dissenting.
I agree with Parts I and II of the Court’s opinion, and thus, like the Court, believe that the case ultimately turns on the meaning of 18 U. S. C. § 3583(e)(3) (1988 ed., Supp. V). I do not agree, however, with the Court’s interpretation of that provision. The section provides that when the conditions of supervised release are violated, the court may “revoke a term of supervised release, and require the person to serve in prison all or part of the term of supervised release without credit for time previously served on postrelease supervision.” Finding in this an authorization for imposition of additional supervised release is an act of willpower rather than of judgment.
The term “revoke” is not defined by the statute, and thus should be construed “in accordance with its ordinary or natural meaning.” FDIC v. Meyer, 510 U. S. 471, 476 (1994). As the Court recognizes, the ordinary meaning of “revoke” is “ ‘to annul by recalling or taking back.’ ” Ante, at 704 (quoting Webster’s Third New International Dictionary 1944 (1981)); see also American Heritage Dictionary 1545 (3d ed. *7161992) (defining “revoke” as “[t]o void or annul by recalling, withdrawing, or reversing; cancel; rescind”). Under this reading, the “revoked” term of supervised release is simply canceled; and since there is no authorization for a new term of supervised release to replace the one that has been revoked, additional supervised release is unavailable.
The Court is not content with this natural reading, however, and proceeds to adopt what it calls an “unconventional” reading of “revoke,” ante, at 706, as meaning “to call or summon back” without annulling, ibid.1 It thereby concludes that the revoked term of supervised release retains some effect, and thus that additional supervised release may be required after reimprisonment. The Court suggests that its abandonment of ordinary meaning is justified by the text, by congressional purpose, and by analogy to pre-Guidelines practice regarding nondetentive monitoring. None of the proffered reasons is convincing.
The Court claims textual support for its “unconventional” reading in the fact that subsection (e)(S), at issue here, uses the term “revoke,” while subsection (e)(1) uses the term “terminate.” Since, the Court reasons, the two terms should not be interpreted to have exactly the same meaning, (1) the statute must intend a “less common” meaning of “revoke,” namely, “call back,” see ante, at 706, and n. 9; and (2) this “less common” meaning authorizes the later imposition of supervised release. Each part of this two-step analysis is patently false.
*717As to the first: The usual, ordinary-English definition of “revoke” is already amply distinguishable from “terminate,” and does not have to be tortured into Old English (or actually, transliteration from Old Latin) in order to explain the choice of words. “Terminate” connotes completion rather than cancellation. See American Heritage Dictionary 1852 (3d ed. 1992) (defining “terminate” as “[t]o bring to an end or a halt” or “[t]o occur at or form the end of; conclude or finish”); Webster’s New International Dictionary 2605 (2d ed. 1942) (defining “terminate” as “[t]o put an end to; to make to cease; to end ... to form the conclusion of . . .”). Using “terminate” in subsection (e)(1) and “revoke” (in its ordinary sense) in subsection (e)(3) is not only not inexplicable; it reflects an admirably precise use of language. In subsection (e)(1), the term of supervised release is “terminated” (“brought to an end”) because termination is warranted “by the conduct of the defendant released and the interest of justice.” The supervised release is treated as fulfilled, and the sentence is complete. In subsection (e)(3), by contrast, the supervised release term is not merely brought to an end; it is annulled and treated as though it had never eristed, the defendant receiving no credit for any supervised release served. It would be hard to pick two words more clearly connoting these distinct consequences than “terminate” and “revoke.”2
*718The first step of the Court’s analysis — its inference that the use of “terminate” in subsection (e)(1) requires its alternative meaning of “revoke” in subsection (e)(3) — is also wrong because the alternative meaning that the Court posits (“to call or summon back,” without the implication of annulment, ante, at 706) is not merely (as the Court says) “less common,” ante, at 706, n. 9; in the context that is relevant here, it is utterly unheard of. One can “call or summon back” a person or thing without implication of annulment, but it is quite impossible to “call or summon back” an order or decree without that implication — which is precisely why the primary meaning of revoke has shifted from its root meaning (“call or summon back”) to the meaning that it bears in its most common context, i. e., when applied to orders or decrees (“cancel or annul”). Of course the acid test of whether a word can reasonably bear a particular meaning is whether you could use the word in that sense at a cocktail party without having people look at you funny. The Court’s assigned meaning would surely fail that test, even late in the evening. Try telling someone, “Though I do not cancel or annul my earlier action, I revoke it.” The notion that Con*719gress, by the phrase “revoke a term of supervised release,” meant “recall but not cancel a term of supervised release” is both linguistically and conceptually absurd.
