cultural opinions to acquire the opinion space. Using this opinion fabric, individuals can be located as inhabitants of particular cultural opinions by applying simple alignment or “diff” techniques between cultures’ reactions and individuals’ reactions.

(Perceptual Aesthetic Space) The Aesthetiscope (Liu & Maes 2005b) is an art robot that renders color grid artwork a la Ellsworth Kelly and early Twentieth Century abstract impressionists (Figure 4). The manner and quality of the generated artwork is guided by a model of the user’s perceptual aesthetics. The perceptual aesthetic space (shown in Figure 1b) has the five dimensions of Think, Sense, Intuit, Feel, and Culturalize—these dimensions are based on Carl Jung’s fundamental modes of perception (1921). Though not yet implemented, the proposed thesis will automatically acquire the user’s aesthetic viewpoint through readings of egocentric text. Currently these dimensions must be specified manually. As a perspectival artifact, the Aesthetiscope reacts to “fodder” given to it, such as a word, a poem, or song lyrics. For example, it

Figure 4. Perspectival aesthetic rendition in the Aesthetiscope. The left column shows how the art robot renders the aesthetic impression of the words “sunset” (above) and “war” (below) through the eyes of a Realist (e.g. Sense=90%, Think=60%, Culturalize=40%, Feel=20%, Intuit=10%). The right column shows the same fodder rendered through the eyes of a Romantic (e.g. Sense=50%, Think=20%, Culturalize=70%, Feel=90%, Intuit=80%).

continuously observes what poetry the user is reading or what songs are queued in the playlist, dynamically changing the color grid artwork to “pair” with the fodder, just as wines are selected to pair with a cheese course. Another perspectival game that can be played is for two individuals both standing in front of the same artwork visualizing some poem to find their shared aesthetic (by averaging their locations), or to violated each other’s aesthetic (by allow one aesthetic viewpoint to corrupt another viewpoint). I am particularly interested on how deeply held aspects such as aesthetics can be exhibited or worn on one’s sleeve so to speak, like a piece of clothing avails identity and taste.

(Cultural Identity & Taste Space) Identity Mirror (Liu, Maes & Davenport 2005; Liu & Davenport 2005) is a mirror to support self-reflection that lets you “see who you are, not what you look like.” As shown in Fig. 5, the mirror’s computed reflection overlays a swarm of keyword descriptors over an abstracted image of the “performer.” The performer can use dance to negotiate his identity—for example, walking to and fro the mirror affects the granularity of the keywords being shown, which describe a far away performer using broad strokes like subculture keywords (e.g. fashionista, raver, intellectual, dog lover), but describe an up-close performer with descriptors like song names, books, food dishes, etc. When movement is slow and deliberate, the keywords more semantically distant from the performer’s ethos appear in the computed reflection, but those keywords are quickly dashed with sudden movements.

The Identity Mirror uses a social network profile to locate

the performer’s viewpoint within the cultural fabric of identity and taste. The cultural “taste fabric” (Liu, Maes & Davenport 2005) is derived by computing the latent semantic connectedness of “interest keywords” (music, books, sports, subcultures, etc) from analysis of the texts of 100,000 social network profiles. The performer’s location on the fabric is calculated by reading his social network profile, mapping that profile onto the nodes of the fabric, and using spreading activation (Collins & Loftus 1975) to define an ethos (a weighted collection of nodal activations). In the mirror artifact, the identity/taste viewpoint of the performer is visualized as a swarm of keywords. The viewpoint “reacts” to changes in the daily news stream. For example, around the time of the summer Olympics, the sports-centered news wire would bias the cultural fabric by highlighting nodes relating to the Olympic sports. The reflection in the mirror simulates the performer’s viewpoint by selectively interpreting the new cultural situation, and displaying just what exists at the intersection of the performer’s ethos and the news-du-jour’s ethos. Ambient

Figure 5. Self-reflexive performance with the identity mirror. A swarm of keywords shows a user’s situation within the cultural fabric of identity/taste, and with respect to the attentional biases of the zeitgeist as calculated by monitoring daily news streams. The user’s social network profile is used to locate the user within the cultural fabric.

Semantics, ( ) another system using the taste fabric, is an artifact that uses viewpoint to predict whether or not one individual would find another person to be sympathetic.

2.3.2 Minor Examples

(Gustatory Space) Synesthetic Recipes (Liu, Hockenberry & Selker 2005) is an interface for browsing for food recipes by imagined tastes of food. For example, typing “old, beautiful, desperate, urgent, alive, primal, homey, organic, nutritious, spicy, sweet, moist, aromatic, easy, zen" yields a recipe for "bohemian stew.” With food dishes, tastes, genres, and cultures arranged into a highly connected semantic network, the network approximates a space of taste-for-food. In Synesthetic Recipes, a viewpoint, called a “tastebud,” can be programmed into one of three avatars. As the user browses for food, the avatars constantly emote their likings and dislikings for suggested recipes. An individual’s tastebud can also be acquired through observational learning of what the user types into the search box. This is a minor viewpoint example and will be included in the thesis for completeness.

(Humor Space) Buffolo is a humor robot that suggests jokes it anticipates an individual will find funny. It does so by having crafted a model of a person’s sense-of-humor relative to the space of jokes. Using a Semantic Sheet representation like What Would They Think?, Buffolo reads an individual’s egocentric texts such as a weblog or email corpus with the goal of extracting (topic, pressure) pairs, much as WWTT extracted (topic, affect) pairs. Pressure is one particular dimension of a full affect measurement. Harkening to psychoanalysis’s hydraulic model of emotions (Freud 1901), an individual’s affective pressure points suggests psychic tensions which need catharsis—humor is a primary means to meet cathartic need (Freud 1905). A major way to structure humor is by culture, since much of one’s embarrassing and tense experiences growing up is shaped by cultural idiosyncrasy, e.g. Asian families and scholastic and work ethic emphasis; overbearing and verbose relatives in Jewish families, narratives of hustling, ghettos, players, and bling in Afro-American culture. Thus, Buffolo senses an individual’s cultural identifications and uses this as a humor viewpoint from which to predict the pleasure of a joke. Buffolo is also a minor viewpoint example and will be included in the thesis for completeness.

