From mcovingt@aisun3.ai.uga.edu Fri Dec 18 14:54:16 1992 Return-Path: Received: from server.uga.edu by aisun1.ai.uga.edu (4.1) id AA06823; Fri, 18 Dec 92 14:54:16 EST Received: by server.uga.edu (5.57/Ultrix3.0-C) id AA19210; Fri, 18 Dec 92 14:54:10 -0500 Received: by aisun3.ai.uga.edu (4.1) id AA12073; Fri, 18 Dec 92 14:54:09 EST From: mcovingt@aisun3.ai.uga.edu (Michael Covington) Message-Id: <9212181954.AA12073@aisun3.ai.uga.edu> Subject: Test message - Please ignore To: dg@ai.uga.edu Date: Fri, 18 Dec 92 14:54:08 EST X-Mailer: ELM [version 2.3 PL11] This is a test mailing to just a few of the participants, plus a number of known invalid addresses. If no problems arise, DG will be opened for use soon. ------------------------------------------------------------------------ From mcovingt@aisun3.ai.uga.edu Fri Dec 18 14:59:16 1992 Return-Path: Received: from server.uga.edu by aisun1.ai.uga.edu (4.1) id AA06880; Fri, 18 Dec 92 14:59:16 EST Received: by server.uga.edu (5.57/Ultrix3.0-C) id AA19233; Fri, 18 Dec 92 14:59:15 -0500 Received: by aisun3.ai.uga.edu (4.1) id AA12137; Fri, 18 Dec 92 14:59:14 EST From: mcovingt@aisun3.ai.uga.edu (Michael Covington) Message-Id: <9212181959.AA12137@aisun3.ai.uga.edu> Subject: Second test message - Please ignore To: dg@ai.uga.edu Date: Fri, 18 Dec 92 14:59:13 EST X-Mailer: ELM [version 2.3 PL11] This is the second test message. DG is not yet open for business. -- :- Michael A. Covington internet mcovingt@uga.cc.uga.edu : Verbum caro :- Artificial Intelligence Programs phone 706 542-0358 : factum est :- The University of Georgia fax 706 542-0349 : et habitavit :- Athens, Georgia 30602-7415 U.S.A. amateur radio N4TMI : in nobis... --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. Log available by anon. FTP from ai.uga.edu, directory /ai.natural.language. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From mcovingt@aisun3.ai.uga.edu Fri Dec 18 22:17:12 1992 Return-Path: Received: from server.uga.edu by aisun1.ai.uga.edu (4.1) id AA08592; Fri, 18 Dec 92 22:17:12 EST Received: by server.uga.edu (5.57/Ultrix3.0-C) id AA20075; Fri, 18 Dec 92 22:17:11 -0500 Received: by aisun3.ai.uga.edu (4.1) id AA12857; Fri, 18 Dec 92 22:17:10 EST From: mcovingt@aisun3.ai.uga.edu (Michael Covington) Message-Id: <9212190317.AA12857@aisun3.ai.uga.edu> Subject: Test mailing - Please ignore To: dg@ai.uga.edu Date: Fri, 18 Dec 92 22:17:09 EST X-Mailer: ELM [version 2.3 PL11] This is a test of the dependency grammar mailing list. -- :- Michael A. Covington internet mcovingt@uga.cc.uga.edu : --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. Log available by anon. FTP from ai.uga.edu, directory /ai.natural.language. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From mcovingt@aisun1.ai.uga.edu Sun Dec 20 22:47:33 1992 Received: from aisun1.ai.uga.edu by aisun3.ai.uga.edu (4.1) id AA14659; Sun, 20 Dec 92 22:47:32 EST Received: by aisun1.ai.uga.edu (4.1) id AA12411; Sun, 20 Dec 92 22:47:31 EST Return-Path: Received: by aisun1.ai.uga.edu (4.1) id AA12407; Sun, 20 Dec 92 22:47:29 EST From: mcovingt@aisun1.ai.uga.edu (Michael Covington) Message-Id: <9212210347.AA12407@aisun1.ai.uga.edu> Subject: ** New dependency grammar mailing list ** To: dg@ai.uga.edu Date: Sun, 20 Dec 92 22:47:28 EST X-Mailer: ELM [version 2.3 PL11] Status: OR A new mailing list has been set up for discussion of dependency grammar and related issues. Basic information follows. Let me know if I can assist you in any way. - Michael Covington, mcovingt@ai.uga.edu (If you received this message directly, you are already a subscriber.) ------------------------------------------------------------------------ ------------------------------------------------------------------------- Basic information about mailing list DG ------------------------------------------------------------------------- The mailing list "DG" is for discussion of dependency grammar and related matters. It was set up by Michael Covington at the suggestion of Richard Hudson. DG is unmoderated. Any message mailed to DG@AI.UGA.EDU will automatically be distributed to all participants. When a message arrives from DG, you have two choices: - You can reply directly to the sender (that's what happens if you hit "reply" or "r" in most mail-reading programs); - You can mail a reply to DG@AI.UGA.EDU to have it distributed to all participants. (This is usually preferable.) (Every mailing from DG ends with a footnote summarizing these two options.) To subscribe to DG, or for other assistance, email MCOVINGT@AI.UGA.EDU. Copies of all mailings sent through DG are available by anonymous FTP from AI.UGA.EDU, directory /ai.natural.language, file dg.log. ------------------------------------------------------------------------- List of subscribers as of December 20, 1992 Michael Covington Richard Hudson sag@csli.stanford.edu guy@cogsci.ed.ac.uk laurie.bauer@vuw.ac.nz norman@logcam.co.uk hajicova@cspguk11.bitnet phopkins@sol.uvic.ca uclyrah@ucl.ac.uk rcj@caen.engin.umich.edu joshi@linc.cis.upenn.edu duda@zisw.wtza-berlin.dbp.de "Mark Liberman" lobin@lili1.uni-bielefeld.de jmiller@cogsci.ed.ac.uk davidm@cogsci.edinburgh.ac.uk "bneving@bbn.com" kpuba@hujivm1.bitnet panevova@cspguk11.bitnet pollard@ling.ohio-state.edu rambow@unagi.cis.upenn.edu rohrer@adler.ims.uni-stuttgart.de A.Rosta@ucl.ac.uk sgall@cspguk11.bitnet hls@ccl.umist.ac.uk t042270@uhccmvs.bitnet voutilainen@cc.helsinki.fi Peter Hellwig manyman@cc.helsinki.fi fddeane@ucf1vm.cc.ucf.edu uclynsg@ucl.ac.uk "Michael C. McCord (Phone 914-784-7808)" "Dan Maxwell" "Richard Sharman, IBM UK" "Henning Lobin" "Ted Briscoe" --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From mcovingt@aisun1.ai.uga.edu Sun Dec 20 22:51:44 1992 Received: from aisun1.ai.uga.edu by aisun3.ai.uga.edu (4.1) id AA14665; Sun, 20 Dec 92 22:51:43 EST Received: by aisun1.ai.uga.edu (4.1) id AA12426; Sun, 20 Dec 92 22:51:41 EST Return-Path: Received: by