Jesse Saba Kirchner
And in Some Language That’s English? Slayer Slang
and Artificial Computer Generation*
 New viewers of Buffy the Vampire Slayer are often struck by the show’s use of distinctive slang. Every episode contains some sort of anomalous English, from the ubiquitous wiggins to the exotic Slaymaster General. Interest in the use of slang in Buffy the Vampire Slayer (henceforth Buffy or BtVS) has occasioned a significant proportion of media coverage about the phenomenon that was Buffy. A more comprehensive treatment of the subject is given in Michael Adams’s Slayer Slang (2003), a book including a linguistic analysis of what Adams calls “slayer slang,” as well as an extensive lexicon of slayer slang terms.
 Despite the attention focused on this aspect of BtVS, most write-ups do not go beyond the descriptive level of listing a few hand picked terms from the cornucopia of slayer slang used in the show. Therefore, work remains to be done to fully understand slayer slang as a linguistic phenomenon and to understand the role it has played in the online BtVS fan community, many of whose members adopted terms from the show for their own use and even invented terms of their own in the style of the slayer slang used in the show.
 In Fall 2003, I initiated a research project to investigate slayer slang in a novel way, by creating a computer program capable of automatically generating sentences in the style of slayer slang. This paper describes the methodology of that program’s construction and operation, the results obtained by it, and a sketch of the implications of this research and future research in this vein for understanding slayer slang from the perspectives of linguistics, sociology, and stylistics.
Slang in Buffy
 The words and phrases that we call slayer slang come in many shapes and kinds. The indispensable compendium of BtVS slang is Adams’s Slayer Slang (2003). Using the data in this book, I divided slayer slang into seven different levels, with each level representing a different kind of slangy deviation from ordinary English. I outline and illustrate those levels in Table 1. (The nature of each level is discussed in the next section.)
table 1: seven levels of Buffy slang
1. Jargon and new vocabulary, e.g.:
2. Affixation (suffixes and prefixes), e.g.:
Go act baity! (“Anne,” 3001)
He’s not one to overshare (“Halloween,” 2006)
He was unmad (“Halloween,” 2006)
3. Changing the part of speech change without affixation, e.g.:
How much the creepy is [that]? (“Out of Mind, Out of Sight,” 1011)
It gives me a happy (“Lie to Me,” 2007)
4. Other syntactic change, e.g.:
Having issues much? (“Ted,” 2011)
5. Truncation, e.g.:
What’s the sitch? (“Welcome to the Hellmouth,” 1001)
6. Discourse sensitivity and semantic shift, e.g.:
Giles: Punishing yourself like this is pointless.
Buffy: It’s entirely pointy! (“When She Was Bad,” 2001)
You are sadness personified ("Two to Go," 6021))
7. Pop-culture references, e.g.:
I can’t believe you of all people are trying to Scully me. (“The Pack,” 1006)
Does anyone feel like we’ve been Keyser Sozed? (“Puppet Show,” 1009)
Of course, the slang in BtVS could be divided in other ways. It would also be reasonable to divide slang by chronology or by the characters who use particular slang items or slang types. Constructing a typology of slayer slang is complicated, because slayer slang contains a great deal of information about the speaker and about what speakers assume of hearers. (I discuss this in more detail in later sections.) The exchange is mediated differently depending on whether one looks at slang used on the show itself, in which case two kinds of audience can be assumed (those on the screen and those watching the screen), or one looks at slang used in the online Buffy fan community, where different assumptions about audience operate.
 Dividing slayer slang by chronology would uncover different facts about the phenomenon. Broadly speaking, slayer slang goes from a period of great innovation and change, in the first two or three seasons, into a period of stability until season five, followed by a period of marked decline in the use of slang (along with an interesting return of some kinds of innovation, particularly jargon) in the final two seasons. Michael Adams (2004) has connected some of these changes with the story arc of the show and with the audience’s developing perspectives on the characters.
 Dividing slang terms according to the characters with which they are associated provides another interesting way to understand the slang system as a whole. On one hand, some aspects of slayer slang are common to all or almost all of Buffy’s main characters; on the other hand, some characters are very strongly distinguished by their relationship with slayer slang generally or with specific elements of it. For example, Giles is distinguished by his nearly complete nonparticipation in slang. In fact, this tendency applies to all adults in Buffy, but Giles takes it to the extreme. Faith also has a strong tie to particular slang, especially her characteristic phrase five-by-five. In the episode “This Year’s Girl” (4015), in which Faith and Buffy switch bodies because of a magical artifact, Faith-in-Buffy reveals herself to the television audience by using that phrase. Faith also idiosyncratically truncates Buffy’s name to B. At the same time, Faith’s inability to use other slang terms in the same way as Buffy, Willow, and Xander is symptomatic of her failure really to become part of the Scooby Gang.
 An interesting character to consider from this sociolinguistic perspective is the Buffybot. Since the bot is physically indistinguishable from Buffy, viewers rely on differences in their speech patterns to distinguish them. This difference is conveyed in part by the bot’s inability to use slang like Buffy’s. In Season Six, before the Buffybot is destroyed, her inability to use slang correctly: for example, she attempts a Buffyesque post-slayage pun but comes up with, “That’ll put marzipan in your pie plate, bingo!” (“Bargaining, Part 1,” 6001) symbolizes her failure to substitute for Buffy. Seen in this light, slang is an essential quality, the kernel of each young character’s personality.
