these children to have simply stored the surface form they heard around them, since these forms are obviously constructed on the basis of a previously stored, relatively correct version of the target. 7 If the children constructed these varying pronunci- ations as attempts at the same target, we can only understand what is happening if we assume that the children have stored something close to the adult phonemic pronunciation as some kind of privileged representation which they aim at each time they speak, and as the ‘‘ideal form’’ (the prototype) that they perceive when others speak. In this respect, Cognitive Phonology is quite different from standard phono- logical theory. Cognitive Phonology argues that phonemes are sounds, that is, not underspecified lists of features, but rather real, fully specified prototypical sounds. 8 We know what a ‘‘t-sound’’ sounds like, and we can hear it in both ‘two’ [t h u]and in ‘mitten’ [mi?nµ] even though the prototype occurs in neither. But representations are mental images of actual words, ‘‘spelled’’ with actual sounds. What Cognitive Phonology accepts from traditional process-oriented phonology is that the actual production normally does not match the mental image because phonemes are im- plemented in contexts and adjusted in real time to fit those contexts. Here is another significant difference from the traditional structuralist view: complementary distribution and phonetic similarity are not definitions of the pho- neme (although they may be useful tools for the linguist attempting to understand the behavior of a language he or she does not speak). As Stampe pointed out in his first works (1968, 1969), complementary distribution is a consequence of the fact that context-sensitive processes apply to underlying forms. Phonetic similarity is a direct result of the fact that processes only minimally change target sounds (al- though chains of processes, like chains of metaphorical extensions in semantics, may lead to very disparate instantiations of a single basic form, as in the case of glottal stop and voiced flap in American English for English /t/). Also note that this view of the phoneme holds that it is a basic level unit, and thus a real, mental image of a sound, not a list of distinctive features. We hear phonemes in our heads (and can generally say them out loud). Otherwise we would not be able to learn to spell, a task which those with relatively phonemic writing systems find very simple. The notion that we have images in our minds has been questioned by some researchers over the years, but in a recent work, Damasio (1999) argues for a coherent view of the notion that answers the traditional questions raised by the so-called homunculus problem. Incidentally, we should point out that this does not mean that features are not real. Donegan (2002: 8) has suggested that features ‘‘can be viewed, not as abstract categories, but as the links of motor and proprioceptive aspects of production, on the one hand, to perceptual properties (auditory, acoustic, or in acquisition, some- times visual) on the other. Such connections may be part of an inborn, ‘prewired’ mechanism like that which appears to link visual stimuli to facial gestures.’’ Pho- netic features then, are not abstract classificatory devices, but rather the mind’s method of unifying oral and aural impressions, a set of connections which are probably acquired during babbling. Jose Mompean (p.c.) has suggested that this is 620 geoff nathan analogous to Gestalt perceptions of all kinds. Just because we see objects as unified wholes does not mean that we cannot also see that they have characteristics, but we do not see them simply as a list of those characteristics, but rather as individual things. 1.5. Neutralization and Overlap Several times in the above discussion I noted that the phoneme /t/ can be pro- nounced in American English with a voiced alveolar flap [Q], as in butter, Betty, electricity, and cognitive. However, it is also the case that the phoneme /d/ can be pronounced in exactly the same way, in words such as rider, validity, and grading. That is, the categories are not completely distinct, but overlap in one area. Figure 23.1 clarifies this relationship (this diagram is based on Mompea ´ n-Gonza ´ lez’s 2004 insightful discussion of neutralization issues within the framework I am discussing). When speakers confront such instances of category overlap, in the absence of other information, they assign the sound to the closest prototype. Thus, the flap [Q] ‘‘sounds like’’ a /d/. Classic phoneme theory denied the possibility of overlapping phonemes, but that is because classic phoneme theory relied on an Aristotelian view of categorization in which a sound could not simultaneously belong to two different categories at once. Note that the operative phrase is ‘‘in the absence of other infor- mation.’’ Lexical access interacts with phonemic perception in complex ways. While a voiced flap sounds like a /d/ in the abstract, if it is in a word that is recognizable as containing a /t/ in the appropriate location (say, by virtue of the spelling, as in city, or by virtue of being related to another form of the word, such as betting being a form of the verb bet), it can also be perceived as a /t/. It is important to recognize that these perceptions are always unidirectional, however. We can think of a ‘‘t-sound’’ as occasionally sounding like a ‘‘d-sound’’ (speaking impressionistically), but no naive speaker would ever say that a ‘‘d-sound’’ sounds occasionally like a ‘‘t-sound.’’ While some phonological theories have argued that in positions of neutrali- zation a special, more abstract (or, in Cognitive Grammar terms, more schematic) sound is stored instead, this is unlikely. It would amount to the claim that there are Figure 23.1. Radial set illustrating the internal structure of the English phoneme /t/ phonology 621 actually three sounds to store: /t/, /d/, and /Q/. But native speakers do not perceive the [Q]inlatter as a third kind of sound, but rather as either a /t/ or a /d/. And again, no language has an orthography that writes morphophonemes or archi- phonemes differently from the regular phonemes of the language; this can scarcely be considered a coincidence. 2. Inventories and Prototypes A further task of phonological theory is to explain why languages select the pho- nemes that they do. Despite the fact that there might be no limit to the possibilities for phoneme inventories in the languages of the world (after all, many American structuralists appeared to believe that languages could differ in quite extraordinary ways), if we make an inventory of inventories, we find that there is in fact a quite limited range of possibilities and that languages appear to have quite restricted possibilities for phoneme inventories. In earlier work, I suggested (Nathan 1989) that this fact was due to the universality of the human vocal tract and its acoustic consequences. To put the argument somewhat briefly, human vocal behavior tends toward the universal because it is subject to constraints imposed by the structure of the anat- omy and physiology that produces the sounds in question. For example, Mac- Neilage (1998) and MacNeilage and Davis (2000) have argued that syllable struc- ture is an adaptation of chewing behavior, itself related to innate sucking behavior. The rhythmic alternation of open and closed vocal tracts provides a scaffolding on which languages have built a framework for linguistic structure in general. Principles of Gestalt psychology, which explain the organization of perception into Figure and Ground, explain why sonorants, particularly high sonority (and therefore louder) segments such as vowels, tend to be selected to serve as syllabic nuclei, while the quieter, but more perceptually discriminable consonants tend to serve as syllable margins (onsets and rhymes). A similar case can be made for the increased perceptibility of voiceless segments when contrasted with voiced vowels in the nucleus, which explains in part why Jakobson ([1941] 1968) found his im- plicational law that voiceless consonants are less marked (more common, earlier learned, etc.) than voiced ones. 9 As we look at the sounds around the world, we find that their distribution follows a typical prototype category structure, with the exception of the fact that the overall categories are universal. Thus, while a prototypical bird will vary from ecosystem to ecosystem, vocal tracts are identical in human beings, and conse- quently stop systems will show universal prototypicality structures. And just as there are nonprototypical birds, there are nonprototypical phonemes—clicks, implo- sives, nasalized vowels, and so on. Each of these kinds of sound will be under some 622 geoff nathan pressure to convert to a less marked one, but historical accidents will also lead to the creation of new versions of such odd sounds. In sum, sounds are subject to prototype effects both at the individual sound level and at the ‘‘selection’’ level—the level of creation of inventories. It is the same set of prototypicality principles at work at both levels. At the individual level, they select one among a number of alternative sounds as the one ideal instance; at the system level, they filter out altogether those suboptimal sounds that traditionally have been labeled ‘‘marked,’’ replacing palatal stops with palatal affricates, high back unrounded vowels with front, and so on. This is discussed in some detail in Nathan (1996). An analogous argument, incidentally, has been made for syllable structure as a prototypical ‘‘syntactic’’ category, subject to similar effects (so that a preference for onsets and a dispreference for codas would behave in the same way). Discussion can be found in Taylor (1995: 234–38). The reader will note that I have carefully limited my examples to phonological facts that traditional phonology would label ‘‘low level’’ processes. I have not given any examples of the more elaborate morphophonemic alternations that one nor- mally finds in an introductory phonology textbook. The reason for this is that, following Stampe’s Natural Phonology, I am not convinced that those alternations are actually phonological. That