aisun1.ai.uga.edu (4.1) id AA12422; Sun, 20 Dec 92 22:51:40 EST From: mcovingt@aisun1.ai.uga.edu (Michael Covington) Message-Id: <9212210351.AA12422@aisun1.ai.uga.edu> Subject: Dependency grammar - Some questions To: dg@ai.uga.edu Date: Sun, 20 Dec 92 22:51:40 EST X-Mailer: ELM [version 2.3 PL11] Status: OR Here are some topics that may help us get the discussion started... (1) What, if anything, does dependency grammar (DG) have to do with Chomsky's new "minimalist" theory? (2) In recent GB theory there seems to be a determination to define all grammatical relations in terms of tree structure. To what extent has this led to the positing of tree structure that is otherwise unmotivated? (Insofar as this has happened, it's an argument that dependency grammar -- which defines the grammatical relations directly -- would be simpler.) (3) Bechraoui has pointed out that there are really two distinctions between constituency and dependency theories: (a) Do you use dependency arcs or constituency trees to represent structure? (b) Do you take structure as basic or relations and functions as basic? (4) What is the dependency analog of ID/LP rules? (5) What dependency parsing algorithms are there? (I know that Norman Fraser is doing research on this now.) Looking forward to hearing from you... -- :- Michael A. Covington internet mcovingt@uga.cc.uga.edu : Verbum caro :- Artificial Intelligence Programs phone 706 542-0358 : factum est :- The University of Georgia fax 706 542-0349 : et habitavit :- Athens, Georgia 30602-7415 U.S.A. amateur radio N4TMI : in nobis... --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From mcovingt@aisun3.ai.uga.edu Sun Dec 20 23:49:00 1992 Return-Path: Received: from server.uga.edu by aisun1.ai.uga.edu (4.1) id AA12614; Sun, 20 Dec 92 23:49:00 EST Received: by server.uga.edu (5.57/Ultrix3.0-C) id AA23110; Sun, 20 Dec 92 23:48:59 -0500 Received: by aisun3.ai.uga.edu (4.1) id AA14792; Sun, 20 Dec 92 23:48:58 EST From: mcovingt@aisun3.ai.uga.edu (Michael Covington) Message-Id: <9212210448.AA14792@aisun3.ai.uga.edu> Subject: Brief scheduled outage To: dg@ai.uga.edu Date: Sun, 20 Dec 92 23:48:57 EST X-Mailer: ELM [version 2.3 PL11] The machine that services the DG mailing list will be down for about 8 hours during the evening/night of December 21-22. --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From @mail-a.bcc.ac.uk:uclyrah@ucl.ac.uk Mon Dec 21 04:17:50 1992 Return-Path: <@mail-a.bcc.ac.uk:uclyrah@ucl.ac.uk> Received: from sun2.nsfnet-relay.ac.uk by aisun1.ai.uga.edu (4.1) id AA12940; Mon, 21 Dec 92 04:17:50 EST Via: uk.ac.bcc.mail-a; Mon, 21 Dec 1992 09:17:32 +0000 Received: from link-1.ts.bcc.ac.uk by mail-a.bcc.ac.uk with SMTP (PP) id <00558-0@mail-a.bcc.ac.uk>; Mon, 21 Dec 1992 09:13:17 +0000 Received: by link-1.ts.bcc.ac.uk (AIX 3.2/UCB 5.64/4.03) id AA28708; Mon, 21 Dec 1992 09:13:15 GMT Message-Id: <9212210913.AA28708@link-1.ts.bcc.ac.uk> To: dg@ai.uga.edu Subject: DG and GB Date: Mon, 21 Dec 92 09:13:15 +0000 From: "R.HudsonIBM-850" 21 December 1992 Can I pick up on Michael Covington's second question, about the possible redundancy of tree structure in GB. I've been struck recently by the similarities between GB `government' and DG `dependency'. (Notice that GB government is NOT the same as DG government, because A can GB-govern B without having any effect on B's inflectional category, i.e. without DG-governing it.) Roughly speaking, if A depends on B then B GB-governs A. The most important difference between GB-government and dependency seems to me to involve adjuncts/modifiers, which aren't GB-governed but do depend on other words. Another difference stems from the role of phrases in GB, so if A depends on B, then B GB-governs not only B but also the phrase of which it's head. A third difference is that if A GB-governs B, then A also GB-governs B's specifier. (This is a very peculiar claim, in my opinion, which seems to create at least as many problems as it solves.) Any of these differences between GB and DG deserves some attention as a possible way of distinguishing between GB and DG, but they shouldn't obscure the basic similarity between GB- government and dependency. Since GB-government is an innovation of the 80's in GB, it is clearly a major shift in the direction of DG. Now if you push behind the notation, one consequence of the recognition of GB-government is that GB already *generates* something remarkably similar to a dependency structure along with the various X- bar structures, a `government-structure'. Admittedly this structure is derived from the constituent structure, but it's formally available thanks to the definition of GB-government. And more importantly, it's referred to directly in a lot of principles of the theory, and seems to be playing an increasingly important part as the theory develops. Is it just a matter of time before someone realises that it would be possible to reverse the relations between government-structure and constituent-structure? Then the government-structures will be generated directly (e.g. as projections of lexical valencies), and constituent structures will be derivable from them. When that point is reached, GB will in effect have become (a very sophisticated version of) DG. Dick Hudson Dept of Phonetics and Linguistics, University College London, Gower Street, London WC1E 6BT (071) 387 7050 ext 3152 --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From @uga.cc.uga.edu:PANEVOVA@CSPGUK11.BITNET Mon Dec 21 08:32:04 1992 Return-Path: <@uga.cc.uga.edu:PANEVOVA@CSPGUK11.BITNET> Received: from uga.cc.uga.edu by aisun1.ai.uga.edu (4.1) id AA13162; Mon, 21 Dec 92 08:32:04 EST Message-Id: <9212211332.AA13162@aisun1.ai.uga.edu> Received: from UGA.CC.UGA.EDU by uga.cc.uga.edu (IBM VM SMTP V2R2) with BSMTP id 1320; Mon, 21 Dec 92 08:31:37 EST Received: from CSPGUK11.BITNET by UGA.CC.UGA.EDU (Mailer R2.08 PTF008) with BSMTP id 9591; Mon, 21 Dec 92 08:31:37 EST Received: from CSPGUK11 (PANEVOVA) by CSPGUK11.BITNET (Mailer R2.07) with BSMTP id 9020; Mon, 21 Dec 92 12:26:23 CET Date: Mon, 21 Dec 92 12:22:58 CET From: PANEVOVA%CSPGUK11.BITNET@uga.cc.uga.edu To: dg@ai.uga.edu The dependency mailing list caused great troubles in our e-mail disks. Now it seems to be O.K.