The dictionary support that the Court seeks to enlist for its definition is fictitious. It is indeed the ease that both the Oxford English Dictionary and Webster’s Third New International Dictionary give as a meaning of “revoke” “to call or summon back”; but neither of them adds the fillip that is essential to the Court’s point — that the thing called back “retain vitality.” Ante, at 707. They say nothing at all about the implication of calling or summoning back — which, in the case of calling or summoning back an order or decree, is necessarily annulment.3 Further, while the dictionaries, the Court mentions do not give its chosen meaning “antiquarian reproach,” ante, at 706, n. 9, many dictionaries do. The New Shorter Oxford shows this usage as obsolete, see New Shorter Oxford English Dictionary 2583 (1993), and the previous edition of Webster’s New International shows it as rare, see Webster’s New International Dictionary 2134 (2d ed. 1942). Other dictionaries also show the Court’s chosen meaning as rare, e.g., Chambers English Dictionary 1257 (1988), as obsolete or archaic, e. g., Cassell Concise English Dictionary 1149 (1992); Funk and Wagnalls New Standard Dictionary 2104 (1957), or do not give it as a meaning at all, e. g., American Heritage Dictionary 1545 (3d ed. 1992).4
*720As for the second step of the Court’s analysis: Even if there were justification for giving "revoke” something other than its normal meaning, and even if the meaning the Court adopts were not unheard of, the latter meaning still does not provide the needed authorization for reimposition of supervised release. The statute does not say that the court may “revoke” (“call back,” as the Court would have it) only part of the term of supervised release, so there is no argument that some portion remains in place for later use. Thus, even if “revoke” means “call back,” a court would need statutory authorization to reimpose this “called back” term of supervised release. But § 3588(e)(3) provides no such authorization. The court is empowered to “revoke” the term; it is empowered to require that “all or part” of the term be served in prison; it is not empowered to reimpose “all or part” of the term as a later term of supervised release.
*721The Court opines that no authorization for further supervised release is needed, because the fact that the district court may require “all or part of the term of supervised release” to be served in prison demonstrates that the revoked term continues to have some metaphysical effect, ante, at 705-706, so that “the balance of it [can] remain effective as a term of supervised release when the reincarceration is over,” ante, at 706. It demonstrates no such thing. In allowing the district court to require that “all or part of the term of supervised release” be spent in prison, the statute simply describes the length of the permitted imprisonment by reference to that now-defunct term of supervised release. It is quite beyond me how the Court can believe that the statute “does not read” this way, ante, at 706, n. 8, and the concurrence that “[t]his ... is not what the text says,” ante, at 714. A “term of supervised release” in what might be called the substantive rather than the temporal sense — i. e., the sentence to a period of supervised release — cannot possibly be served in prison. To be in prison is not to be released. The only sense in which “all or part of the term of supervised release” can be served in prison is the temporal sense. Cf. United States v. Johnson, ante, at 57 (“To say respondent was released while still imprisoned diminishes the concept the word intends to convey”). The Court’s unrealistic reading is also undermined by the fact that § 3583(g) provides for serving in prison part of “the term of supervised release,” in spite of the fact that the term there has been “terminated,” so that even the Court would not claim it has ongoing vitality. See n. 2, supra. And finally, in concluding that the term of supervised release remains in place, the Court essentially reads the phrase “revoke a term of supervised release” out of the statute, treating the subsection as if it did no more than authorize the court to “require the person to serve in prison all or part of the term of supervised release” originally imposed, § 3583(e)(3). Of course the statute could have been drafted to say just that — allowing the court to require *722part of the term of supervised release to be served in prison, with the rest of the term remaining in place to be served on supervised release. In the text actually adopted, however, the supervised-release term is not left in place, but is explicitly “revoked.”5
Further, if one assumes, as the Court does, that a revoked term somehow “survives the . . . order of revocation,” ante, at 705, and retains effect (even without any statutory authorization for reimposition or reactivation), then it would follow that whatever part of it is not required to be served in prison is necessarily still in effect. Thus the district court would have no discretion not to require the remainder of the term to be served on supervised release. Yet the Court seems to view further supervised release as only an “option.” Ante, at 704, 713, n. 13; accord, ante, at 713-714 (Kennedy, J., concurring in part).
The Court’s confusing discussion of how § 3583(a) would produce consequences similar to those its opinion achieves— and consequences that are entirely reasonable — if § 3583(e)(3) read differently from the way it does read, ante, at 707-708, is entirely irrelevant. I do not contend that the result the Court reaches is any way remarkable, only that it is not the result called for by the statute. The Court carefully does not maintain — and it could not, for reasons I need not describe — that subsection (a) justifies imposition of post-*723revocation supervisory release given the actual text of subsection (e)(3), and nothing more is pertinent here.. Hypothetical discussion of what role § 3583(a) might play had Congress legislated differently is beside the point.