Studying slayer slang
 One of the most important aspects of Slayer Slang (2003) is that it went beyond simply listing every instance of slayer slang to attempting a linguistic analysis. Slayer Slang explains a great deal about what is characteristic of slayer slang and why it is an interesting topic of study. But it raises more questions than it answered: Is slayer slang really governed by rules, as Chomskyan linguistics asserts that all human languages are? Or is it just the product of writers taxing their imaginations to try to sound like hip teenagers? As Adams (2003) writes, “the hazard of fictional jargon for a fictional profession is that it does not develop naturally, but rather in authors’ imaginations” (16). No one gave Buffy’s writers a handbook on how to write like Buffy talks. Instead, the writers had to rely on invention or imitation of what had been written before. As BtVS writer and producer Jane Espenson notes in the introduction to Slayer Slang, “the only thing that gives us coherence is that we’re all writing segments of the same story and that we’re all doing our darnedest to do a Joss Whedon impersonation” (Adams 2003, ix). Is that coherence enough to produce a consistent, rule-governed kind of slang?
 The balance between repetition and innovation constitutes another problem for slayer slang. Repetition of terms and patterns is crucial for making slang understandable and usayable by wider groups of people. But some of Buffy’s most memorable slangy lines are single-use nonce-formations that refer to pop culture or cleverly extend linguistic commonplace:
Buffy: I’m the one getting single-white-femaled here. (“Faith, Hope and a Trick,” 3003)
Giles: Punishing yourself like this is pointless.
Buffy: It’s entirely pointy! (“When She Was Bad,” 2001)
Ubiquitous slayer slang like Scooby Gang (with all its derivative forms) traces back to a single line written by a single author. Of course, every item of slang has to begin with some single use. But there is a contradiction in any slang system between the need to retain its novelty and edge, to resist incorporation into mainstream language and, at the same time, to exert its own conservative force, one that promotes maximum comprehensibility by keeping meanings static and discouraging innovation. The research described here was designed to help us understand how slayer slang maintains the balance between innovation and repetition.
 The question of whether or not slayer slang is really rule-based was what first intrigued me about studying the use of language in BtVS. As I soon discovered, the importance of this question goes beyond a purely linguistic level of interest: it is also important for understanding the relationship between slayer slang as used in episodes of the show (what I will call “canonical slayer slang,” as explained in the next section) and the slayer slang used by fans and other followers of Buffy, including writers of fanfiction and others in the online Buffy community, writers of the official BtVS novels, and also the writers of Angel. If slayer slang can be called a rule-based slang system, then it should be possible to see how faithfully different users of slayer slang follow the rules laid down by canonical slayer slang and how those users challenge or develop those rules. If, however, slayer slang does not follow predictable rules, then it is necessary to ask how we can identify the slayer slang used by fans and other non-canonical speakers as such. The answer is important to any accurate understanding of the relationship between BtVS and its fans and followers. It also bears on the projection of lexicon into syntactic structures, as the latter are systematic, but development of a lexicon is relatively unpredictable.
 As I began to study this question, my interest in the use of language and slang in BtVS expanded into other areas. Studying the adoption of slayer slang led to the study of the group of people adopting it. What could we learn about Buffy fans from the stylistics of slayer slang? Like Dick Hebdige (1979) in his study of style in youth subcultures, I became “intrigued by the most mundane objects [such as language] which, none the less, . . . take on a symbolic dimension, becoming a form of stigmata” (2). Although less stigmatized than the punk rock styles studied by Hebdige, slayer slang can also be understood as a “signifying practice,” the nature of which reveals information about the individuals and groups who adopt the practice.
 These were the general questions driving my research: How much of slayer slang was derivable from consistent rules? And what could we learn about the users of slayer slang from the slang itself? Answers to these questions originate in the project’s design.
 In designing my project, I had the good fortune of following the work of other writers who had already begun to study slayer slang in a rigorous way. Thus, I could approach slayer slang from a new direction, using computational methods. The mandate of the slayer slang sentence generator was to answer this question: Can a computer, given the vocabulary of the show and the patterns of slayer slang, produce original sentences that sound like real dialogue from BtVS?
 This question was not picked quite as much at random as it might seem. Computers are useful tools for the investigation of language phenomena because they lack some of the amazing language abilities that all normal human beings take for granted. Humans have highly evolved brains that can accommodate and make sense of sentences that are awkward or ungrammatical. But a computer is entirely constrained by the rules given to it and cannot accommodate beyond them. This limitation is an advantage because it very quickly reveals any flaws in a linguistic analysis. However, it also presents certain liabilities: perhaps the most significant of these is that computers do not adhere to the Cooperative Principle that governs conversation and similar modes of discourse unless explicitly programmed to do so. Because slang is often unique to the conversational exchange in which it occurs, the computer’s conversational ineptitude hampers its production of slang. I return to this point below.
 The resources needed for this project were as complicated as the requirements for it. Two kinds of data were needed: slang terms used in BtVS and information about how patterns of slang were formed and with what frequencies. Adams’s Slayer Slang (2003) serves as a point of entry for both types of data. Its extensive lexicon is the most comprehensive list of slayer slang assembled to date, drawing on several different sources; the lexicon is also large enough to provide statistical information about frequency and patterns of use. The lexicon lacks some slang terms that should have been included (for example, items from the show’s last season), so statistics derived from it are not completely accurate. However, they are more than adequate for a computer program that includes proof of concept as a major purpose.