(after clearing of our disks during last week). I hope now the connection will function well. Marry Christmas Jarmila Panevova --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From ltb!maxwell@relay.nluug.nl Tue Dec 22 09:36:01 1992 Received: from sun4nl.nluug.nl by aisun1.ai.uga.edu (4.1) id AA00877; Tue, 22 Dec 92 09:36:01 EST Received: from ltb by sun4nl.nluug.nl via EUnet id AA17205 (5.65b/CWI-3.3); Tue, 22 Dec 1992 15:35:56 +0100 Return-Path: X-Organisation: BSO/Language Technology BV X-Address: P.O. Box 543 X-City: NL-3740 AM Baarn X-Country: The Netherlands X-Phone: (+31) 2154 84411 X-Fax: (+31) 2154 16781 Received: from lt5 by ltb.ltb.bso.nl (5.65b/LT-1.0) id AA14001; Tue, 22 Dec 92 15:11:47 +0100 Received: by lt5.ltb.bso.nl (5.65b/LT-1.0) id AA20349; Tue, 22 Dec 92 15:11:45 +0100 From: maxwell@ltb.bso.nl Message-Id: <9212221411.AA20349@lt5.ltb.bso.nl> Subject: questions To: dg@ai.uga.edu Date: Tue, 22 Dec 92 15:11:43 MET X-Mailer: ELM [version 2.3 PL11] Michael raises some questions about dependency grammar. I plead ignorance on (1), (2), and (5), but feel qualified to provide some answers to (3) and (4). "(3) Bechraoui has pointed out that there are really two distinctions between constituency and dependency theories: (a) Do you use dependency arcs or constituency trees to represent structure? I would definitely not use constituency trees, since then I think you would be doing some kind of constituency grammar. Actually some kinds of constituency grammar are not that far from dependency grammar. Carl Pollard was quoted as saying that HPSG is almost a dependency grammar, and this is to some extent true for any constituent grammar which makes use of X-bar theory. As for dependency arcs, as used in Dick Hudson's word grammar, for example, these appear to me to be a notational variant of what I'm used to working with, namely dependency trees. At least it is not obvious to me what the empirical differences between them are. (b) Do you take structure as basic or relations and functions as basic?" I think you need a combination of both: arcs or branches to indicate structure, but usually with some kind of label to indicate function. There is then less structure than in constituency trees, but somewhat more function marking. (4) What is the dependency analog of ID/LP rules? linearization rules which are based either on the governor-dependent relationship or the sibling relationship between nodes with the same governor. -- This is in constrast to ID/LP rules which make exclusive use of sibling relationships in GPSG, but in HPSG there is also a second kind of LP rule, based on a grammatical relations hierarchy. Dan Maxwell@ltb.bso.nl --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From @mail-a.bcc.ac.uk:uclyrah@ucl.ac.uk Tue Dec 22 16:29:31 1992 Return-Path: <@mail-a.bcc.ac.uk:uclyrah@ucl.ac.uk> Received: from sun2.nsfnet-relay.ac.uk by aisun1.ai.uga.edu (4.1) id AA01767; Tue, 22 Dec 92 16:29:31 EST Via: uk.ac.bcc.mail-a; Tue, 22 Dec 1992 21:29:26 +0000 Received: from link-1.ts.bcc.ac.uk by mail-a.bcc.ac.uk with SMTP (PP) id <06680-0@mail-a.bcc.ac.uk>; Tue, 22 Dec 1992 21:29:23 +0000 Received: by link-1.ts.bcc.ac.uk (AIX 3.2/UCB 5.64/4.03) id AA17024; Tue, 22 Dec 1992 21:29:21 GMT Message-Id: <9212222129.AA17024@link-1.ts.bcc.ac.uk> To: dg@ai.uga.edu Subject: Notation Date: Tue, 22 Dec 92 21:29:21 +0000 From: "R.HudsonIBM-850" Dan Maxwell raises the question, inter alia, of notation. Is there any empirical difference between dependency arcs and dependency trees? Since he mentions me as an arc-user, I should react. a. I agree that in principle there's no difference, provided we make the usual assumptions about possible dependency structures (and in particular, that no word has more than one head). E.g. \ : \ : \ ------> --------> : \ book about linguistics : : : \ or: : : \ = : : : : \ /---\ /------- : : \ | | | | : : : | V | V book about linguistics book about linguistics (Graphic substance leaves something to be desired in both notations when you're aiming at a simple DOS file!) b. However the picture changes (literally) if, like me, you think there are cases where more complex structures are needed, involving double-headed words or even inter-dependent words. Then you obviously can't use the vertical dimension to show dependency because if A and B are inter-dependent they'd each need to be higher than the other. I think this is what's needed in relative clauses, where the relative pronoun or the antecedent noun is both the head of the relative verb, and also one of its dependents. Arcs or flat arrows can show such relations, but I don't think dependency trees can. /---------- | /--------\ | --------> || /--\ || <-------- || | ||| <----- |V V ||V book I bought = book I bought The empirical question, then, is whether such analyses really are needed in order to deal with the data (in this case, relative clauses). If you're convinced they aren't, then you'll prefer dependency trees, as a more restrictive notation. That's why I do believe there's an empirical difference between the notations. Dick Hudson Dept of Phonetics and Linguistics, University College London, Gower Street, London WC1E 6BT (071) 387 7050 ext 3152 --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From mcovingt@aisun3.ai.uga.edu Mon Jan 4 11:51:40 1993 Return-Path: Received: from server.uga.edu by aisun1.ai.uga.edu (4.1) id AA24506; Mon, 4 Jan 93 11:51:40 EST Received: by server.uga.edu (5.57/Ultrix3.0-C) id AA13240; Mon, 4 Jan 93 11:51:37 -0500 Received: by aisun3.ai.uga.edu (4.1) id AA03968; Mon, 4 Jan 93 11:51:36 EST