The Court next turns to questions of policy — framed as an inquiry into “congressional purpose.” Ante, at 708. Citing legislative history (although not legislative history discussing the particular subsection at issue), ante, at 709-710, the Court explains what it views as the policies Congress seeks to serve with supervised release generally, and then explains how these general policies would be undermined by reading § 3583(e)(3) as written. “Our obligation,” the Court says, “is to give effect to congressional purpose so long as the congressional language does not itself bar that result.” Ante, at 710, n. 10. I think not. Our obligation is to go as far in achieving the general congressional purpose as the text of the statute fairly prescribes — and no further. We stop where the statutory language does, and do not require explicit prohibition of our carrying the ball a few yards beyond. In any event, as read by any English speaker except one who talks of revoking a dog, the statute does “bar” the result the Court reaches here. The proper canon to govern the present case is quite simple: “[W]here, as here, the statute’s language is plain, ‘the sole function of the courts is to enforce it according to its terms,’ ” United States v. Ron Pair Enterprises, Inc., 489 U. S. 235, 241 (1989) (quoting Caminetti v. United States, 242 U. S. 470, 485 (1917)).
Perhaps there is a scrivener’s error exception to that canon, see, e. g., Holloway v. United States, 526 U. S. 1, 19, n. 2 (1999) (Scalia, J., dissenting); Green v. Bock Laundry Machine Co., 490 U. S. 504, 527-528 (1989) (Scalia, J., concurring in judgment), but the words of today’s author in another case well describe why that is inapplicable here: “This case is a far cry from the rare one where the effect of implementing the ordinary meaning of the statutory text would be patent absurdity or demonstrably at odds with the inten*724tions of its drafters.” BFP v. Resolution Trust Corporation, 511 U. S. 531, 563 (1994) (Souter, J., dissenting) (citations and internal quotation marks omitted). It would have been entirely reasonable for Congress to conclude that a prisoner who had broken the terms of supervised release seriously enough to be reincarcerated should not be trusted in that status again; and that a judge should not be tempted to impose an inappropriately short period of reimprisonment by the availability of further supervised release. Congress might also have wished to eliminate the unattractive prospect that a prisoner would go through one or even more repetitions of the violation-reimprisonment-supervised-release sequence — which is avoided by requiring the district court confronted with a violation either to leave the prisoner on supervised release (perhaps with tightened conditions and lengthened term, as § 8583(e)(2) permits) or to impose imprisonment, but not to combine the two. Because the interpretation demanded by the text is an entirely plausible one, this Court’s views of what is prudent policy are beside the point. And that is so whether those policy views are forthrightly stated as such (“[I]f any prisoner might profit from the decompression stage of supervised release, no prisoner needs it more than one who has already tried liberty and failed,” ante, at 709), or whether, to give an interpretive odor to the opinion, they are recast as policies that it “seems very unlikely” for Congress to have intended (“Congress . . . seems very unlikely to have meant to compel the courts to wash their hands of the worst cases at the end of reimprisonment,” ante, at 709-710).
Finally, the Court appeals to pre-Guidelines practice with regard to nondetentive monitoring. But this cannot cure the lack of statutory authorization for additional supervised release. Even if the language of § 3583(e)(3) were ambiguous (which it is not), that history would be of little relevance, since the Sentencing Reform Act’s adoption of supervised release was meant to make a significant break with prior *725practice, see Mistretta v. United States, 488 U. S. 361, 366 (1989) (describing the Act’s “sweeping reforms”); Gozlon-Peretz v. United States, 498 U. S. 395, 407 (1991) (“Supervised release is a unique method of postconfinement supervision invented by the Congress for a series of sentencing reforms”).6 The Court’s effort to equate parole and supervised release, ante, at 710-712, is unpersuasive. Unlike parole, which replaced a portion of a defendant’s prison sentence, supervised release is a separate term imposed at the time of initial sentencing. Compare 18 U. S. C. § 3583(a) with 18 U. S. C. §§ 4205(a), 4206 (1982 ed.) (repealed); see also USSG ch. 7, pt. A, intro, comment. 2(b). This distinction has important consequences for the present question, since when parole was “revoked” (unlike when supervised release is revoked), there was no need to impose a new term of imprisonment; the term currently being served (on parole) was still in place. Similarly, there was no occasion to impose a new term of parole, since the possibility of parole was inherent in the remaining sentence. See 18 U. S. C. § 4205(a) (1982 ed.) (“Whenever confined and serving a definite term *726or terms of more than one year, a prisoner shall be eligible for release on parole after serving one-third of such term . . The question whether further supervised release may be required after revocation of supervised release is so entirely different from the question whether further parole may be accorded after revocation of parole, that the Court’s appeal to the parole practice demonstrates nothing except the dire scarcity of arguments available to support its conclusion.7
*727* * *
This is not an important case, since it deals with the interpretation of a statute that has been amended to eliminate, for the future, the issue we today resolve. But an institution that is careless in small things is more likely to be careless in large ones; and an institution that is willful in small things is almost certain to be willful in large ones. The fact that nothing but the Court’s views of policy and “congressional purpose” supports today’s judgment is a matter of great concern, if only because of what it tells district and circuit judges. The overwhelming majority of the Courts of Appeals — 9 out of 11 — notwithstanding what they might have viewed as the more desirable policy arrangement, reached the result unambiguously demanded by the statutory text. See ante, at 698, n. 2. Today’s decision invites them to return to headier days of not-too-yore, when laws meant what judges knew they ought to mean. I dissent.