 The lexicon in Slayer Slang lists slayer slang gathered from a number of more or less distinct sources: from episodes of BtVS; from the authorized novels set in the Buffyverse; and from the online Buffy fan community, particularly the two official fan forums for the show (namely, the Bronze during Buffy’s five years on the WB, and the Bronze: Beta after the show moved to UPN). Slayer Slang presents all of these terms as examples of slayer slang, which indeed they are. However, in designing this program, I was more selective. For reasons explained in some detail below, I used only the slayer slang words and patterns that actually appeared in episodes of BtVS, what I will call “canonical slayer slang.” This category is distinct from slayer slang in BtVS novels, online fan forums, fanfiction, and other discourse, most of which is more or less derivative from the slang used on episodes of Buffy. The decision to use only canonical slayer slang as the basis for the program was intended to keep linguistic analysis of slayer slang as unproblematic as possible.
 Canonical slayer slang was the source of data that I divided into the seven different levels shown in Table 1. Those levels divide the slang according to types of linguistic deviation from mainstream American English and also largely correlate with the predictability or novelty of member forms.
 The first level, comprised of jargon, is the easiest to include in such a program. All that is necessary is to compile a list of the particular words. It is also one of the least productive and least innovative areas of slayer slang. There are very few enduring jargon terms introduced later than Scooby Gang in Season Two (and Scooby Gang could be considered a Level Seven pop-culture derived form, as much as just jargon). The levels grow more and more difficult to incorporate into a sentence generator until they reach a point of near-impossibility (although the exact frontier depends on the skill of the programmer). Very few people in the world would invent the term Undead-American, and no computer can match such innovation yet.
 The scale does not correspond exactly to the frequency with which each kind of slang is used. The most productive and most frequently used kinds of slang are the suffixes -age and -y (in words like slayage and crayon-breaky). These suffixes are only Level Two slang in terms of implementation difficulty. All the words at Level Two, however, are products of relatively few prefixes or suffixes, which generally work in predictable ways. If jargon and affixation were not restricted, slayer slang would become so complicated and cumbersome that only full-time Buffyologists could understand it.
 On the other hand, pop-culture derived words and utterances admit no such limits. Counting their occurrences is tricky, but it’s easy to think of many examples, from the first season (“She’s our Sabrina” [“The Witch,” 1003]) to the last (“It’s like somebody slaughtered an Abercrombie & Fitch catalog” [“Selfless,” 7005]). If Buffy had gone on past seven seasons, quips like these would have continued to pile up, while new jargon terms and new affixes would have been comparatively rare.
 Slayer slang from Levels One and Two is an integral part of the program. The vocabulary used by the program began with the list of jargon and Buffy-specific terms that would be used and then was expanded to include other generally useful words. Affixation and other patterns of forming new words (or finding new uses for old words) were also included in the basic functioning of the sentence generator. The sentences generated by this program are like canonical slayer slang in that most instances of slang come from these two levels. Slang from Levels Three, Four, and Five was implemented with only partial success. Slang from these levels was included only when there existed an explicit precedent in BtVS. The program itself has no ability to create new words by means of truncation or changing parts of speech. Levels Six and Seven forms are more or less absent from the program. This is an area that particularly needs to be addressed in future versions of the sentence generator.
 It is important to note that none of these kinds of slang from the higher levels is impossible to characterize linguistically or to implement in a program. For example, the nouning of words like creepy and happy works in a straightforward way from their usual uses as adjectives. This phenomenon has been the subject of detailed investigation at least as far back as Clark and Clark (1979). Their analysis not only treats the functional shift common nouns, but also proposes a theory to explain the use of proper names and other nouns surfacing as verbs with meanings that are not straightforwardly predictable. They illustrate with examples from real texts: “You’re in danger of being Hieronymous Bosch’d”; “She wasn’t Krishna’d out, she was only hippied out.” Clark and Clark propose one analysis for what I have identified as elements from two different levels of slayer slang—changing parts of speech and making pop-culture references. This indicates that some of the distinct problems faced by the Buffy sentence generator are manifestations of a smaller number of significant gaps in the program’s capabilities.
 Clark and Clark’s analysis actually identifies what is now the main gap in the Buffy sentence generator’s output, sentences whose meaning depends on active cooperation in a conversation. Such cooperation in turn depends on shared knowledge of the world and the most salient properties of the objects and people in it. This kind of slang—Level Seven in my hierarchy—has the most obvious sociological importance. Slang depending on shared knowledge serves to create group solidarity and to separate knowledgeable insiders from ignorant outsiders. It is also the kind of slang that a program like the current version of this sentence generator cannot generate, because it has no sense of dialogue or cooperative speech.
 What is needed to fill the gap is a computational system that builds conversation and stores information that serves as shared information for the cooperative process of generating higher level slang. Perhaps most appropriate would be an implementation of Hans Kamp’s Discourse Representation Theory (Kamp 1981). It should be noted, however, that any substantial improvement in the sentence generator could be obtained only by implementing such a system (at considerable expense). Short of that, electronic artifice will not scale the highest ramparts of BtVS linguistic creativity, the sort represented in “I know Faith’s not gonna be on the cover of Sanity Fair” (“Doppelgangland,” 3016).
Program design and operation
 The details of how such a program is put together are more technical than would bear the telling. Nevertheless, it is useful to have a conceptual understanding of how computer-generated slayer slang is created. Inherent in this method are the project’s strengths and weaknesses, which are apparent in the results it has produced.