From: mcovingt@aisun3.ai.uga.edu (Michael Covington) Message-Id: <9301041651.AA03968@aisun3.ai.uga.edu> Subject: New year greetings To: dg@ai.uga.edu Date: Mon, 4 Jan 93 11:51:36 EST X-Mailer: ELM [version 2.3 PL11] There are now a total of 73 subscribers to DG, the dependency grammar mailing list. Now let's get some discussion going... Happy new year to all. -- :- Michael A. Covington internet mcovingt@uga.cc.uga.edu : ***** :- Artificial Intelligence Programs phone 706 542-0358 : ********* :- The University of Georgia fax 706 542-0349 : * * * :- Athens, Georgia 30602-7415 U.S.A. amateur radio N4TMI : ** *** ** --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From ellalain@nuscc.nus.sg Tue Jan 5 04:10:18 1993 Return-Path: Received: from nuscc.nus.sg by aisun1.ai.uga.edu (4.1) id AA28006; Tue, 5 Jan 93 04:10:18 EST Received: by nuscc.nus.sg (5.65/1.34) id AA01534; Tue, 5 Jan 93 17:09:34 +0800 Date: Tue, 5 Jan 93 17:09:34 +0800 From: ellalain@nuscc.nus.sg (Alain Polguere) Message-Id: <9301050909.AA01534@nuscc.nus.sg> To: Subject: Translation of Tesniere Dear DGs, It seems to me that there is no English translation of Tesniere's _Elements_de_Syntaxe_Structurale_. Is that right? I would love to have one (for teaching purposes). AP --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From @mail-a.bcc.ac.uk:uclyrah@ucl.ac.uk Wed Jan 6 05:38:05 1993 Return-Path: <@mail-a.bcc.ac.uk:uclyrah@ucl.ac.uk> Received: from sun2.nsfnet-relay.ac.uk by aisun1.ai.uga.edu (4.1) id AA02622; Wed, 6 Jan 93 05:38:05 EST Via: uk.ac.bcc.mail-a; Wed, 6 Jan 1993 10:36:42 +0000 Received: from link-1.ts.bcc.ac.uk by mail-a.bcc.ac.uk with SMTP (PP) id <14146-0@mail-a.bcc.ac.uk>; Wed, 6 Jan 1993 10:36:39 +0000 Received: by link-1.ts.bcc.ac.uk (AIX 3.2/UCB 5.64/4.03) id AA44077; Wed, 6 Jan 1993 10:36:37 GMT Message-Id: <9301061036.AA44077@link-1.ts.bcc.ac.uk> To: dg@ai.uga.edu Subject: Translation of Tesniere Date: Wed, 06 Jan 93 10:36:37 +0000 From: "R.HudsonIBM-850" Alain Polguere seems to me to be right - there is no English translation of Tesniere, and there ought to be one (perhaps abbreviated?). I have contacts with publishers. Can anyone suggest a good person to do such a translation? The ideal person would be an experienced English-French translator who could also relate Tesniere's work to more recent developments in English-language syntactic theory. Dick Hudson Dept of Phonetics and Linguistics, University College London, Gower Street, London WC1E 6BT (071) 387 7050 ext 3152 --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From @mail-a.bcc.ac.uk:uclyrah@ucl.ac.uk Wed Jan 6 06:09:27 1993 Return-Path: <@mail-a.bcc.ac.uk:uclyrah@ucl.ac.uk> Received: from sun2.nsfnet-relay.ac.uk by aisun1.ai.uga.edu (4.1) id AA02654; Wed, 6 Jan 93 06:09:27 EST Via: uk.ac.bcc.mail-a; Wed, 6 Jan 1993 11:08:52 +0000 Received: from link-1.ts.bcc.ac.uk by mail-a.bcc.ac.uk with SMTP (PP) id <14621-0@mail-a.bcc.ac.uk>; Wed, 6 Jan 1993 11:08:48 +0000 Received: by link-1.ts.bcc.ac.uk (AIX 3.2/UCB 5.64/4.03) id AA98055; Wed, 6 Jan 1993 11:08:47 GMT Message-Id: <9301061108.AA98055@link-1.ts.bcc.ac.uk> To: dg@ai.uga.edu Subject: Dependency statistics Date: Wed, 06 Jan 93 11:08:46 +0000 From: "R.HudsonIBM-850" I've just been doing some work on the statistics of dependencies in texts, which turn out to be rather interesting. After analysing 5 different very short texts (each about 180 words +/- 60), I'm prepared to hazard the following generalisations about any written English text of at least this length: - 71% (+/- 8%) of all dependents will follow their heads; - 59% (+/- 5%) of dependents will be complements; - 8% (+/- 4% of dependents will be subjects; - 67% (+/- 3% !! ) of dependents will be next to their heads; - 79% (+/- 4% !! ) of dependents will be no more than one word away from their heads; - at no point in the chain of words will there be more than 6 `open' dependencies, i.e. dependencies for which a head or dependent is expected but not yet found on a left-right parse (this needs some explanation, but I think it makes sense; it can actually be made more interesting by weighting the dependencies as complements, adjuncts, etc.). This figure varies considerably between texts, according to their syntactic complexity. For simple texts the upper limit is 4 and most words have no more than one. Notice how precise some of these figures are - and therefore how easy it should be to refute them. I'm not sure that I really believe them myself, but that should increase your confidence in them since I was the analyst who produced them. It would be very interesting to have similar figures for other languages. I have analysed two similar texts in German and found: - contrary to expectations, a small majority of words follow their heads; - the proportion of complement and subject dependencies is within the same range as for English; - the proportion of words that are next to their heads or only one word away from their heads is about 5% less than the minimum for English - i.e. German dependencies are less closely tied to physical proximity than English ones are; - the number of open dependencies is higher than in English - i.e. processing load for the German reader is greater than for the English reader. (This is a rather mind-blowing conclusion, and needs to be checked carefully before being announced as a discovery!) Needless to say, the analyses reported are based on my own grammar (on the whole as reported in my "English Word Grammar" 1990), some of which are controvertial - e.g. I take a determiner as the head of the accompanying common noun. (I think I have some weak statistical support for this analysis, in fact.) You have to read the figures with these assumptions in mind - e.g. if you're sure that determiners depend on their common noun, then the figures for complements and head-first have to be reduced by between 10 and 20%. I'd be very interested to hear