(Personality Space) Character Affect Dynamics Analysis (CADA) (Liu & Mueller, forthcoming) is a cognitive linguistics system which reads novels and infers the personalities of its characters into the Big Five personality inventory (John 1990) whose dimensions are agreeableness, neuroticism, openness, conscientiousness, extraversion. The system models the actions and interactions of characters as affective token passing. For example, the

sentence “Mary swindled Jack” is parsed into a Who-Dun-What representation, Mary and Jack are recognized as characters, and swindle is recognized as a displeasurable aggressive act. In CADA’s representation, the utterance equates to Mary sending Jack a negative attack token. If the narrative continues to show Jack negatively affected and submissive, then the system learns that Jack is vulnerable. Statistically over the long haul of a novel, stable personality characterizations can be made about each character. CADA demonstrates how personality qua viewpoint can be acquired in a sophisticated way from text, but it does not work well over too small a corpus of text, nor does it work well over most egocentric texts like weblogs. It is a minor system example that will nonetheless illuminate the idea of “psychoanalytic readings of text.”

2.4 EvaluationTo successfully defend the computational theory of point-of-view to be presented in the proposed thesis, I propose three lines of evaluation—literature evaluation, model validation, and task-based evaluation.

Literature evaluation. When presenting a new theory on a subject as basic as point-of-view, a primary task is to properly situate the theory within all of its proper literatures. I believe it is fair to call this activity “evaluation” even though it is not quantitative. Literature evaluation means to scrutinize the implications of the computed viewpoint models to the viewpoint theories presented in the literature, not only stuffing the literature narrowly into “related work,” but sustaining dialog with the literature all the way through the thesis, constantly check-pointing how this thesis’s account and the literature’s accounts mirror and inform each other. The literature on point-of-view is so extensive that even if my theory could provoke any new thinking in the existing frameworks on point-of-view, it should be considered a huge success. Because this is a computational thesis, there is in particular a tremendous opportunity to reify humanistic theories into computational structures and processes.

Model validation. Among the major viewpoint systems to be discussed in the thesis, space models are computed for perceptual aesthetic space, for opinion space, and for cultural identity/taste space. Location models are computed from individuals’ egocentric texts. I propose to evaluate how well both the space models and location models present accurate pictures of viewpoint spaces and individual viewpoints. I will collect human ratings of the computer-generated models as baselines for computing the accuracy of model acquisition. I will also evaluate the quality of reactions produced through simulated viewpoints by employing human judges. A few of these evaluations are already obtained, including a human-rated evaluation of the quality of attitude prediction in What Would They Think?

Task-based evaluation. The usefulness of viewpoint artifacts speaks to the significance and potential impact of computing viewpoint. User studies will illuminate how well viewpoint artifacts can support a diverse set of tasks such as self-reflection, taste-based recommendation, learning about others, decision support, artistic portrayal, and others. Some of these evaluations are already obtained, including: a taste-based recommendation task using the viewpoint space of cultural identity & taste; a learning about others task using What Would They Think?; an artistic portrayal task in the Aesthetiscope.

3 ContributionThe proposed thesis aspires to be the first comprehensive and computed theory of point-of-view. The theory will be well supported by built viewpoint models for several domains such as aesthetics, cultural identity, and opinions, implementations of automated viewpoint acquisition from readings of text, and implementations of several interactive viewpoint artifacts, which demonstrate broad and significant implications for this line of research. Specifically, I hope to show that

The slippery notion of point-of-view, well covered in the humanities literatures, can be represented, captured, and reified into computational artifacts.

A point-of-view can be computed most elegantly as an individual’s collective situations [sic] within latent semantic spaces of viewpoint such as OpinionSpace, PerceptualAestheticSpace, and CulturalIdentitySpace.

An individual’s point-of-view and the topology of viewpoint spaces can be acquired automatically through what I call psychoanalytic machine reading.

Interactive viewpoint artifacts that simulate an individual’s judgments and react to an individual’s actions just-in-time and just-in-context can afford powerful new tools for learning about others, for self-reflection, for inspiration, and for deeper user modeling.

4 Background and Related WorkThe proposed thesis research is articulated against several back-grounds. Given that I proposed ‘literature evaluation’ as an important contribution of this thesis, the following section will be extensive. In the following, I revolve discussion around several basic topics treated in this thesis, present both computational and non-computational work.

4.1 Psychoanalytic reading for viewpointI suggest that an individual’s point-of-view within a particular realm—inter alia, aesthetic, cultural identity, or opinion—can be located by psychoanalytic readings of egocentric text. This feat builds on work in the text

interpretation division of literary theory called hermeneutics, and on story understanding work in Artificial Intelligence.

Non-computational workWhile a preponderance of modernist theorists still cling to the view that textual communication is rational, is objective, and that meaning can be deciphered from text through logical disambiguation, post-Enlightenment thinkers and many computational semantics practitioners believe that meaning is just a collective hallucination—it is not present in text objectively, only inter-subjectively. Friedrich Schleiermacher (1809)—founder of the modern science of hermeneutics—posed textual interpretation as a Romantic enterprise. He believed that, more than objective communication, each text avails of its author. To unravel the author from the text, it is necessary to read deeply, holding in mind the various layers of context which underlie each authorial choice, and using each context like a colored lens to illuminate a single dimension of the text’s kaleidoscope of meanings. In the last century, hermeneutic theorists formalized the notion of reading-through-context in Speech Acts theory. John Austin’s formulation of the theory (1962) distinguished between and utterance and subtext in an act of speech. When your boss says “there’s no room in this company for mediocrity,” the surface utterance, or locution, seems to state a moral truth that the company is not mediocre, but the subtext, or illocutionary force, of the utterance can mean to threaten. Jurgen Habermas has broadened Speech Acts Theory into a Theory of Communicative Action (1981), which re-views all social and cultural interactions as speech acts with locutionary and illocutionary components. In my research, I invoke Speech Acts Theory by attempting to unravel illocutions that emanate from the homunculus of author’s viewpoint—for example, someone telling you their favorite things on a social network profile can help you (or a machine) to infer their taste: an individual listing “Marilyn Manson’s Antichrist Superstar” as a favorite album (Manson is a goth rock musician) and “Salinger’s Catcher in the Rye” as a favorite book constitutes two factual utterances, but illocutionary force underlying the two utterances seems contemptuous of social and ethical norms, and so the illocutionary act here, is potentially one of rebellion.