 The sentence generator takes a “seed” generated at random and turns it into a grammatical (and slangy) English sentence by passing it through a series of modules tailored for different tasks. We can look at the design of the program by following an example sentence (taken from the real data in the appendix) as it goes through its path through the different modules. A list of these modules is given in Table 2.
table 2: component modules of the Buffy slang generator
Main program: scoobyage.cgi
The sentence we will follow through the program is the following:
118. I think some researchy gals are not loving that rushy werewolf.
 The program is initiated when a user loads the Website where the program resides into her Internet browser. The program here is represented by the main Webpage, called scoobyage.cgi. This page calls various modules, each of which performs some special tasks, and then prepares the sentence it has generated for display to the user:
scoobyage.cgi starts up, loads all the modules it will use later on, and then summons the first module: Pattern_Generator ;
Pattern_Generator has a list of possible sentence patterns and it picks one at random: NP.VPb, that is, “Noun Phrase” and “A ‘to be’ Verb Phrase” (i.e., a verb phrase including some form of the verb be), the most basic skeletal pattern on which the eventual sentence is built;
NP.VPb is passed to the next module: Structure_Builder; the pattern is expanded by another random pick from a list of possible structures based on NP.VPb. NP is expanded to D.AJ.N—a determiner (such as “the” or “some”), an adjective, and a noun; VPb expands to VB.VS.PROG.NP, meaning the verb “to be” followed by a singly-transitive verb in the progressive aspect, followed by the noun phrase which is the object of that verb.
Since there is still an unexpanded phrase—the NP at the end of VPb—the module reiterates. It expands that NP into D.AJ.N. Now there are no more unexpanded phrases, so the concatenated structure goes to the next module—Lexicon—which will turn it into almost-English.
 The input into Lexicon is this structure:
Lexicon fills each placeholder with a word of the appropriate type. Each word is chosen at random from a list of words of that given type. This means that the sentence being created might end up grammatically correct but semantically unlikely, either because it is unidiomatic English or because it is implausible said of or in the Buffyverse. This problem will be dealt with in a later module, but we need not worry about it at this point, because we know that the words in this sentence make sense together.
 The skeleton from Structure_Builder is replaced by this pseudo-English sentence: a research+y gal be love+ing that rush+y werewolf. Concealed in this sentence is one of the most characteristic features of slayer slang, namely rampant suffixation, particularly of -y. Of the terms included in Slayer Slang that were found in BtVS episodes, almost one-third of them were produced just by adding the -y suffix. This very important suffix is represented in this program by two possible ways of picking adjectives, either from a list of standard English adjectives or by adding -y to any of the words in the noun or verb lists. Simple random choice makes the first noun phrase plural but keeps the rushy werewolf singular. Another random choice makes the sentence negative instead of positive, so the result looks like this:
a+pl research+y gal+pl be not love+ing that rush+y werewolf
 The next step is to fix the orthography of the sentence, for which purpose we have the module Orthographizer. Simple substitutions turn a+pl into some, research+y into researchy, etc. They also find the appropriate conjugation for be given a plural subject, and the final product of this module is
some researchy gals are not loving that rushy werewolf
which is then sent to the last module. The sentence at this point has already been put in working order as far as grammaticality is concerned. The last step before preparing the sentence for presentation is to make sure that the content of the sentence makes some sense as well. The last module is therefore named Theta_Checker, after the linguistic notion of “theta roles,” which refers to those restrictions on classes of words that are based on the meaning of other words in the sentence. These are the controls that, given a verb like eat, will limit the possible subjects to living things that can physically consume other things. Checking that all proper theta restrictions are adhered to is the main, though not the only, task for this module.
 Any sentence generator for human language must perform a number of theta role checks, but a Buffy sentence generator must perform some additional checks. For instance, a standard theta role check makes sure that nouns in certain positions must be animate. Verbs like hit or kick require an animate subject: in this sense, they are analogous to verbs like sire and feed which (in the world of slayer slang) require an undead subject. Other slang terms may have non-slang meanings that allow a wider range of use, but their usage in BtVS tends to adhere to the slang definition. So, while one could imagine a situation in which it would be necessary to say, “She staked Joyce!” it is much more common for the verb stake to be used only when the stakee is a vampire and prone to dusting. Theta_Checker therefore disfavors the non-slangy use of slayer slang.
 Theta_Checker also checks other aspects of the sentence for acceptability. Certain words are syntactically plausible but improbable because they sound like standard English words or because they are phonologically difficult. Words like “hottiey” (as in “She has a hottiey brother”) and “destinyey” are disallowed because of the difficulty in pronunciation. The module rules out sentences that repeat a vocabulary element more than once, to avoid typical machine-produced awkward sentences like “Buffy kicked Buffy.”
 These and other checks cull the sentences that are most likely to be unacceptable for reasons of semantics, phonology, or logic. In a more powerful program, such sentences could be fixed, but because the processing power for this whole program is negligible, when Theta_Checker detects an error it instructs scoobyage.cgi to reject the entire sentence and begin from another random generation. Thus, when this program is run, it may consider and discard any number of problematic sentences before finding an acceptable sentence to display.
 The test sentence we have followed is almost ready for display now. The last step is another random check to decide whether to add an “interjection” to the sentence. The “interjections” used by scoobyage.cgi vary widely in their exact functions and could be said to interact variously with the rest of the sentence. But all of them have in common with real interjections the attribute of being included in a sentence mostly to indicate the speaker’s feelings or thoughts about the statement or question being made (surprise, happiness, doubt). This time the random check picks out I think, which is effectively an interjection serving to distance the speaker from the factuality of the statement being made. This is concatenated with the sentence, punctuation is added, and capitalization is checked. The sentence is finally ready for display and, along with HTML tags for Web display, the sentence is displayed for the program user:
I think some researchy gals are not loving that rushy werewolf.