from anyone who has similar figures they'd be willing to share with me, or to send more details to anyone who'd like to examine other texts. I have a Prolog system for manually parsing texts and counting the results which may be helpful. Dick Hudson Dept of Phonetics and Linguistics, University College London, Gower Street, London WC1E 6BT (071) 387 7050 ext 3152 --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From Edmund.Grimley-Evans@cl.cam.ac.uk Wed Jan 6 06:37:47 1993 Return-Path: Received: from swan.cl.cam.ac.uk by aisun1.ai.uga.edu (4.1) id AA02685; Wed, 6 Jan 93 06:37:47 EST Received: from duncan.cl.cam.ac.uk (user etg10 (rfc931)) by swan.cl.cam.ac.uk with SMTP (PP-6.4) to cl; Wed, 6 Jan 1993 11:37:22 +0000 Received: by duncan.cl.cam.ac.uk (5.57/SMI-3.0DEV3) id AA01571; Wed, 6 Jan 93 11:37:09 GMT Date: Wed, 6 Jan 93 11:37:09 GMT From: Edmund.Grimley-Evans@cl.cam.ac.uk Message-Id: <9301061137.AA01571@duncan.cl.cam.ac.uk> To: dg@ai.uga.edu Cc: uclyrah@ucl.ac.uk Subject: Dependency statistics How frequent were discontinuous dependency trees? (The strongest and simplest definition of continuity that I know of is that the set of (indirect) dependents of every word should be a contiguous set of words in the sentence. I understand that there are also weaker ways of defining continuity.) ======================================================================= Edmund GRIMLEY EVANS Edmund.Grimley-Evans@cl.cam.ac.uk ======================================================================= --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From @sun.central-services.umist.ac.uk:harold@language-linguistics.umist.ac.uk Wed Jan 6 06:38:47 1993 Return-Path: <@sun.central-services.umist.ac.uk:harold@language-linguistics.umist.ac.uk> Received: from sun2.nsfnet-relay.ac.uk by aisun1.ai.uga.edu (4.1) id AA02696; Wed, 6 Jan 93 06:38:47 EST Via: uk.ac.umist.central-services.sun; Wed, 6 Jan 1993 11:38:16 +0000 Received: from honshu.ccl.umist.ac.uk by cclsun.ccl.umist.ac.uk; Wed, 6 Jan 93 11:38:07 GMT From: Harold Somers Message-Id: <4100.9301061138@honshu.ccl.umist.ac.uk> Subject: Re: Translation of Tesniere To: uclyrah@ucl.ac.uk (R.HudsonIBM-850) Date: Wed, 6 Jan 93 11:38:04 GMT Cc: dg@ai.uga.edu In-Reply-To: <9301061036.AA44077@link-1.ts.bcc.ac.uk>; from "R.HudsonIBM-850" at Jan 6, 93 10:36 am X-Mailer: ELM [version 2.3 PL11] > > > Alain Polguere seems to me to be right - there is no English translation of > Tesniere, and there ought to be one (perhaps abbreviated?). I have contacts I would have thought that it would be better to have a detailed commentary on Elements, with extensive extracts, rather than a faithful translation. There's so much in there that is naive, wrong or out-dated. 10 years ago, when I was doing research for my PhD, I wrote [2] a short paper (11 pages) discussing in detail the parts of Elements which I thought werte directly relevant to Valnecy and Case grammar (the topic of my thesis), viz. Book A 'Preamble' Ch1-6 (pp 11-20) Book B 'Structure of the verb phrase' Ch 48-62 (pp 102-143) Book D 'Valency' (pp 238-282 partim) I ignored all the stuff on 'translation', which I - and earlier commentators like Baum [1] - took to be very similar to the early ideas of Chomsky; this wasnt of interest to me at the time, so I skipped over it. Being a fairly thorough young chap (those WERE the days), I chased up quite a lot of reviews of Elements as well (including one in Hebrew which I never managed to get translated! - sadly, I dont seem to be able to lay my hands on my bulging file of Tesniere-related stuff), and also managed to find copies of the two non-posthumous publications [3,4]. If anyone is interested in this old paper of mine (I wrote it in 1982), please let me know your surface address, and I will send it to you. > with publishers. Can anyone suggest a good person to do such a translation? > The ideal person would be an experienced English-French translator who could Better would be a French-English translator! :-) References [1] R. Baum. 1976. Dependenzgrammatik: Tesnie`re's Modell der Sprachbeschreibung in wissenschaftlicher und kritischer Sicht (Beihefte zur Zeitschrift fu"r Romanische Philologie 151). Tu"bingen: Max Niemeyer Verlag. [2] H.L. Somers. 1982. Lucien Tesnie`re: a review of his 'oeuvre'. Report No. 82/8, Centre for Computational Linguistics, UMIST, Manchester. [3] L. Tesnie`re. 1934. Comment construire un syntaxe. Bulletin de la Faculte' des Lettres de Strasbourg 12, 219-229. [4] L. Tesnie`re. 1953. Esquisse d'une syntaxe structurale. Paris: Librairie C. Klincksieck. --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From @sun.central-services.umist.ac.uk:harold@language-linguistics.umist.ac.uk Wed Jan 6 06:47:24 1993 Return-Path: <@sun.central-services.umist.ac.uk:harold@language-linguistics.umist.ac.uk> Received: from sun2.nsfnet-relay.ac.uk by aisun1.ai.uga.edu (4.1) id AA02708; Wed, 6 Jan 93 06:47:24 EST Via: uk.ac.umist.central-services.sun; Wed, 6 Jan 1993 11:46:57 +0000 Received: from honshu.ccl.umist.ac.uk by cclsun.ccl.umist.ac.uk; Wed, 6 Jan 93 11:46:51 GMT From: Harold Somers Message-Id: <4143.9301061146@honshu.ccl.umist.ac.uk> Subject: valency of nouns To: dg@ai.uga.edu Date: Wed, 6 Jan 93 11:46:49 GMT X-Mailer: ELM [version 2.3 PL11] Can anyone mail me a bibliography on the subject of nominal valency. I did some work on this a few years ago, and so have a few articles up to about 7 years ago, the latest being a manuscript by Jane Grimshaw (Nouns, Arguments and Adjuncts, May 1986). I wonder if that was ever published anywhere? A more up-to-date bibliography would be much appreciated. --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From bnevin@ccb.bbn.com Wed Jan 6 09:15:50 1993 Return-Path: Received: from BBN.COM by aisun1.ai.uga.edu (4.1) id AA03097; Wed, 6 Jan 93 09:15:50 EST Message-Id: <9301061415.AA03097@aisun1.ai.uga.edu> Received: from CCB.BBN.COM by BBN.COM id