Describing the Court’s reading as “unconventional” makes it sound perfectly O. K. There are, after all, unconventional houses, unconventional hairdos, even unconventional batting stances, all of which are fine. Houses, hairdos, and batting stances, however, have an independent existence apart from convention, whereas words are nothing but a convention— particular sounds which by agreement represent particular concepts, and (in the case of most written languages) particular symbols which by agreement represent particular sounds. Thus, when the Court admits that it is giving the word “revoke” an “unconventional” meaning, it says that it is choosing to ignore the word “revoke.”
The Court is correct, ante, at 705, n. 7, that my suggested explanation of the difference between “terminate” and “revoke” does not comport with the use of “terminate” in § 3583(g). But the use of the term in that subsection also contradicts the Court’s explanation of the difference between the two terms — viz., that “terminate,” unlike in its view “revoke,” “eon-elude[s] any possibility of supervised release later,” ante, at 704. For the Court evidently believes (contrary to the use of “terminate” in § 3583(g)) that further supervised release is available when a supervisee is reimpris-oned for possession of a controlled substance. It would be “fundamentally contrary” to the congressional scheme, the Court asserts, if supervised release following reimprisonment were not available for “one who has already tried liberty and failed,” ante, at 709. But the use of “terminate” in *718§ 3588(g) prescribes just that. Further, § 3583(g) undermines the Court’s argument that because § 3583(e)(3) authorizes the court to “revoke a term of supervised release” and then to require “all or part of the term” to be served in prison, the revoked term must retain some metaphysical vitality. See ante, at 705-706. This is so because § 3583(g) provides that the court shall “terminate the term of supervised release” (hence extinguishing it even in the Court’s view), and yet goes on to provide that the court shall require the defendant to serve at least one-third of “the term of supervised release” in prison. See infra, at 721. So on either the Court’s interpretation of the difference between “terminate” and “revoke” or on mine, the use of “terminate” in § 3583(g) was a mistake — which is why Congress has since amended it to read “revoke.” See § 110505,108 Stat. 2017. See also Brief for United States 25, n. 20 ("Congress apprehended that the term ‘terminate’ was inappropriate [in § 3583(g)]”). If we both concede it was a mistake, that leaves my explanation of the difference between “terminate” in § 3583(e)(1) and “revoke” in § 3583(e)(3) uncontradicted.
As the Court suggests in its quotation of Webster’s Third’s definition of “RECALL,” see ante, at 706, the annulment may be only temporary (a “suspension”); but that is so only if there is some authority for repromúlgation after the revocation — which leaves the Court no further along than it was before it dipped into the more obscure meanings of “revoke”: it must identify some authority to reimpose supervised release. This blends into the next point made in text.
Whether one attributes any currency to “revoke” in the sense of “call back” depends, I think, on whether one counts as current usage figurative usage. The OED, while not showing the meaning “to call back” as obsolete, does indicate that its current usage is “chiefly fig[urative].” 13 Ox*720ford English Dictionary 838 (2d ed. 1989) (OED). Just as current usage would allow one to say that "the emperor called back his decree,” so also it would allow one to say that the emperor “revoked” his decree in that figurative sense of “calling it back” — i e., in the sense of canceling it. It. is assuredly not current usage, however — I think not even rare current usage — to use “revoke” to connote a literal calling back. (“Since my bird dog was ranging too far afield, I revoked him.”)