This process, although tortuous to fully explain in prose, takes only a fraction of a second to be computed, making the response to the user's request for a sentence almost instantaneous.
 The corpus of 150 sentences generated in sequence by this program (included as an appendix to this essay) provides a representative sample of both the strengths and weaknesses of the Buffy sentence generator. A corpus of this size is large enough to suggest the rate of success of this program and to suggest directions for future development.
 Out of the full 150 sentence corpus, three sentences are positively ungrammatical:
(57) Faith didn’t love who.
(102) Her freaky sires pretty don’t like some demons.
(133) Her pretty much don’t like Cordelia.
These sentences are marked in the appendix with the symbol “*.”
 In addition to these three ungrammatical sentences, there are eleven sentences almost impossible to imagine in a BtVS episode. These sentences may not be grammatically impossible, but they are implausible for other reasons. Nine have the appearance of being contrived; they seem to be slayer slang run amuck, to the point where they are almost impossible to understand. Consequently, they are hard to imagine as actual BtVS sentences in any context. The other two questionable sentences are unlikely because of the real personalities and chronologies of the BtVS characters to whom they refer. Although almost anything is possible in the Buffyverse, these utterances seem basically implausible. Examples of the two types of unlikely sentences include the following:
(1) Your totally violent guys don’t love her gatheringy ashes.
(148) The pretty sirey slayer didn’t like Giles.
These sentences are marked in the appendix with the symbol “?.”
 The shortest sentences in the corpus are two words long. These include
(33) Cordelia freaked.
(64) Warren babbled.
Easy to understand, they are not particularly interesting. The reverse is true of the longest sentences in the corpus, weighing in at 11 words. There are two sentences of this length:
(77) Don’t you think Anya pretty much didn’t like her baddie minions?
(80) Okay, the unusual vampires pretty much don’t love this witchy bad.
The average sentence length is 6.227 words. To determine whether these tendencies correspond with those of canonical slayer slang, it would be necessary to analyze the entire body of Buffy transcripts very carefully. These lengths are certainly consistent with ordinary casual English.
Remaining issues to be researched
 Two areas on which this research does not shed light are differences in slayer slang among BtVS characters and changes in slayer slang over the course of seven seasons, not to mention in its extensions on Angel, into Buffy fandom, and perhaps even into mainstream speech. The approach taken in the construction of this program, treating all the canonical terms in Slayer Slang as of equal weight, ignores the important differences in speakers and contexts.
 The original design for the sentence generator included a plan to produce sentences in the peculiar styles of particular characters. After all, it is obvious to any Buffy fan that each major character has a unique way of talking. One of Xander’s lines would never sound right were it spoken by Riley. But capturing these stylistic differences systematically would be the work of a much larger project. It would entail using Buffy scripts or transcripts to build an individual corpus of utterances for each major character. Then each corpus could be compared with the others to find the salient differences among them. One might even wish to go further and to treat each utterance with sensitivity to the preceding dialogue, and to the other characters present, as well as to other aspects of the context. Each step in this direction would produce a program better trained, not only to generate sentences that successfully imitate slayer slang, but actually to write plausible dialogue according to the principles used by the show’s writers. But it also would entail ever-growing corpora and constantly increasing program complexity.
 Changes in slayer slang over time include changes in the composition of the slang and changes in the role of slayer slang in BtVS. Examples of the former are the replacement of old slang by new words or patterns, such as when slayerette gave way to Scooby Gang, or when the suffix -age declined while the suffix -y grew increasingly common (Adams 2003, 19 and 42). An example of the change of the role of slayer slang in Buffy is the general decline of slang in later seasons (Adams 2004). It is impossible to analyze the slang in BtVS fully without taking both kinds of change into consideration. What really must be sought is an understanding of slayer slang that transcends the superficial (transcript) level in multiple dimensions: it must extend to some understanding of the birth and re-birth of slayer slang in the minds of Buffy’s writers and must accommodate adaptation of slayer slang by Buffy’s fans.
Interpretation and analysis
 The results of the slayer slang sentence generator tended to give a strong affirmative answer to the question of whether computer-generated sentences could really sound like slayer slang. Although the program is imperfect, its failures do not undermine the possibility of such generation. So, to answer this article’s original question, how much can we consider slayer slang to be shaped by consistent rules? The computer sentence generator shows that slayer slang is heavily rule-based, since the computer is only capable of following rules: it is certainly possible to create slayer slang sentences by adherence to consistent rules.
 But this picture is incomplete. We must acknowledge that although the sentence generator can reliably create novel sentences in the style of slayer slang, it cannot generate every kind of slayer slang that occurs in other sources. Some kinds of this slang (those determined by pop culture references and discourse sensitivity, for example) entirely evade computer generation. And the sentence generator also produces implausible and impossible sentences. Thus, the rules used by the slayer slang sentence generator are necessary but not sufficient to produce the entire corpus of slayer slang and nothing else. The relationship between these predictable rules and the unpredictable aspects of slayer slang is one of the key elements in the organization of the slayer slang speaker community.
 The questions this program was designed to study are critical for understanding the nature of Buffy fandom, and they go to the heart of what made Buffy different from other TV shows. With regard to a community like that of Buffy fans, connected through online messageboards, fansites, and other Internet-based communication, the importance of language cannot be overstated. This is true not only in terms of content, but also in terms of language used as a sign of affiliation and a factor of group cohesion. Slayer slang is the secret handshake of the Buffyverse, and for that reason understanding language in the show and understanding the two-way relationship between BtVS and its fandom is much more important than studying the slang for its own sake, as an object of linguistic interest.