aa02577; 6 Jan 93 9:11 EST Date: Wed, 6 Jan 93 09:07:15 EST From: "Bruce E. Nevin" Subject: determiner as head To: dg%ai.uga.edu@bbn.com Cc: bn@ccb.bbn.com I have some independent support for your analysis of the determiner as the head of the noun phrase. In operator grammar, "the" is analyzed as a noun in apposition to the following noun, by way of a reduction from "that which is": The family doctor is fast disappearing. That which is a family doctor is fast disappearing. Compare other appositional constructions, viz.: My friend John My friend who is John My friend the doctor My friend who is the doctor For discussion and supporting argument, see Z. Harris, _A Grammar of English on Mathematical Principles_, Wiley (1982), 236-243. > From: "R.HudsonIBM-850" > Date: Wed, 06 Jan 93 11:08:46 +0000 > Needless to say, the analyses reported are based on my own grammar (on > the whole as reported in my "English Word Grammar" 1990), some of > which are controvertial - e.g. I take a determiner as the head of the > accompanying common noun. (I think I have some weak statistical > support for this analysis, in fact.) You have to read the figures with Bruce Nevin bn@bbn.com --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From rsharman@vnet.ibm.com Wed Jan 6 09:53:34 1993 Return-Path: Received: from vnet.ibm.com by aisun1.ai.uga.edu (4.1) id AA03237; Wed, 6 Jan 93 09:53:34 EST Message-Id: <9301061453.AA03237@aisun1.ai.uga.edu> Received: from WINVMD by vnet.ibm.com (IBM VM SMTP V2R2) with BSMTP id 8885; Wed, 06 Jan 93 09:51:35 EST Date: Wed, 6 Jan 93 14:52:37 GMT From: rsharman@vnet.ibm.com To: dg@ai.uga.edu Subject: Dependency Statistics After some discussions with Dick Hudson a few weeks ago I became interested in the question of how adjacent and non-adjacent words might be related, and whether it was possible to shed some light on the (thorny) question of whether a phrase-structure or a dependency-graph representation of the word associations in a sentence was better. It occurred to me that some techniques currently in use in Language Modelling for Speech Recognition might come in useful, so I proposed the following problem: The WORD PAIR ASSOCIATION problem: what is the average degree of association between pairs of words, taken over the language as a whole, when the constraint is applied that the pairs must be adjacent words? This generalises to: The GENERAL WORD ASSOCIATION problem: what is the average degree of association between pairs of words, taken over the language as a whole, when the pairs are taken at arbitrary distances (one word intervening, two words intervening, etc)? From experience in predicting words for Speech Recognition I would have expected that the further apart the words in a pair are, the less their association. But how far do associations go? what is the precise nature of the relationship? etc. Dick Hudson's figures on dependency relations seem to suggest that local dependencies are very strong, if not the dominant type of relationship. But it could be that the figures he gives have something to do with the type of grammatical analysis he has assumed. So, independently of any grammatical theory, what does one word tell you about other words in its local context? I decided to try a small experiment calculating the Mutual Information between pairs of words as an indicator of this association. (Mutual Information between two items, x and y, is defined as the log of the ratio of the joint probability of the two items occuring to the product of their separate a priori probabilities. Any textbook on Information Theory gives the details). Then average mutual information for all words taken from a corpus would give some figures for the general amount of association in a language. Here are the results for a sample corpus of 192,723 words of English Newspaper text, with a vocabulary of 22,996 unique lexical items: 1 Self-information of a word with itself 10.25 2 adjacent word pairs 3.00 3 a pair separated by 1 word 2.83 4 a pair separated by 2 words 2.73 5 a pair separated by 3 words 2.68 The measure is in bits per word. So, knowing a word is worth about 10 bits, and it tells you about 3 bits worth about the next word, and so on. In general, the hypothesis appears correct: the further words are away from each other the less related they are, INDEPENDENTLY of any assumptions about grammatical theory. Any comments? Richard Sharman --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From @mail-a.bcc.ac.uk:ucleaar@ucl.ac.uk Wed Jan 6 16:12:25 1993 Return-Path: <@mail-a.bcc.ac.uk:ucleaar@ucl.ac.uk> Received: from sun2.nsfnet-relay.ac.uk by aisun1.ai.uga.edu (4.1) id AA05897; Wed, 6 Jan 93 16:12:25 EST Via: uk.ac.bcc.mail-a; Wed, 6 Jan 1993 21:09:18 +0000 Received: from link-1.ts.bcc.ac.uk by mail-a.bcc.ac.uk with SMTP (PP) id <22719-0@mail-a.bcc.ac.uk>; Wed, 6 Jan 1993 21:09:16 +0000 Received: by link-1.ts.bcc.ac.uk (AIX 3.2/UCB 5.64/4.03) id AA60538; Wed, 6 Jan 1993 21:09:13 GMT Message-Id: <9301062109.AA60538@link-1.ts.bcc.ac.uk> To: dg@ai.uga.edu, "Bruce E. Nevin" Subject: Re: determiner as head In-Reply-To: (Your message of Wed, 06 Jan 93 09:07:15 EST.) <9301061415.AA03097@aisun1.ai.uga.edu> X-Caution: the 'from' field may be garbled, but 'sender' is probably OK. Date: Wed, 06 Jan 93 21:09:12 +0000 From: And Rosta Bruce Nevin: > I have some independent support for your analysis of the > determiner as the head of the noun phrase. In operator grammar, > "the" is analyzed as a noun in apposition to the following noun, > by way of a reduction from "that which is": > > The family doctor is fast disappearing. > That which is a family doctor is fast disappearing. > > Compare other appositional constructions, viz.: > > My friend John > My friend who is John > My friend the doctor > My friend who is the doctor > > For discussion and supporting argument, see Z. Harris, _A Grammar > of English on Mathematical Principles_, Wiley (1982), 236-243. Why is this an argument for the determiner being the head? Rather, the examples seem to show the attractiveness of treating the determiner as *coreferential* with the common noun. Or is _that which is_ actually *present* at some stage & then deleted? What are the grounds for deciding which phrase's head is the root of the restrictive appo construction? Is the funny prosody of (1a) relevant? 