is, the examples I have discussed so far are all re- sponses by the speaker to the inherent constraints on speech production dictated by the articulatory and perceptual apparatus. As Stampe (1969, 1979, 1987) has said, they are what the speaker brings to the language. The relationships among mor- phologically related forms, on the other hand, are not derived at all from the facts of phonetics, but are rather leftovers of earlier phonetically based processes that have lost their phonetic basis. Speakers discover them by extracting schemas from sim- ilarities among forms irrespective of the articulatory or acoustic consequences of the patterns. Within the tradition of Generative Phonology, such patterns were de- scribed by what Anderson (1981) called ‘‘crazy’’ rules. Within Cognitive Grammar such patterns are not rules at all, but schemas extracted from patterns that are al- ready stored. A classic example is the English verb paradigm /i :æ:¼/ found in such verbs as sing, sink, and so on. Research has shown that this pattern is so pervasive that it is somewhat productive, and speakers tend to produce new examples when presented with novel verbs such as gring. However, this behavior is radically dif- ferent from the exceptionless insertion of aspiration on initial voiceless stops in novel words or in words borrowed from other languages. Important work on this view of relatedness of forms can be found in Bybee and Moder ( 1983) and Bybee (2001), also in the work of Jaeger (1980, 1984) on the psychological reality of the vowel shift rule. This amounts to a claim for a limited version of modularity within Cognitive Grammar. It is, however, a motivated modularity. Phonology is about entrenched motor skills, which are quite different from entrenched patterns of similarity among forms. Phonology is about what is hard or easy for a human vocal tract to perform, while morphology is about recognition of similarities among already stored variants. phonology 623 3. Usage-Based Models 3.1. The Basic Model The original impetus for this view of phonological structure was a work by Lan- gacker (1988), 10 which argued that the number of units of language was much greater than generative theories of linguistics claim and that the reduction in re- dundancy that fuels much of the theoretical apparatus of Generative Grammar is mistaken. Language is, instead, massively redundant, and much more is stored, and less generated, by rule than any current competing theory would admit. In particular, a usage-based model argues that individual instances are stored in large numbers as well as rules. In fact, Langacker argues that rules can only arise as abstractions from overtly occurring expressions (Langacker 2000: 3). Thus, rules cannot be established by speakers without prior storage of a large enough number of individual instances to permit an extraction of the regularities in the form of a schema that the instances support. The abstractions from numbers of similar examples are known as schemas, and the establishment of particular schemas constitutes schematization. An important psychological mechanism made use of in a usage-based theory is the notion of entrenchment. Entrenchment refers to the fact that some highly complex event can ‘‘coalesce into a well-rehearsed routine that is easily elicited and reliably executed. When a complex structure comes to be manipulable as a ‘pre- packaged’ assembly, no longer requiring conscious attention to its parts or their arrangement it has the status of a unit’’ (Langacker 2000: 3–4). The result of this view of linguistic organization is ‘‘that repeated applications of such processes, occurring in different combinations at many levels of organiza- tion, result in cognitive assemblies of enormous complexity. The vision that emerges is one of massive networks in which structures with varying degrees of entrench- ment, and representing different levels of abstraction, are linked together in rela- tionships of categorization, composition and symbolization’’ (Langacker 2000: 5). Further, these ‘‘linguistic categories are usually complex, and develop from prototypical structures via such processes as extension, the extraction of schemas, and the articulation of coarse-grained units into more specific ones. Complex categories are networks in which linguistic structures of any kind and size are linked. These structures—the ‘nodes’ or vertices of the network—might consist, for example, of the allophones of a phoneme’’ (Langacker, 2000: 13). Following the general principles of usage-based grammar, Bybee (2000, 2001; this volume, chapter 36), Palmer (1996), and Langacker (2000) argue that words are stored with extensive phonetic detail, and if words are pronounced in a number of different ways, each individual pronunciation is stored separately, while com- monalities (such as a possible phonemic representation) are extracted as common, underspecified schemas. Bybee says: 624 geoff nathan Phonemes, then, do not exist in the representations of words; they are not units of lexical representation. Instead phonemes are abstract patterns that