Slayer slang and slayer style
 Dick Hebdige (1979) describes style (and particularly linguistic style) in subcultures as a “signifying practice” (118). In other words, style (including slang) is best understood as an action or a process, taking place between a speaker and an audience. It is a process in which a message is transmitted. Although this may be a literal message—as in the case of any utterance of slayer slang—the literal content of the message does not exhaust its meaning: the manner of the message, that is to say its style, transmits information about the speaker and the audience, as well. This information may pertain to their respective sex or class or their worldview or relationship to mainstream culture and is conveyed in the active practice that is style. As Hebdige says, “subcultural styles do indeed qualify as art but as art in (and out of) particular contexts; not as timeless objects, judged by the immutable criteria of traditional aesthetics, but as ‘appropriations’, ‘thefts’, subversive transformations, as movement” (129).
 What can we learn about the speakers of slayer slang from the nature of the slang itself? In this question, we are concerned with speakers of canonical slayer slang—slayer slang as used in episodes of Buffy—and with fans of the show who adopt slayer slang as an idiom of their own. The practice of slayer slang suggests a speaker (and an audience) invested in pop culture. Pop culture references are used frequently and usually without an explanation. Slayer slang tends to use pop culture references, not just as metaphors, but also as building blocks of language, nouns or verbs that can be affixed onto or put into another tense or case. This is no less true of slayer slang as used by BtVS fans than it is of the slang used in the show. Slayer Slang includes examples of pop culture used in these ways in Buffy online communities. One example is Ewanage, seen in the Bronze and defined by Michael Adams (2003) as “Exposure to . . . Ewan McGregor” (175).
 These references are sometimes so casual that the pop culture terms have become literally a part of the language; yet they often require a complex understanding of the reference. To understand Buffy’s statement, “I can’t believe that you of all people are trying to Scully me,” not only must the listener must know what character is being referred to and what television show she comes from, but furthermore must be able to pick out from among all of that character’s attributes (being female, being a government employee, being fervently Roman Catholic, etc.) the one relevant to the moment (being skeptical).
 Remembering Hebdige’s description of subcultural style, we should consider the message conveyed by this kind of slang at a level beyond the literal. What is conveyed by these common yet complex and layered pop culture references? They suggest a deep investment in popular culture and require thorough familiarity with a wide range of American popular culture. This suggestion is quite ironic, given the many patterns in slayer slang that are distinct from standard English and therefore marks its distance from mainstream culture; but the contradiction represents a central aspect of the sense of identity and cultural affiliation encompassed by the use of slayer slang.
 Slayer slang also tends to develop new usages and creative twistings of words already incorporated into the slang. For example, the new term Scooby Gang was changed and played with on Buffy (Scoobies, Scoobs, Scoobycentric, Scooby-sense, etc.), as well as in the BtVS fan community. Slayer slang both illustrates and depends upon a speaker’s cleverness and discourse sensitivity. This aspect of slayer slang more than any other derailed the idea of analyzing full BtVS episode transcripts to gather complete statistical data about the patterns of slayer slang, given the frequency and complexity of exchanges like these:
Giles: Let’s not lose our perspective here, Xander.
Xander: I’m Perspective Guy. Angel’s a killer. (“Becoming, Part 1,” 2021)
A similar but distinct kind of slayer slang in BtVS fan communities is exemplified in posts from the Bronze saying Slay you later or Don’t let the bedvamps bite. Since online message board conversations cannot have the temporal immediacy of real-time discourse, perhaps these examples of slayer-slangy plays on English clichés are the closest possible thing in online fandom. Certainly they suggest a similar cleverness and a keen awareness of language.
 Significantly, these sentences use the least mechanical aspects of slayer slang. In the terms in which I proposed my original problem, they are characteristics that rely on innovation instead of repetition. The question might be raised of whether it is fair to pick out these characteristics as the most essential parts of slayer slang. Are they not, after all, less common than slayer slang characteristics like jargon and novel affixation? The relevant measure, however, is not frequency. The most important elements of slayer slang (as of any kind of slang) are those that reveal something of the speaker’s social position: -y suffixation could be just as easily part of thieves’ cant or an elite affectation. The most frequent items of slayer slang are revealing about slayer slang speakers only if they are typical of youth slang or have some other sociological relevance.
 The fact that such nuances fall through the cracks of a computerized version of slayer slang may seem like a disheartening result. Yet the program acquits itself well enough: after all, it produces reasonable slayer slang much of the time. Instead of throwing up our hands over this result, we can let it lead us to a deeper understanding of the slayer slang speech community. Recall that the data used to design this program came only from canonical slayer slang, and data from BtVS online communities was not included. Sentences generated by this program therefore are a sort of “second-order” slayer slang. In this sense, they are like the slayer slang sentences used in BtVS fan communities. Both are weakly bound by canonical slayer slang: they can be innovative but cannot depart too radically from the text of BtVS or they will become unrecognizable as slayer slang. Of course, the members of fan communities can and do innovate in more interesting ways than this computer program. Fan innovations have also occasionally been cited as inspiring slayer slang later used in Buffy. This truly dialectical behavior seems rare, however. We can at least impressionistically see a general adherence to the patterns laid down by canonical slayer slang, in both second-order slayer slang sources.