1 a. MY friend eLIza b. my FRIEND the DOCtor ---- And Rosta --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From mcovingt@aisun3.ai.uga.edu Wed Jan 6 16:36:50 1993 Return-Path: Received: from server.uga.edu by aisun1.ai.uga.edu (4.1) id AA06270; Wed, 6 Jan 93 16:36:50 EST Received: by server.uga.edu (5.57/Ultrix3.0-C) id AA22292; Wed, 6 Jan 93 16:36:49 -0500 Received: by aisun3.ai.uga.edu (4.1) id AA08438; Wed, 6 Jan 93 16:36:48 EST From: mcovingt@aisun3.ai.uga.edu (Michael Covington) Message-Id: <9301062136.AA08438@aisun3.ai.uga.edu> Subject: Natural language software registry (announcement) To: dg@ai.uga.edu Date: Wed, 6 Jan 93 16:36:47 EST X-Mailer: ELM [version 2.3 PL11] Forwarded message: From registry@dfki.uni-sb.de Wed Jan 6 05:18:03 1993 Date: Wed, 6 Jan 93 11:20:54 +0100 Message-Id: <9301061020.AA07617@disco-sun5.dfki.uni-sb.de> Organization: DFKI Saarbruecken GmbH, D-W 6600 Saarbruecken From: registry@dfki.uni-sb.de (The Software Registry) To: mcovingt@aisun3.ai.uga.edu Subject: NL Software Registry Reply-To: registry@dfki.uni-sb.de (Dear participants: I'm passing along the following, which seems interesting. -- Michael Covington) Remark: NATURAL LANGUAGE SOFTWARE REGISTRY The Natural Language Software Registry is a catalogue of software implementing core natural language processing techniques, whether available on a commercial or noncommercial basis. The current version includes + speech signal processors, such as the Computerized Speech Lab (Kay Electronics) + morphological analyzers, such as PC-KIMMO (Summer Institute for Linguistics) + parsers, such as Alveytools (University of Edinburgh) + knowledge representation systems, such as Rhet (University of Rochester) + multicomponent systems, such as ELU (ISSCO), PENMAN (ISI), Pundit (UNISYS), SNePS (SUNY Buffalo), + applications programs (misc.) This document is available on-line via anonymous ftp to ftp.dfki.uni-sb.de (directory:registry), by email to registry@dfki.uni-sb.de, and by physical mail to the address below. Now our request: we would like to subscribe Dependency Grammar in order to be fully informed about the activities in this area. Yours, Christoph Jung, Markus Vonerden Natural Language Software Registry Deutsches Forschungsinstitut fuer Kuenstliche Intelligenz (DFKI) Stuhlsatzenhausweg 3 D-W-6600 Saarbruecken Germany phone: +49 (681) 303-5282 e-mail: registry@dfki.uni-sb.de -- :- Michael A. Covington internet mcovingt@uga.cc.uga.edu : ***** :- Artificial Intelligence Programs phone 706 542-0358 : ********* :- The University of Georgia fax 706 542-0349 : * * * :- Athens, Georgia 30602-7415 U.S.A. amateur radio N4TMI : ** *** ** --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From @mail-a.bcc.ac.uk:uclyrah@ucl.ac.uk Thu Jan 7 03:01:57 1993 Return-Path: <@mail-a.bcc.ac.uk:uclyrah@ucl.ac.uk> Received: from sun2.nsfnet-relay.ac.uk by aisun1.ai.uga.edu (4.1) id AA08067; Thu, 7 Jan 93 03:01:57 EST Via: uk.ac.bcc.mail-a; Thu, 7 Jan 1993 08:00:16 +0000 Received: from link-1.ts.bcc.ac.uk by mail-a.bcc.ac.uk with SMTP (PP) id <26499-0@mail-a.bcc.ac.uk>; Thu, 7 Jan 1993 08:00:09 +0000 Received: by link-1.ts.bcc.ac.uk (AIX 3.2/UCB 5.64/4.03) id AA67697; Thu, 7 Jan 1993 08:00:05 GMT Message-Id: <9301070800.AA67697@link-1.ts.bcc.ac.uk> To: dg@ai.uga.edu Subject: Apposition Date: Thu, 07 Jan 93 08:00:05 +0000 From: "R.HudsonIBM-850" 7 January 1993 Bruce Nevin says that operator grammar recognises the common noun as a noun in apposition to the determiner; e.g. in "the family doctor", "doctor" is in apposition to "the". Splendid - just what I say in "Word Grammar" (1984: 91). But then And Rosta asks what evidence there is for the next step in the argument, for taking the common noun as a dependent of the determiner. The evidence that I would use involves uncontroversial examples of apposition like (1). (1)a The village of Trumpington lies next to the M25. b The fact that he was late is beyond dispute. In (a), "Trumpingon" is clearly in apposition to "village" (i.e. coreferential with it), but equally clearly "Trumpington" is subordinated to "village" by the preposition "of". And in (b) it is quite clear that "that (he was late)" is a dependent of "fact", because it follows the same rules for extraposition as a relative clause like "that he reported" in (2); but at the same time it is clear that it is in apposition to "fact". (2) The fact that he reported is beyond dispute. In both cases the "that" clause can be extraposed: (3)a The fact is beyond dispute that he was late. b The fact is beyond dispute that he reported. Moreover, in both (1b) and (2) the relation between "that" and the preceding noun would act as a barrier to extraction if the whole noun- phrase had been in an (otherwise) extractable position: (4)a *Who did you discuss the fact that he had contacted? b *Who did you discuss the fact that he had reported to? In other words, the argument for determiners as head would run as follows: - there are some cases of apposition in which it is clear that the second element is dependent on the first; - the semantic relation between a common noun and its preceding determiner is the same as in standard examples of apposition; - there is no clear evidence that the common noun is head of the determiner; - therefore we can treat the relation between them as a special case of the already established apposition relation. Dick Hudson Dept of Phonetics and Linguistics, University College London, Gower Street, London WC1E 6BT (071) 387 7050 ext 