Since the last century, reading is posed as multiple because each reader posturing results in a different meaning gleaned. In De l’interpretation, Paul Ricoeur (1965) distinguished between a hermeneutics of ‘retrieval’ versus a hermeneutics of ‘suspicion’, practiced chiefly by Nietzsche and Freud. Nietzsche grandfathered existentialism because he dissociated from socially condoned ‘readings’ of life’s meaning and instead ‘read’ life and the world through a social biologist’s lens—for example he makes a famous argument in Beyond Good and Evil (1886) that evil is actually good when viewing the world as a

body because it undermines the false power of society, returning human kind to the homeostasis of entropy, that universal end. Freud (1901) was fascinated with the unconscious mind as illocutionary force, and his methodology of psychoanalytic reading to infer the neuroses of his patients is invoked in this thesis. Resembling Ricoeur’s contribution, reading psychologist Louise Rosenblatt (1978) distinguished between ‘efferent’ and ‘aesthetic’ reading. ‘Efferent’ like Ricoeur’s ‘retrieval’ reading means objective reading—reading with the modus operandi to take away something from the text. ‘Aesthetic’ reading means to allow the reader to live through the text—this is the evocative mode of reading which the Aesthetiscope uses to liberate from the text, meanings which come out of the reader, not out of the author. Rosenblatt’s ‘aesthetic’ and Ricoeur’s ‘suspicious’ readings are superficially oppositional but they are actually doppelganger ideas, as will become clear when deep reading is computationalized in this thesis.

Computational workClassical AI works in Story Understanding are Terry Winograd’s SHRDLU blocks world understander (1971), Eugene Charniak’s children’s story understander (1972), Mike Dyer’s BORIS goal-oriented understander (1983) which manifests Schank and Abelson’s scripts, goals, and plans construction (1977), and Wendy Lehnert’s plot units strategy (1982) for viewing the macroscopic semantic structure of narratives. These classical systems and theories were overly symbolic, logical, and brittle, treated understanding as logical theorem proving, and thus failed to work over a broad range of natural texts.

Reading is such a high level description of a complex cognitive machinery surrounding the interpretation of text that only recently has Artificial Intelligence researchers dared to term their work “reading.” Moorman and Ram (1994) present a sophisticated model of machine reading where their reader robot ISAAC can focus and attend, can willfully suspend disbelief, and can use analogy to creatively force knowledge into a current understanding framework. Moorman and Ram lament that ISAAC does not have enough background knowledge to perform baseless analogy, a problem that is addressed in this thesis using ConceptNet as a ‘base’ for analogy. Ram’s system, AQUA (Ram 1994), introduces a computational workflow for interleaving reading with understanding—in AQUA, reading with some understanding framework produces anomalies which prompt questions which help to direct the explanation and understanding process. ISAAC and AQUA both have the idea that reading is an activity which constructs, populates, and occasionally revises a situation model (Zwaan & Radavansky, 1998), a construct meant to demonstrate unification of comprehension. Reading using situation models should be considered ‘retrieval reading’ (Ricoeur), ‘efferent reading’ (Rosenblatt), and ‘objective’. This thesis

poses its computed readings as ‘aesthetic’, ‘suspicious’ and overarchingly ‘psychoanalytic’. One interesting example of metaphorical reading that is closer to psychoanalytic reading is Srinivas Narayanan’s KARMA system (1997). Using multiple, synchronized Petri-nets, KARMA ‘reads’ text by simulating state trajectories in Petri-nets—for example, the utterance “Japan’s economy stumbled” reifies in KARMA’s walking and stumbling machinery, allowing the inference that Japan’s economy is off-balance and the situation may not correct immediately but will correct eventually. Along these same lines, Erik Mueller’s ThoughtTreasure system (2000) reads-by-visualization—creating a 2D ASCII-art rendition of a read passage, which is a lucid representation which can be used to answer many questions. In my view, psychoanalytic reading is also trying to read-by-visualization because once the author can be envisaged as occupying a location in some viewpoint space, the representation is lucid enough to be able to answer many questions about the author and allow authorial reactions to be predicted.

Other relevant technologies and literature that supports psychoanalytic readings at the mechanistic level are Natural Language Processing, Common Sense Reasoning, and Textual Affect Sensing. The related works for these have already been given in a previous section.

4.2 Viewpoint spacesI pose an individual’s viewpoint as her situation within latent semantic spaces that serve as realms of viewpoint. The idea that an individual can only be understood by understanding the whole culture or bundle of potentialities that surrounds her is a sociological understanding. Relevant non-computational work includes theories of culture’s structure and language-qua-culture’s structure; computational work includes the modeling of latent semantic spaces.