 How much is the slang of scoobyage.cgi like the slang used in the Bronze and other BtVS fan sites? Once we look past the jargon particular to the online fan communities (bezoar, bitca, etc.), we find firm conclusions evasive—therein lies another research project.
 What is at stake in the current research? We seek to understand a basic cleavage in the community of slayer slang speakers, between innovators and imitators. This is another aspect of the same question with which I began my research. Although I was interested in the overall balance of innovation and conservatism in slayer slang, it is also important to understand the relationship between them. This perspective places slayer slang in the context of other slang systems. Any group with slang of its own and in which privilege differentials exist (in slayer slang, one group of speakers—the Scooby Gang— certainly has a privileged status for its speech) may tend to separate innovation and repetition among distinct groups of speakers.
 My research began with the goal of answering questions about the nature of slayer slang and about its relationship to its community of speakers. To find these answers, I constructed a computer program that uses the vocabulary of BtVS and the patterns of slayer slang used in the show to construct original sentences in the style of slayer slang. Theoretically, this program can generate an infinite number of never-before-spoken sentences that are unmistakably in the style of slayer slang. This gives a solid answer to the first question raised in this paper: yes, slayer slang is largely based on rules which are as consistent as the rules of dialects and other variation from mainstream speech in the real world.
 More difficult questions concern the relationship of slayer slang to its community of speakers and the different levels of command and authority over the slang given to different groups and individuals within that community. Here the products of the slayer slang sentence generator are inconclusive: more work must be done in this area before we can answer all of the attendant questions, including those about the reciprocity between canonical slayer slang and the slayer slang used by fans and other language users, those underprivileged within the community of slayer slang speakers. How much do their extensions of slayer slang and their original coinages feed back into canonical slayer slang? A greater influence by BtVS fans on the development of canonical slayer slang would change our understanding of the relationship between the different groups of speakers. Work must also be done to investigate the development of slayer slang within the show over the course of seven seasons, along with its parallel development on Angel.
 This article explores many new ways to study longstanding questions concerning the characteristics of slang and subculture. Like Adams’s Slayer Slang (2003), this work problematizes its own conclusions by looking forward. Thus I do not apologize if this article raises more questions about slayer slang than it has answered: some other student of the Buffyverse will feel the need to solve the remaining problems and, in doing so, will advance our shared understanding of that world and what it tells us about ourselves.
Appendix: Buffy sentence generator corpus
(Gathered August 9, 2004, 4:36 - 4:40 p.m.)
(1) Your totally violent guys don’t love her gatheringy ashes.
(2) Who vamped Faith?
(3) Alright! Her yawnworthy ashes didn’t dust Angel.
(4) Your demons are not brooding.
(5) Xander researches.
(6) Okay, her sombernesses totally freaked.
(7) I think the sillinesses are vampirey.
(8) You know, her sadness doesn’t show.
(9) Don’t you think her somberness totally doesn’t love your majornesses?
(10) I think Xander doesn’t miss that freaksome guy.
(11) Her freaksomenesses don’t watch your slayers.
(12) Okay, Angel so dusts her sirey school.
(13) The girlfriend didn’t stake Spike.
(14) And yet some totally clue-free gatherings had been feeding.
(15) Alright! A minion-free gal very much doesn’t like Cordelia.
(16) Don’t you think her girlfriend totally wigs?
(17) Some witchy werewolves had very not been feeding.
(18) The really watchery covens are feeding.
(19) Whoa, he really researches.
(20) Okay, Willow didn’t brood.
(21) He really avoids.
(22) Principal Wood pretty much didn’t freak.
(23) This sadness has pretty much been feeding.
(24) Cordelia doesn’t watch her wickednesses.
(25) That destiny-free girlfriend kinda didn’t like his minion-free crazy.
(26) Xander didn’t like your sitches.
(27) A slayer pretty much loved your dollsomeness.
(28) Your coveny vampires were wigging.
(29) And yet Oz didn’t wig.
(30) Maybe the Mayor freaked.
(31) Spike vamps her dead demons.
(32) Whoa, your girlfriend-free sitches have really not been feeding.
(33) Cordelia freaked.
(34) These sadnesses have not been feeding.
(35) Okay, her bad didn’t hit the witch.
(36) The yawnworthinesses had really been feeding.
(37) And yet the violent slayers totally freaked.
(38) Joyce was Anya.
(39) Okay, some minion-free slayers pretty much like Xander.
(40) Okay, these slayers love these sitches.
(41) That gal pretty much didn’t like her vampires.
(42) Buffy was not wigging.
(43) Xander totally rushed.
(44) And yet Xander very much didn’t avoid Buffy.
(45) Whoa, her miniony demon liked some girlfriendy gals.
(46) Oz totally showed.
(47) You know, her freaksome clues very much don’t babble.
(48) You know, Willow is coveny.
(49) You know, a stakey guy so doesn’t avoid.
(50) Her skankinesses don’t like that yawnworthiness.
(51) Okay, a demon dusts Spike.
(52) Spike muchly likes those pretty wiggy clues.
(53) Don't you think Anya doesn’t dust her vampirey weirds?
(54) Buffy didn’t dust her broody library.
(55) Don’t you think Anya loved her dollsomenesses?
(56) And yet Cordelia didn’t love his crazinesses.
(57) Faith didn’t love who.
(58) Joyce likes this freakedness.
(59) And yet Principal Wood totally didn’t dust her ampedness.
(60) Cordelia totally freaked.
(61) Cordelia didn’t save Buffy.