emerge in the phonological organization of the lexicon (see Langacker [2000]). To the ex- tent that abstract phonetic units are grouped together into more abstract units, this is done on the basis of the phonetic implementation schemata, and is not a strict matter of complementary distribution. (Bybee 2000: 72) The image that I use to make this view concrete is that each instance is like a footstep in soft ground. Each successive footstep deepens the mark made, but there is never complete overlap—nonetheless a generalized pattern emerges after a sufficient number of steps. This leads Bybee and Langacker to argue that ‘‘phonemes’’ do not exist as units. Instead, they claim that the phenomena that phonemes are intended to de- scribe are relations of similarity among parts of phonetic strings. These relations of similarity can be captured by lexical connections and schemas just as other relations of similarity are. Complementary distribution, rather than a criterion for deciding on lexical status of a phone, is just a consequence of the fact that articulatory adjustments are conditioned by the surrounding environment (Bybee 1994; Bybee 2000: 82; this volume, chapter 36). From a somewhat different perspective, this is, of course, the same view I argued for above within the Natural Phonology–oriented discussion on the nature of categories. 3.2. The Nature of Online Processes in a Usage-Based Model In her study of English /t, d/ deletion, Bybee (2000, 2001) argues that some indi- vidual past-tense forms are stored as a whole (that is, even regular past-tense forms may not be generated through online morphological rules—the more frequent verbs may have regular past-tense forms stored as wholes; Bybee 2001: 112). The deletion of /t, d/ is also not an online process; rather, individual pronunciations of past-tense forms of verbs are stored, with different versions having different (sub- phonemic) lengths of closure. Deletion is simply an end point on a continuum of shorter and shorter alveolar gestures, and some words are stored with zero closure as an alternative pronunciation. Consequently, variable deletion of /d/ is not dele- tion at all, but rather storage of widely varying forms with a wide range of implementations: Lexical entries containing a final coronal stop are gradually accommodating to the changing input, and will gradually restructure, losing the stop entirely. Thus there are three sources of surface variation: the articulatory change is gradually reducing the gesture involved; the phonetic environment conditions whether or not the gesture can be perceived, and the lexical items themselves have differ- ent degrees of reduction. (Bybee 2000: 73) phonology 625 Here a difference arises between the embodied online view I have argued for above and the usage-based model, in that Bybee (this volume, chapter 36) argues that each ‘‘derived’’ version created by online processing is stored and that a sufficient number of these stored versions will lead to a language change. There are means of eval- uating the differences between these views and, perhaps, of attempting to reconcile them; this research is ongoing. 3.3. Morphophonemics As discussed above, Bybee and Langacker have argued that there are no morpho- phonemic rules per se, but rather that speakers extract commonalities among related forms to form higher-level schemas. Rubba (1993) has shown for Modern Aramaic, for example, that the complex patterns of alternations in a classic Semitic ‘‘triconsonantal root’’ language emerge out of natural processes of schematization from actual instances. For example, the words plaxa ‘work’ (infinitive), palxa ‘work’ (jussive), and palax ‘work’ (agentive) permit the extraction of the abstract schema /p l x/, abstracting away from the specific vowels. 4. Synthesis—Phonology as Human Action I will conclude this review by arguing that the nature of phonology as conceived of by the founders of the field, from Baudouin through Sapir to Stampe, is to account for the way that speakers of languages perceive and produce the sounds of their languages. Langacker has stated that the goal is to account for ‘‘actual distribu- tional phenomena,’’ but I would argue that accounting for the phenomena at hand requires a more active role for speakers in the production and perception of their speech. It is a fact that German speakers not only ‘‘don’t happen to have’’ voiced obstruents word-finally (as is proposed in Kumashiro 2000 as well as Langacker 2000) but that they are unable to pronounce word-final obstruents, even if they try. Further, we need to explain why native speakers of Vietnamese, who have no word- final obstruents at all, devoice those that they do manage to say while, for example, speaking English. If phonology consists simply of patterns gathered from data already acquired, Vietnamese speakers should have no opinions at all on how to pronounce final voiced obstruents. Similarly, German speakers replace initial [Z] in their English (and in borrowings) with [d Z ], even though that sound does not 626 geoff nathan exist in German either. Since neither sound