Non-computational workOf the three major viewpoint spaces considered in the present thesis work, perceptual aesthetic space is a formal semiotic space because I was inspired by Carl Jung’s dimensional model of perception (1921)—though that model has been verified sociologically and psychologically as it is the basis of the widely used Myers-Briggs Type Indicator of temperament (Briggs & Myers 1976). As for cultural identity space and opinion space, these are ethnographically acquired by analyzing large corpora of cultures. Here, we invoke ‘culture’ to mean the collective symbolic creative product of humanity, and not to mean a mode of superior intellect or taste. Our interpretation is in line with the word ‘Kultur’ as invoked by Wittgenstein and Nietzsche, and is in line with Clifford Geertz’s interpretation. In The Interpretation of Cultures (1973), Clifford Geertz motivated the significance of culture to the self thusly, “man is an animal suspended in webs of significance he himself has

spun, I take culture to be those webs” (Geertz, 1973: 4-5). “Webs of significance” is the inspiration for the semantic fabric representation of the cultural identity space—a fabric is a super densely connected semantic web, and ‘significance’ is embodied in the 12,000 nodes of cultural symbols such as book authors, book titles, musical genres, etc. By exposing the vastness of tastes captured by the cultural identity space, I illustrate Grant McCracken’s (1997) point that our contemporary consumer-driven world is in late capitalism, where all tastes and identities that can be imagined are fulfilled by consumables—this is Plato’s prediction of Plenitude. Whereas Geertz conceived of culture around the interconnectedness of symbols, Roland Barthes’s conceptualization of culture concerns the valences of symbols (1964). Barthes’s approach, which he calls Semiology, views culture as a system of signification. That is to say, words and objects are signifiers, but culture supplies a way to map signifiers into signifieds, or underlying meaning. For example, in Western cultures, ‘rich’ is a signifier that maps into a positive affect, or privilege, as Structuralists calls it. The Semantic Sheet representation of viewpoint spaces such as opinion follows Barthes’s interpretation of culture, i.e. the space of opinion is a collection of pairs of (topic qua signifier, affect qua signified).

Computational workThe Internet contains many resources such as weblog communities, social networks, recipe corpora, humor corpora, political corpora – these resources are reflections of the cultures of the offline everyday world—thus mining these resources can provide us with working models of viewpoint spaces. The topology of these latent semantic spaces can be inferred through statistical modeling techniques such as Latent Semantic Analysis (Deerwester et al. 1990), Support Vector Machines (Joachims 1998), Multi-Dimensional Scaling (Kruskal & Wish 1978), and the mathematical method of Principle Components Analysis. Explaining culture has been the providence of ethnographers (Kluckhohn 1949), but the process is nearly identical to linguistic modeling as culture is a language with symbols, meaning, syntax, sentences, and discourse—hence in this thesis work, I term language modeling of cultures, ‘linguistic ethnography’. My idea of linguistic ethnography is close to a movement in the Semantic Web community called ‘emergent semantics’ (Aberer et al. 2004) which advocates the countervailing view that semantic ontology should be shaped from the ground-up, a posteriori, and in accordance with the natural tendencies of the unstructured data—such a resource is often called a folksonomy when built by humans (e.g. dmoz.org, allrecipes.com).

4.3 Point-of-view as ‘locations’ in space

Knowing the topology and constitution of viewpoint spaces, I pose an individual’s point-of-view as locations and situations within this space.

Non-computational workThe idea that a self is just a particular emanation of the social and cultural milieu is embraced by situational theorists—in their discourse, viewpoint is called perspective and subject-position. Experiential situationists believe that a self is formed out of its prior experiences in the world, and these ideas originate in David Hume’s (1748) empiricism. Memory-based reasoning (Stanfill & Waltz 1986) and case-based reasoning (Riesbeck & Schank 1989) and reinforcement learning (Kaelbling, Littman & Moore 1996) in Artificial Intelligence are examples of experiential situationalism. Social situationists believe that a self is a ‘socially and culturally mediated construction.’ Although situationalism can be traced back to the Sixth Century in the Indian linguistic tradition, the recent episode of the movement finds ground in Jacques Lacan’s notion that ‘the ego is formed out of the other’ (1957) (‘other’ meaning environment in Lacan’s discourse), although Nietzsche’s social biology also implied social situationalism, even existential situationalism. Georg Simmel (Levine 1971) presaged the field of sociology by posing an individual as only knowable through fragmented reflections against his milieu such as his job, his church membership, his social status, etc. In more recent work, Mihaly Csikszentmihalyi and Eugene Rochberg-Halton (1981) studied how significant objects in a family’s domestic setting constitutes a ‘symbolic environment’, which echoes and reinforces each individual’s identity. Narrative psychologists like Kevin Murray (1990) suggest that cultural narratives like romantic and comedic stories serve as materials out of which an individual constructs an identity. Sarah Thornton’s study (1996) of underground club culture reveals how hipsters politicize their locations within the music ‘scene’ because their location is something to be signaled to others, and something capable of winning them social capital. Frederic Jameson (1998) poses situationalism most bleakly by equating language to a ‘prison-house’ and lamenting that cultural space is so fractured that individuals are increasingly bucketed into their idiolects and experience Marx’s alienation.

Situationalism coming from the sociological literature seems sometimes to obscure the agency and creativity of individuals. In the philosophical literature, it is shown that being situated in cultural milieu is not a helpless act resembling K’s fate in Franz Kafka’s The Trial (1925). Jean-Francois Lyotard (1984) proclaims that the end of hegemonic ‘meta-narratives’ means that individuals can shape their own viewpoint by selecting mini-narratives to cloth themselves in. Jacques Derrida and Claude Levi-Strauss (Derrida 1966) conceive of the individual as a bricoleur who makes cultural bricolage by opportunistically

choosing what ideas and positions to import into their viewpoint space. In other words, much of postmodern philosophy is concerned with prognosticating that individuals either will (Lyotard, Derrida) or will not (Jameson) be able to control how the space in which their perspective lives is constructed. Recent ethnographic examples of empowered individuals appropriating their own situational spaces include Certeau (1997) and Grodin and Lindlof (1996). My viewpoint models do not reflect the full range of dynamicity and agility that individuals are really capable of, though in describing an individual’s location as multiple and complex rather than singular and categorical, and in predicting reactions creatively through analogy, I believe this thesis will support postmodern optimism.