(62) The girlfriend-free slayer didn’t wig.
(63) Whoa, Angel didn’t like your wicked clues.
(64) Warren babbled.
(65) The really demony girlfriend muchly researched.
(66) Alright! Spike totally fed.
(67) That watchery guy really kicked Oz.
(68) Maybe Willow totally didn’t like that book.
(69) Don’t you think that wiggy gal really doesn’t love Angel?
(70) Joyce is wigging.
(71) Don’t you think her very violent happies don’t see Willow?
(72) Buffy really doesn’t rush.
(73) Don’t you think those freaksomenesses had been feeding?
(74) Maybe Spike pretty much doesn’t love a freaksomeness.
(75) Those vampires fed.
(76) I think Buffy muchly saw her destinies.
(77) Don’t you think Anya pretty much didn’t like her baddie minions?
(78) Don’t you think her girlfriend totally freaks?
(79) You know, her demons freak.
(80) Okay, the unusual vampires pretty much don’t love this witchy bad.
(81) Her dollsomenesses don’t vamp this really wiggy slayers.
(82) I think her messedness really loved this pretty yawnworthy slayer.
(83) Buffy was his clue.
(84) Some clue-free guys kinda didn’t rush.
(85) I think Xander didn’t freak.
(86) Alright! The slayers are watching those demons.
(87) Anya rushed.
(88) The demon totally saved her.
(89) Oz so bails.
(90) Maybe your crazinesses have pretty much been feeding.
(91) I think Cordelia totally researches.
(92) Dawn doesn’t brood.
(93) Her dollsomeness muchly likes your wickedness.
(94) Her sirey library very much didn’t like Willow.
(95) Alright! The minion-free demons very much babble.
(96) That vampire pretty much doesn’t bail.
(97) Whoa, Cordelia pretty much loved your demony zombies.
(98) I think that dollsome boyfriend doesn’t research.
(99) And yet the slaggedness kinda was not girlfriend-free.
(100) Warren stakes her bloody school.
(101) Don’t you think his yawnworthinesses pretty much were miniony?
(102) Her freaky sires pretty don’t like some demons.
(103) Dawn really slays the Master.
(104) Your slayer doesn’t avoid.
(105) The deadness feeds.
(106) Her somber covens didn’t avoid these boyfriend-free witches.
(107) Alright! The Hellmouth had really not been feeding.
(108) Okay, these slayers didn’t see her sires.
(109) These witches loved your slayy happies.
(110) The Master loved those slayers.
(111) Whoa, those covens had been feeding.
(112) Whoa, your slayers were not rushing.
(113) Buffy liked her so freaksome school.
(114) The Mayor really didn’t nap.
(115) I think Willow really sees these yawnworthy guys.
(116) I think that exactness pretty much was not kicky.
(117) Willow so doesn’t babble.
(118) I think some researchy gals are not loving that rushy werewolf.
(119) Those watchery vampires brooded.
(120) And yet the Mayor avoided some nappy witches.
(121) Okay, some hitty slayers avoid your dollsome girlfriend.
(122) Alright! Willow likes your slayers.
(123) Alright! Angel pretty much doesn’t freak.
(124) Angel doesn’t bail.
(125) Your watcher didn’t miss his slayer.
(126) The Master vamped Oz.
(127) Her crazinesses bail.
(128) Her majornesses don’t like Angel.
(129) Angel kinda doesn’t like your pretty coveny crazy.
(130) His yawnworthinesses totally was not sire-free.
(131) I think she likes Cordelia.
(132) The Master pretty much doesn’t bail.
(133) Her pretty much don’t like Cordelia.
(134) The pretty bloody girlfriend kinda bailed.
(135) Maybe an uber guy doesn’t babble.
(136) Her ampedness didn’t avoid.
(137) The Master pretty much doesn’t like her covens.
(138) And yet Willow doesn’t research.
(139) Okay, the girlfriend really doesn’t wig.
(140) Her book-free demon muchly loved that vampire.
(141) Okay, the sucky gal doesn’t nap.
(142) Her werewolf has been researching.
(143) Whoa, Angel pretty much watches Principal Wood.
(144) The girlfriend-free guy pretty much loved the boyfriend-free slayers.
(145) Willow pretty much staked her slaggedness.
(146) Oz totally didn't love that gal.
(147) Some violent slayers pretty much don’t love Cordelia.
(148) The pretty sirey slayer didn’t like Giles.
(149) I think her creepies pretty much avoided the slayy watchers.
(150) And yet these vampires kinda babble.
*Thanks are due to Kevin Sandler and D. Terence Langendoen, current and former faculty at the University of Arizona, where the research described in this paper began. This research would not have been possible without their advice and insight into Buffy the Vampire Slayer and Natural Language Processing, respectively.
Adams, Michael. 2003. Slayer Slang: A Buffy the Vampire Slayer Lexicon. New York: OUP.
___. 2004. “Don’t give me songs/Give me something to sing about: Buffy the Vampire Slayer and the Death of Style.” Paper presented at Slayage Conference on Buffy the Vampire Slayer. Nashville, TN, May 28-30.
Clark, Eve V. and Herbert J. Clark. 1979. “When nouns surface as verbs.” Language 55.4: 767-811.
Hebdige, Dick. 1979. Subculture: The Meaning of Style. London: Methuen.
Kamp, Hans. 1981. “A Theory of Truth and Semantic Representation.” In Formal Methods in the Study of Language, edited by J. Groenendijk and others, 277-322 Amsterdam: Mathematisch Centrum.