exists, no possible schema could have been extracted to deal with it. Where does this behavior come from? Questions like these form the next step in the development of phonology in the Cognitive Grammar framework. Although the ‘‘embodied’’ and ‘‘usage-based’’ models agree on many areas, such as the importance of viewing phonology as embodied human action and not merely the assignment of sounds and lexical items to abstract categories, the question of the extent to which phonemes are idealized mental targets shaped by natural constraints of the vocal tract and per- ceptual system, versus the extent to which they are built up through similarity among repeated varied instances, is not yet settled and remains an area for future research. One of the major goals of Cognitive Grammar is to describe the nature of lan- guage as it actually occurs in speakers, accounting for all the complexity and re- dundancy that characterize it. The term ‘‘cognitive’’ implies that Cognitive Gram- mar’s goal is to come up with a model of language that is in accord with what is known about human cognition in general and to attempt, as far as possible, to ex- clude from the analysis entities that appear to have no independent existence out- side of other modes of cognition. Cognitive Grammar has had enormous success through its use of a few independently motivated tools, including categorization, schematization, and mechanisms of extension such as metaphor, metonymy, and image schema transformation. Lakoff and Johnson (1980) have argued that such abstract modes of mental activity as set theory are based in bodily experience with containers and small objects. Similarly, Johnson (1987) has shown that moral ques- tions of fairness are based on our bodily experience with gravity. In short, Cognitive Grammar has achieved a tremendous amount through the recognition that human beings understand the world and its complexity through reasoning and categor- izing based on basic-level categories of bodily and other reality-based experience. Unfortunately, there has been a tendency to omit the body (and its perceptual apparatus) when discussing phonology, even though this branch of linguistics ought to be the most physically grounded part of language, consisting as it does of movements of the vocal tract that produce sounds. Phonology is more than the classification and categorization of sounds that are just ‘‘there,’’ as if phonology were a kind of stamp collecting. Phonology comprises a description of the imple- mentation of motor plans by speakers and the recovery of those motor plans by listeners. Motor plans are not implemented simpliciter, but are molded and mod- ified to fit the surroundings that they find themselves in. And those modifications, while themselves motivated by facts about how the vocal tract can move in real time and by facts about what information can be retrieved by our perceptual systems, are language-specific and constitute the ‘‘phonological grammar’’ of a language. To do otherwise is to remove the body from the speaker and to surrender to the generative temptation of making language into an abstract manipulation of meaningless symbols. Baudouin and Sapir’s original view of phonology as the implementation in real time of idealized mental images, and the sympathetic reconstruction of those phonology 627 language-specific images of vocal tract gestures and the sounds associated with them, stands as the best way for Cognitive Grammar to understand human pho- nological behavior as it is illustrated in observable behavior. NOTES I am grateful to Margaret Winters and Jose ´ Antonio Mompean for their helpful, and at times challenging, comments and to David Stampe and Patricia Donegan for their inspiration. 1. There is another alternative view that should be mentioned, namely that phonemes are the invariant features found in the normally variable physical speech chain. As soon as acoustic phonetics became a relatively easily accessible science, however, it became obvious that there were no physical invariants available to be stored by speakers. 2. There is one additional strand of Cognitive Phonology that was begun in Lakoff (1993) but did not lead to subsequent work within the general Cognitive Grammar framework, although it did spawn a certain amount of further work within more gener- ative-oriented frameworks and had some influence on Optimality Theory, a model that will be discussed below. 3. Donegan and Stampe (1979) used the word innate to describe the knowledge that native speakers have of the processes of their language, but emphasized that he did not mean the Chomskyan sense, but rather simply that the knowledge was not ‘‘learned’’ like the alphabet, but rather flowed from children’s experience with acquiring control over their vocal tracts. Children learn that it is difficult to produce final voiced obstruents because they try to do so and fail, at least at first. It is in this sense that final devoicing is innate, according to Stampe. 4. There are actually other possible variants, such as a voiceless flap in outhouse, but this level of complexity should suffice. The voiced coronal flap raises other special issues that will be discussed below. 