The Court chastises this example, suggesting that only a tippling hunter would “revoke” his bird dog, as “dogs cannot be revoked, even though sentencing orders can be.” Ante, at 707, n. 9. I could not agree more. However, the definition the Court employs (“call back” without the implication of cancellation) envisions that dogs can be revoked — thus illustrating its obscurity. The OED definition on which the Court relies, see ante, at 706, n. 9, defines “revoke” as “to recall; to call or summon back ... an animal or thing.” 13 OED 838 (2d ed. 1989). The first example it gives of this usage is as follows: “These hounds ... being acquainted with their masters watchwordes, eyther in revoking or imboldening them to serve the game.” Ibid. Of course the Court’s “not unheard of” usage, ante, at 706, is not limited to recalling dogs — oxen can be revoked as well, as the OED’s third example illustrates: “Ye must revoke The patient Oxe unto the Yoke.” 13 OED 838.
The concurrence adjusts for that inconvenient fact by simply changing the object of the verb, concluding that “after the right to be on supervised release has been revoked there is yet an unexpired term of supervised release that can be allocated ... in whole or in part to confinement and to release . . . Ante, at 713 (Kennedy, J., concurring in part) (emphasis added). The statute, however, does not revoke “the right to be on supervised release”; it revokes the “term of supervised release” itself, see § 3583(e)(3), which is utterly incompatible with the notion that the term remains in place. Switching the object of “revoke” is no fair in itself, and it leaves the provision entirely redundant, since revoking “the right to be on supervised release” adds nothing to “requir[ing] the person to serve in prison all or part of the term,” § 3583(e)(3).
United States Sentencing Commission, Guidelines Manual eh. 1, pt. A, intro, comment. S (Nov. 1998) (USSG), is not to the contrary. The Court quotes the comment for the broad proposition that “[t]he Sentencing Guidelines, after all, ‘represent an approach that begins with, and builds upon,’ pre-Guidelines law.” Ante, at 710. The comment itself, however, makes the much more narrow point that data on sentences imposed pre-Guidelines were used as a “starting point” in devising sentencing ranges under the Guidelines. The sentence from which the Court quotes states: “Despite . . . policy-oriented departures from pre-guidelines practice, the guidelines represent an approach that begins with, and builds upon, empirical data.” USSG ch. 1, pt. A, intro, comment. 3. This sheds no light on the extent to which prior practice in matters other than length of sentence underlay the Guidelines, much less on the extent to which such prior practice is a meaningful guide to statutory interpretation in general — and even less to statutory interpretation pertaining to supervised release, which the Guidelines elsewhere refer to as “a new form of post-imprisonment supervision created by the Sentencing Reform Act,” id., eh. 7, pt. A, intro, comment. 2(b).
The Court also appeals to pre-Guidelines practice regarding probation and special parole. Ante, at 711-712, n. 11. The pre-Guidelines probation practice is altogether inapt, since the governing statute explicitly provided for resentencing, after violation, and specifically allowed the court to “impose any sentence which might originally have been imposed.” 18 U. S. C. §3653 (1982 ed.) (repealed). This makes it quite impossible for probation practice to support the Court’s “broader point that a court’s powers at the original sentencing are the baseline from which powers at resentencing are determined,” ante, at 711, n. 11; all it proves is that they are the baseline where the statute says so. Indeed, the fact that the statute found it necessary to say so tends to contradict the Court’s position.
Special parole, while more akin to supervised release than either parole or probation, hardly provides dear support for the Court’s reading of § 3583(e)(3). In fact, the majority of Courts of Appeals have read the relevant statute regarding special parole, 21 U. S. C. § 841(c) (1982 ed.) (repealed), as not allowing reimposition of special parole in circumstances analogous to those at issue here. See Manso v. Federal Detention Center, 182 F. 3d 814, 817 (CA11 1999) (citing eases). The Court’s reliance on the Parole Commission’s 1977 interpretation of the special parole statute, see 28 CFR § 2.57(e) (1999), is misplaced. The principle that Congress is presumed to legislate in light of existing administrative interpretations does not stretch to cover an administrative interpretation of a statute dealing with a different subject, of recent vintage, and unsupported by judicial opinion. Cf. Bragdon v. Abbott, 524 U. S, 624, 645 (1998) (repetition of existing statutory language assumed to incorporate “uniform body of administrative and judicial precedent” that had “settled the meaning” of existing provision); Haig v. Agee, 453 U. S. 280, 297 (1981) (assuming congressional awareness of “longstanding administrative construction”). Further, some courts have found it unclear whether the Parole Commission’s regulation itself envisions reimposition of special parole. See, e. g., Fowler v. United States Parole Commission, 94 F. 3d 835, 841 (CA3 1996).