Computational workThe computation of individuals as locations within latent semantic spaces is connected to the user modeling literature. The most directly applicable work on situated viewpoint is, found in a non-computational but quite computable psychological theory by H. Montgomery. In “Towards a perspective theory of decision making and judgment” (1994), Montgomery writes “Three determinants of perspectives in thinking are identified: (a) the subject, i.e., subject orientation, (b) the object, and (c) psychological distance between subject and object” (Montgomery 1994: abstr.). Likewise, this thesis simulates viewpoints by computing the relationship between individual’s location and the fodder.

The user modeling literature computes and attempts to predict user actions and likings. It should be noted that this thesis models individuals whereas the notion of user implies that a user model is inherently only a narrow description of an individual behaving within the context of a narrow application domain. The user modeling literature knows two prevailing paradigms for representing users – firstly using frames to model intrinsic attributes, as in demographics and psychographics, and secondly using vector-based statistics to model extrinsic behavior, as used in collaborative filtering (Shardanand & Maes 1995). However, reducing a person to but a few categorical attributes lacks specificity, while most behavior modeling is too domain or task-specific and the learned features do not rise the generality of describing an individual as he exists outside applications. Point-of-view models developed in this thesis work are semiotic like categorical user models, but can be reasoned with robustly like statistical vector-based user models. Viewpoint modeling further distances itself from typical behavior modeling since the work is concerned with characterizing an individual out of any application’s context.

4.4 Simulating judgmentViewpoint models can simulate judgments of the individual being modeled. There is substantial background literature

on how people simulate the judgments of others, and how machines can cognitively model reactions.

Non-computational workThe ability of individuals to mentally model the thoughts and behaviors of other individuals is an evolutionary adaptation often called mindreading. Two core components of mindreading recently popular in the Cognitive Science literature are intentionality—the ability to infer a speaker’s reference through gaze and other cues—and theory of mind—the name of the mental faculty for modeling other minds. Opinions on how humans implement theory-of-mind are split between the Theory-Theory camp believing that predictive rules are hard-coded into minds, and the Simulation-Theory camp (Gallese & Goldman 1998) believing that other minds can be simulated on the self like how the same Java code can be run on different computers. The Simulation-Theory camp was boosted by a recent finding in neurobiology that lower primates have mirror-neurons—neurons which fire both when an action like grasping a banana is taken, and when the action is seen to be taken by a conspecific (Gallese & Goldman 1998). In Minsky’s Society of Mind Theory (1988), he introduces the abstraction of mental critics to explain how minds access the expertise and wisdom of other minds. Minsky seems to ally with the Simulation-Theory camp, and even goes further to theorize that all of our actions are proactively simulated by mental critics in the cognitive background, who censor our present stream-of-consciousness with advice; this is not unlike what my What Would They Think? system tries to achieve.

Daniel Dennett poses mindreading in a manner sympathetic to viewpoint-based simulation. In The Intentional Stance (1987), Dennett suggests that individuals can ‘read’ the same situation differently through different lenses of interpretation (=viewpoint-spaces?), which he calls ‘stances’. For example, a robbery witnessed through the ‘physical stance’ yields physical perceptions like a convenience store opening and a man-object storming out. Witnessed through the ‘design stance’, telic and agentic aspects are illuminated and it is seen that robbers are designed to rob stores, which are designed to carry money. Finally, witnessed through the ‘intentional stance’, it is noticed that the robber is a willful and rational person who robbed this store out of some motivation or habit, and that he is running because he is fleeing from the scene of the crime. Parallel to this are Clifford Geertz’s notions of ‘thin description’ and ‘thick description’ (1973). Using Gilbert Ryle’s example of a narrative that recounts a person winking, Geertz remarks that a ‘thin description’ renders the wink as not differentiable from an unmotivated twitch in the body, but ‘thick description’, importing context regarding cultural motivations for winks into the narrative, allows a wink’s cultural meaning to be seen. ‘Thick description’, then, constitutes something like a ‘cultural stance’ in Dennett’s framework, and ‘thin description’—dissociated

from human intention and meaning—maps to the ‘physical stance’. Stance theory—including Ron Edward’s Actor Stance model and Kevin Hardwick’s Narrative Stance model—is also well developed as a practical methodology in the Acting discipline, where the lifeblood of actors is the successful and convincing rendition of characters. Improvisatory actors must in particular possess something resembling point-of-view models of the characters whose shoes they must perform spontaneously in.

Computational workDennett’s intentional stance treats persons as agents who follow rational principles, thus suggesting a way to simulate what an individual might think, want or do next given a prior model. This computation is called the Belief-Desire-Intention model (Georgeff 1998) of agency in the Intelligent Agents literature, and simulation of actions is called ‘action selection’ (Maes, 1994). Similarly, Pollack (1992) advocates the use of plans as packets of rational actions. A plan is guided by two principles—filtering the space of possible actions to eliminate those which conflict with goals, and constructing a plan to satisfy the greatest number of goals, called overloading. Notable examples of rational simulation systems include Allen Newell and Herbert Simon’s (1963) General Problem Solver, and Newell’s SOAR (1990) cognitive architecture. When some prior symbolic knowledge in the form of expert rules, knowledge captures of prior states, goals, or scripts are infused, rational simulation is called Case-based Reasoning (Riesbeck & Schank 1989) or Memory-based Reasoning (Stanfill & Waltz 1986).

Outside of rational simulation are baby machines, behavior-based models (closely related to behavior-based user modeling), cognitive-affective architectures, and knowledge-based models. Gary Drescher’s baby machine (1991) tried to reach the holy grail of unifying behaviorism and symbolicism by learning abstractions out of behavior through marginal attribution. Rodney Brooks (1991) advocated a ‘reactive’ approach to intelligence, which relies completely on reinforcement learning of hidden Markov models to drive action. Marvin Minsky (forthcoming), Push Singh (2005), and Aaron Sloman (1981) have proposed three-tier cognitive-affective heterogeneous architectures for simulating minds. Cyc (Lenat 1995), ThoughtTreasure (Mueller 2000) and ConceptNet (Liu & Singh 2004b) represent a large-scale knowledge approach to simulating thought, guided by a topology of thinkables. This thesis work practices a knowledge-based and associative approach to simulation. Knowledge is not acquired formally but uses statistical and reinforcement learning to model the ‘knowledge’ of viewpoint spaces. Viewpoints are simulated through associative approaches like spreading-activation (Collins & Loftus 1975) and knowledge-based approaches like analogical reasoning (Gentner 1983; Fauconnier & Turner 2002).