5. The F 1 Â F 2 space refers to the overall vowel space defined by the acoustic di- mensions of the first and second formants (roughly equivalent to height and backness in traditional terms.) 6. The title of Stampe’s (1968) paper is a reference to the famous newspaper article ‘‘Yes, Virginia, there is a Santa Clause.’’ Stampe’s paper was presented shortly after Gen- erative Phonology concluded that the traditional phoneme did not exist. Although the paper was never actually published, most of the arguments appear in Stampe (1987). 7. These variations only make sense, of course, if we treat each of the variants as motivated extensions from a central prototype forming a radial prototype structure of the kind we discussed above. 8. It is important to remember that I am speaking of mental sounds here—auditory/ articulatory images, not actual physical sounds. But they are real sounds, not abstract lists of features. 9. There is also a further reinforcing fact, namely that voiceless consonants are physiologically simpler to produce, since it is relatively more difficult to achieve airflow across the glottis if the upper portion of the vocal tract is closed; see Keating, Linker, and Huffman (1983) for discussion. 10. For a current discussion, see Langacker (2000). 628 geoff nathan REFERENCES Anderson, Stephen R. 1981. Why phonology isn’t ‘natural’. Linguistic Inquiry 12: 493–539. Baudouin de Courtenay, Jan. [1895] 1972. An attempt at a theory of phonetic alternations. In Edward Stankiewicz, ed. and trans., A Baudouin de Courtenay anthology: The be- ginnings of structural linguistics 144–213. Bloomington: Indiana University Press. Bloomfield, Leonard. 1933. Language. New York: Holt, Rinehart and Winston. Boersma, Paul. 1998. Functional phonology: Formalizing the interactions between articulatory and perceptual drives. The Hague: Holland Academic Graphics. Bybee, Joan L. 1994. A view of phonology from a cognitive and functional perspective. Cognitive Linguistics 5: 285–306. Bybee, Joan L. 1999. Usage-based phonology. In Michael Darnell, Edith Moravcsik, Fre- derick J. Newmeyer, Michael Noonan, and Kathleen M. Wheatley, eds., Functional- ism and formalism in linguistics, vol. 1, General Papers 211–42. Amsterdam: John Benjamins. Bybee, Joan L. 2000. The phonology of the lexicon: Evidence from lexical diffusion. In Michael Barlow and Suzanne Kemmer, eds., Usage-based models of language 65–86. Stanford, CA: CSLI Publications. Bybee, Joan L. 2001. Phonology and language use. Cambridge: Cambridge University Press. Bybee, Joan L., and Carol Lynn Moder. 1983. Morphological classes as natural categories. Language 59: 251–70. Carnie, Andrew, and Norma Mendoza-Denton. 2003. Functionalism is/n’t formalism: An interactive review of Darnell et al. (1999). Rev. of Michael Darnell, Edith Moravcsik, Frederick J. Newmeyer, Michael Noonan, and Kathleen M. Wheatley, eds., Func- tionalism and Formalism in Linguistics (Amsterdam: John Benjamins, 1999). Journal of Linguistics 39: 373–89. Chomsky, Noam, and Morris Halle. 1968. The sound pattern of English. New York: Harper and Row. Damasio, Antonio. 1999. The feeling of what happens: Body and emotion in the making of consciousness. New York: Harcourt Brace. Donegan, Patricia J. 1986. On the natural phonology of vowels. New York: Garland. Donegan, Patricia J. 2002. Normal vowel development. In Martin J. Ball and Fiona E. Gibbon, eds., Vowel Disorders 1–36. Boston: Butterworth-Heinemann. Donegan, Patricia J., and David Stampe. 1979. The study of natural phonology. In Dan Dinnsen, ed., Current approaches to phonological theory 126—73. Bloomington: Indiana University Press. Fromkin, Victoria A. 1973. Speech errors as linguistic evidence. The Hague: Mouton Fromkin, Victoria A. 1980. Errors in linguistic performance: Slips of the tongue, ear, pen and hand. New York: Academic Press Fromkin, Victoria A. 1988 The grammatical aspects of speech errors. In Frederick J. Newmeyer, ed., Linguistics: The Cambridge survey 2: 117–38. Cambridge: Cambridge University Press. Gussenhoven, Carlos, and Heike Jacobs. 1998. Understanding phonology. London: Arnold. Hayes, Bruce. 1999. Phonetically-driven phonology: The role of optimality theory and inductive grounding. In Michael Darnell, Edith Moravscik, Michael Noonan, Fre- derick Newmeyer, and Kathleen Wheatly, eds., Functionalism and formalism in linguistics, vol. 1, General papers 243–85 Amsterdam: John Benjamins. phonology 629 . these children to have simply stored the surface form they heard around them, since these forms are obviously constructed on the basis of a previously stored, relatively correct version of the. discussion of neutralization issues within the framework I am discussing). When speakers confront such instances of category overlap, in the absence of other information, they assign the sound to the. containing a /t/ in the appropriate location (say, by virtue of the spelling, as in city, or by virtue of being related to another form of the word, such as betting being a form of the verb bet),