4.5 Interactive viewpoint artifactsViewpoint models captured into interactive artifacts affords a tools for self-reflection, learning about others, and inspiration. The design of interactive artifacts is informed by knowledge of Interaction Design and Human-Computer Interaction, and builds upon previous computational work in Interaction and Software agents, and Responsive Environments.

Non-computational workBorn from the Bauhaus and Ulm Schools, Interaction Design theorizes the psychological, cognitive, and social ramifications of design. The design of interactive viewpoint artifacts must be informed by the various politics of what makes an artifact useable, useful, and inspiring. A unique design challenge for this work is that these artifacts rest somewhere between the categories of “humanistic agent” and “tool.” The human metaphor must be sustained since viewpoint is primary a human competency and is only fully appreciable as such. But the tool metaphor prescribes that the boundaries and capabilities of the interface are made clear and transparent to ensure that the artifacts are useable in a predictable manner. Byron Reeves and Clifford Nass (1996) advocate that computer-human interaction be illuminated by understanding human-human communicative contracts. Don Norman (1989) explains the importance of a tool’s aesthetics and emotional evocations to its usability. Because interacting with someone’s or the self’s viewpoint is by its nature so provocative and engaging, it allows ample opportunity for constructivist ‘tinkering’ (Papert & Harel 1991), ludic or playful activity (Gaver 2001) (), and allows users to explore unusual values or avenues which Anthony Dunne (1999) calls ‘value fictions’.

Computational work“Software agents” (Maes 1994) are computed embodiments of stereotyped human capabilities, and Pattie Maes explored how they could interactively support human choices such as music selection or browsing the Web, and augment human intelligence (I.A., not A.I.). Bradley Rhodes (Rhodes & Maes 2000) and Henry Lieberman (1997) describe interaction agents could observe user actions such as typing or browsing, and serendipitously and proactively give advice or suggestions. Another line of computational work in Responsive and Reflective Environments (Krueger 1983) investigates how interfaces such as Identity Mirror can engage an individual to ‘perform’ self-reflection.

5 TimelineI plan to finish refactoring technical implementations and complete all evaluations by the end of January 2006, and the thesis and defence by the end of the spring term 2006.

6 ResourcesNo additional resources are required, beyond typical access to Media Lab resources, and opportunities to travel to meet with non-local readers.

ReferencesAberer, K. et al.: 2004, Emergent semantics. Proc. of 9th International

Conference on Database Systems for Advanced Applications (DASFAA 2004), LNCS 2973, 25-38 Heidelberg.

Austin, J. L.: 1962, How to Do Things with Words. Cambridge (MA): Harvard UP.

Bakhtin, M.: 1935, The Dialogic Imagination, Austin, TX: University of Texas Press.

Barthes, R.: 1964/1967, Elements of Semiology. (Translated by Annette Lavers & Colin Smith). London: Jonathan Cape.

Bradley, M.M., & P.J. Lang: 1999, Affective norms for English words (ANEW): Instruction manual and affective ratings. Technical Report C-1, The Center for Research in Psychophysiology, University of Florida.

Briggs, K. C., & I. B. Myers: 1976, Myers-Briggs Type Indicator: Form F, Palo Alto: Consulting Psychologists Press

Brooks, R.: 1991, Intelligence Without Representation, Artificial Intelligence Journal (47): 139-159

Certeau, M. de: 1997, Culture in the Plural. Ed. and intro. Luce Giard. Trans. and afterword Tom Conley. Minneapolis: U of Minnesota Press.

Charniak, E.: 1972, Toward a Model of Children Story Comprehension. MIT School of EECS PhD Thesis.

Collins, A. M., and E. F. Loftus: 1975, A spreading-activation theory of semantic processing, Psychological Review 82: 407-428.

Csikszentmihalyi, M., E. Rochberg-Halton: 1981, The Meaning of Things: Domestic Symbols and the Self, Cambridge University Press.

Deerwester, S., S. T. Dumais, G. W. Furnas, T. K. Landauer, R. Harshman: 1990, Indexing by Latent Semantic Analysis, Journal of the Society for Information Science, 41(6), 391-407.

Deleuze, G. and F. Guattari: 1987, A Thousand Plateaus: Capitalism and Schizophrenia, Brian Massumi (transl.), Minneapolis: The University of Minnesota Press.

Dennett, D.: 1987, The Intentional Stance, Bradford Books.Derrida, J.: 1966/1978, Structure, Sign, and Play in the Discourse of the

Human Sciences". In: Writing and Difference, trans. Alan Bass: 278-294. London: Routledge.

Doyle, J.: 1980/1987, A truth maintenance system, in Reading in nonmonotonic reasoning: 259-279, Morgan Kaufmann.

Drescher, G.: 1991, Made-Up Minds: A Constructivist Approach to Artificial Intelligence. MIT Press.

Dunne, A.: 1999, Hertzian tales: Electronic products, aesthetic experience and critical design. London, RCACRD Research Publications.

Dyer, M.G.: 1983, In-depth understanding. Cambridge, Mass.: MIT Press.Fauconnier, G. and M. Turner: 2002, The Way We Think: Conceptual

Blending and the Mind’s Hidden Complexities, Basic Books.Freud, S.: 1901, The psychopathology of everyday life.Freud, S.: 1905/1963, Jokes and their relation to the unconscious, WW Norton.

Gallese, V. and A. Goldman: 1998, Mirror neurons and the simulation theory of mind-reading, Trends in Cognitive Sciences 2(12): 493-501Gärdenfors, P.: 2000, Conceptual Spaces, MIT Press.Gärdenfors, P. and K. Holmqvist: 1994, Concept formation in dimensional

spaces, Lund University Cognitive Studies 26.

Gaver, W.: 2001, Designing for Ludic Aspects of Everyday Life, ERCIM News No. 47.

Geertz, C.: 1973, The interpretation of cultures. New York: Basic.Georgeff, M.P. et al.: 1998, The Belief-Desire-Intention Model of Agency. In

N. Jenning, J. Muller, and M. Wooldridge (eds.), Intelligent Agents V. Springer.

Gentner, D.: 1983, Structure-mapping: A theoretical framework for analogy, Cognitive Science 7: 155-170.

Goffman, E.: 1959, The Presentation of Self in Everyday Life. Garden City, NY: Doubleday.

Goldstone, R. L. and B.J. Rogosky: 2002, Using relations within conceptual systems to translate across conceptual systems, Cognition 84: 295–320

Grodin, D. and T. Lindlof (eds.): 1996, Constructing the Self in a Mediated World, Thousand Oaks, CA: Sage.

Habermas, J.: 1981/1985, The Theory of Communicative Action, Volume 1, Beacon Press.

Hegel, G. W. F.: 1807, Phänomenologie des Geistes. Hume, D.: 1748, An Enquiry Concerning Human Understanding.Jameson, F.: 1998, Postmodernism and Consumer Society, in: The Cultural

Turn: Selected Writings on the Postmodern 1983-1998. Verso.Joachims, T.: 1998, Text Categorization with Support Vector Machines:

Learning with Many Relevant Features, Proceedings of ECML-98, 10th European Conference on Machine Learning, edited by Claire Nédellec and Céline Rouveirol: 137-142, Springer-Verlag.

Johannesson, M.: 1996, Obtaining Psychological spaces with MDS - a pilot study with perceptual stimuli, Lund University Cognitive Studies 45.

John, O. P.: 1990, The “Big Five” factor taxonomy: Dimensions of personality in the natural language and questionnaires, Handbook of personality: Theory and research, Guilford Press.

Jung, C. G.: 1921/1971, Psychological Types, trans. by H. G. Baynes, Princeton, NJ: Princeton University Press.

Kaelbling, L.P., L.M. Littman and A.W. Moore: 1996, Reinforcement learning: a survey, Journal of Artificial Intelligence Research, 4:237—285

Kafka, F.: 1925/1999, The Trial, Schocken.Kluckhohn, C.: 1949, Mirror for Man, McGraw-Hill Book Co.Krueger, M.W.: 1983, Artificial Reality, Addison-Wesley. Kruskal, J. B., and M. Wish, 1978, Multidimensional Scaling, Sage University

Paper series on Quantitative Application in the Social Sciences, 07-011. Beverly Hills and London: Sage Publications.

Lacan, J.: 1957, The agency of the letter in the unconscious, or reason since Freud, La Psychoanalyse 3: 47-81.

Lakoff, G. and M. Johnson: 1980, Metaphors We Live by, University of Chicago Press.

Lehnert, W.G.: 1982, Plot units: A narrative summarization strategy. In W. G. Lehnert & M. H. Ringle (Eds.). Strategies for natural language processing. Hillsdale, NJ: Lawrence Erlbaum Associates.

Lenat, D.: 1995, Cyc: a large-scale investment in knowledge infrastructure. Communications of the ACM 38(11). ACM Press.

Levenshtein, V.: 1965/1966, Binary codes capable of correcting deletions, insertions, and reversals, Doklady Akademii Nauk SSSR, 163(4):845-848, 1965 (Russian). English translation in Soviet Physics Doklady, 10(8):707-710.

Levine, D.N. (ed.): 1971, On Individuality and Social Forms: Selected Writings of Georg Simmel, University of Chicago Press.

Lieberman, H.: 1997, Autonomous Interface Agents, Proceedings of CHI 1997.

Liu, H.: 2002, MontyLingua: Commonsense-Enriched NLP, Toolkit and API. Accessed at: http://web.media.mit.edu/hugo/montylingua/

Liu, H.: 2003, Unpacking meaning from words: a context-centered approach to computational lexicon design, Blackburn et al. (Eds.): Modeling and Using Context, LNCS 2680: 218-232, Springer.

Liu, H., G. Davenport: 2005, Self-reflexive performance: dancing with the computed audience of culture , International Journal of Performance Arts and Digital Media 1(3), Intellect Ltd.

Liu, H., M. Hockenberry & T. Selker: 2005, Synesthetic Recipes: foraging for food with the family, in taste-space, Proceedings of the 32nd Annual Conference on Computer Graphics and Interactive Techniques, Los Angeles, CA.

Liu, H., H. Lieberman and T. Selker: 2003, A model of textual affect sensing using real-world knowledge, Proceedings of the 7th International Conference on Intelligent User Interfaces, IUI 2003, 125-132, ACM Press.

Liu, H. and P. Maes: 2004, What Would They Think? A Computational Model of Attitudes. Proc. of the 2004 ACM Conference on Intelligent User Interfaces, 38-45. ACM Press.

Liu, H. and P. Maes: 2005a, InterestMap: harvesting social network profiles for recommendations, Proceedings of ACM Beyond Personalization 2005: A Workshop on the Next Stage of Recommender Systems Research, 54-59, ACM Press.

Liu, H. and P. Maes: 2005b, The Aesthetiscope: Visualizing Aesthetic Readings of Text in Color Space. Proceedings of FLAIRS2005: 74-79, AAAI Press.

Liu, H., P. Maes, G. Davenport: 2006, Unraveling the taste fabric of social networks, International Journal on Semantic Web and Information Systems 2(1), Hershey, PA: Idea Academic Publishers.

Liu, H. and E. T. Mueller: forthcoming, Character Affect Dynamics Analysis.Liu, H. and P. Singh: 2004a, Commonsense reasoning in and over natural

language, M Negoita, RJ Howlett, LC Jain (Eds.): Knowledge-Based Intelligent Information and Engineering Systems, LNCS 3215:293-306, Springer.

Liu, H. and P. Singh: 2004b, ConceptNet: A Practical Commonsense Reasoning Toolkit. BT Technology Journal 22(4), 211-226. Kluwer Academic Publishers.

Lyotard, J-F: 1984, The Postmodern Condition: A Report on Knowledge, Minneapolis: University of Minnesota Press.

McCracken, G.: 1997, Plenitude. Toronto: Periph: Fluide.Maes, P.: 1994, Modeling Adaptive Autonomous Agents, Artificial Life

Journal, C. Langton, ed., Vol. 1, No. 1 & 2, MIT PressMiller, G. A., R. Beckwith, C. Fellbaum, D. Gross, & K. Miller: 1990, Five

papers on WordNet, CSL Report 43, Cognitive Science Laboratory, Princeton University.

Minsky, M.: 1988, Society of Mind, Simon and Schuster.

Minsky, M.: 1990, Logical vs. Analogical or Symbolic vs. Connectionist or Neat vs. Scruffy, Artificial Intelligence at MIT., Expanding Frontiers, Patrick H. Winston (Ed.), Vol 1, MIT Press.

Minsky, M.: 1992, Future of AI Technology, Toshiba Review 47(7).Minsky, M.: forthcoming, The Emotion Machine, Pantheon.Montgomery, H.: 1994, Towards a perspective theory of decision making

and judgment, Acta Psychol (Amst) 87(2-3):155-78.Moorman, K. and A. Ram: 1994, Integrating creativity and reading: A

functional approach. In Sixteenth Annual Conference of the Cognitive Science Society.

Mueller, E. T.: 2000, ThoughtTreasure: A natural language/commonsense platform. Accessed on 11 November 2005 from http://www.signiform.com/tt/

Murray, K.: 1990, Life as fiction, Ph.D. Dissertation, Department of Psychology, University of Melbourne

Narayanan, S.S.: 1997, Knowledge-based action representations for metaphor and aspect (KARMA) (Unpublished doctoral dissertation). University of California, Berkeley

Newell, A.: 1990, Unified Theories of Cognition, Cambridge, MA: Harvard University Press.

Newell, A. and H. A. Simon: 1963, GPS, a program that simulates human thought. In E. A. Feigenbaum and J. Feldman (eds.), Computers and Thought: 279-293. New York: McGraw-Hill.

Nietzsche, F.: 1886, Jenseits von Gut und Böse (Beyond Good and Evil).Norman, D.: 1989, The Design of Everyday Things, Currency Doubleday.Ortony, A., G.L. Clore, M.A. Foss: 1987, The Referential Structure of the

Affective Lexicon, Cognitive Science 11(3): 341-364.Ortony, A., G. L. Clore, A. Collins: 1988, Cognitive Structure of Emotions,

Cambridge University Press.Papert, S., and I. Harel: 1991, Situating constructionism, in Constructionism,

Ablex Publishing Corp.Pollack, M.: 1992, The uses of plans, AI Journal:57

Ram, A.: 1994, AQUA: Questions that drive the explanation process. In Roger C. Schank, Alex Kass, & Christopher K. Riesbeck (Eds.), Inside case-based explanation: 207-261, Hillsdale, NJ: Erlbaum.

Reeves, B. and C. Nass: 1996, The media equation: How people treat computers, television and new media like real people and places, CSLI / Cambridge University Press.

Rhodes, B. and P. Maes: 2000, Just-in-time information retrieval agents, IBM Systems Journal 39(4): 685-704.

Ricoeur, P.: 1965/1970, Freud and Philosophy: An Essay on Interpretation, transl. Denis Savage, New Haven: Yale University Press.

Riesbeck, C.K. and R.C. Schank: 1989, Inside Case-Based Reasoning. Lawrence Erlbaum Associates, Hillsdale.

Roget, P.: 1911, Roget’s Thesaurus of English Words and Phrases. Retrieved from gutenberg.net/etext/10681

Rosenblatt, L.: 1978, Efferent and Aesthetic Reading, The Reader, The Text, The Poem: A Transactional Theory of the Literary Work: 22-47, Carbondale: Southern Illinois UP.

Schank, R. C. and R. P. Abelson: 1977, Script, plans, goals, and understanding. Hillsdale, NJ: Erlbaum.

Schleiermacher, F.: 1809/1998, General Hermeneutics, in A. Bowie (Ed.) Schleiermacher: Hermeneutics and Criticism: 227-268. Cambridge University Press.

Shardanand, U. and P. Maes. (1995). Social information filtering: Algorithms for automating `word of mouth'. Proceedings of the ACM SIGCHI Conference on Human Factors in Computing Systems: 210-217.

Singh, P.: 2003, Reaching for dexterous manipulation. EECS Area Exam Paper. Accessed 12 November 2005 at http://web.media.mit.edu/~push/Reaching.pdf.

Singh, P.: 2005, EM-ONE: An Architecture for Reflective Commonsense Thinking, MIT Media Lab PhD Thesis.

Singh, P., B. Barry, H. Liu: 2004, Teaching machines about everyday life, BT Technology Journal 22(4), 227-240, Kluwer Academic Publishers.

Singh, P., T. Lin, E. T. Mueller, G. Lim, T. Perkins, W. L. Zhu: 2002, Open Mind Common Sense: knowledge acquisition from the general public. Proceedings of ODBASE’2002.

Sloman, A.: 1981, Why robots will have emotions. Proceedings of the Seventh International Joint Conference on Artificial Intelligence.

Stanfill, C. and D. Waltz: 1986, Toward memory-based reasoning, Communications of the ACM 29(12): 1213-1228.

Thornton, S.: 1996, Club Cultures: Music, Media and Subcultural Capital, Wesleyan University Press

Winograd, T.: 1971, Procedures as a Representation for Data in a Computer Program for Understanding Natural Language. MIT AI Technical Report 235.

Zwaan, R.A. and G.A. Radvansky: 1998, Situation models in language comprehension and memory. Psychological Bulletin, 123(2): 162-185