3152 --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From @mail-a.bcc.ac.uk:uclyrah@ucl.ac.uk Thu Jan 7 03:02:02 1993 Return-Path: <@mail-a.bcc.ac.uk:uclyrah@ucl.ac.uk> Received: from sun2.nsfnet-relay.ac.uk by aisun1.ai.uga.edu (4.1) id AB08067; Thu, 7 Jan 93 03:02:02 EST Via: uk.ac.bcc.mail-a; Thu, 7 Jan 1993 08:01:33 +0000 Received: from link-1.ts.bcc.ac.uk by mail-a.bcc.ac.uk with SMTP (PP) id <26540-0@mail-a.bcc.ac.uk>; Thu, 7 Jan 1993 08:01:26 +0000 Received: by link-1.ts.bcc.ac.uk (AIX 3.2/UCB 5.64/4.03) id AA67746; Thu, 7 Jan 1993 08:01:24 GMT Message-Id: <9301070801.AA67746@link-1.ts.bcc.ac.uk> To: dg@ai.uga.edu Re: Discontinuous phrases Date: Thu, 07 Jan 93 08:01:24 +0000 From: "R.Hudson" 6 January 1993 Edmund Grimley-Evans asks how many discontinuous dependency trees I found in my texts. The brief answer is either (a) that I don't know, or (b) that I didn't find any, according to what theoretical assumptions you make. The most obvious place where the difference arises is in cases of `extraction' like (1), where "who" depends on "contact", so the phrase rooted in "contact" is discontinuous. (1) Who did you tell John he should contact? (a) If you assume that dependencies are by definition `deep', i.e. closely linked to semantics and valency, and that each word has only one head, then "who" has just one head, "contact". I don't know how frequent such phrases are, because I would have recorded "John" as a dependent of "did". This dependency is essential for the position of "John" (i.e. "John" takes its position from "did"), but has nothing to do with the semantic role of "John", so you won't recognise it as a proper dependency. But then you have to decide what to say about examples like (2), where "John" is a `raised' subject of "seems", and has all the characteristics of a syntactic subject (and is therefore a dependent) of "seems" without having any semantic relation to it. (2) John seems to like Mary. (b) If you allow a word to have more than one head (as I do - see for example my "English Word Grammar" 1990) then "who" can have (at least) two heads, "did" and "contact". One is responsible for its position, and the other for its semantic role. If you allow purely surface dependencies of this kind, then you find that every English sentence is built around a dependency tree that is totally continuous. The continuous trees are the only ones I recorded in my analysis, because there are a lot of other grafted-on (but easily recoverable) dependencies (of which the relation between "who" and "contact" is just one example). I felt that it was reasonable to record just the "basic skeleton" because the other dependencies are reasonably predictable from it; e.g. if you know that "who" depends on "did", and that "contact" is subordinate to "did" (i.e. indirectly dependent on it) then it is easy to work out from the grammar that "who" can also depend on "contact". This gives the grammar just a bit more power than a context-free phrase structure grammar, and ought to make it computationally tractable. However it would indeed be interesting to know what the answer to Edmund's question would be given the more standard assumptions of (a), as a measure of the difficulty of the parsing task without my extra assumptions. Dick Hudson Dept of Phonetics and Linguistics, University College London, Gower Street, London WC1E 6BT (071) 387 7050 ext 3152 --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From ellalain@nuscc.nus.sg Thu Jan 7 04:34:02 1993 Return-Path: Received: from nuscc.nus.sg by aisun1.ai.uga.edu (4.1) id AA08170; Thu, 7 Jan 93 04:34:02 EST Received: by nuscc.nus.sg (5.65/1.34) id AA22001; Thu, 7 Jan 93 17:33:56 +0800 Date: Thu, 7 Jan 93 17:33:56 +0800 From: ellalain@nuscc.nus.sg (Alain Polguere) Message-Id: <9301070933.AA22001@nuscc.nus.sg> To: Subject: Tesniere Concerning Tesniere (1959), Harold Somers writes: > I would have thought that it would be better to have a detailed > commentary on Elements, with extensive extracts, rather than a faithful > translation. There's so much in there that is naive, wrong or > out-dated. 10 years ago, when I was doing research for my PhD, I wrote > [2] a short paper (11 pages) discussing in detail the parts of Elements > which I thought werte directly relevant to Valnecy and Case grammar > (the topic of my thesis), Thanks for your comment. You may be right, even though I am not so sure that I necessarily would agree with you on what to consider "naive, wrong or out-dated" in Tesniere's work... Nevertheless, two remarks: 1) It would be nice if all linguistics books which are translated could contain at least a quarter of what can be found AND USED in Tesniere (1959). In that respect, I think it won't lower the standards. 2) I still maintain that a full translation is more desirable; I prefer to select myself what is relevant for my own needs. By the way, I would be glad to help as a reviewer for the translation. I am very interested in having a copy of your '82 paper. I include the .signature below... - Alain POLGUERE -------------------------------------------------------- | Department of English Language and Literature | | National University of Singapore | | 10 Kent Ridge Crescent -- Republic of Singapore 0511 | | tel. (65) 772-3700 / fax (65) 773-2981 | | ellalain@nuscc.nus.sg / ellalain@nusvm.bitnet | ------------------------------------------------------------------------- --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From bnevin@ccb.bbn.com Thu Jan 7 11:47:13 1993 Return-Path: Received: from BBN.COM by aisun1.ai.uga.edu (4.1) id AA08967; Thu, 7 Jan 93 11:47:13 EST Message-Id: <9301071647.AA08967@aisun1.ai.uga.edu> Received: from CCB.BBN.COM by BBN.COM id ab28811; 7 Jan 93 11:45 EST Date: Thu, 7 Jan 93 11:41:44 EST From: "Bruce E. Nevin" Subject: Re: determiner as head In-Reply-To: Your message of Wed, 06 Jan 93 21:09:12 +0000 To: dg%ai.uga.edu@bbn.com Cc: bn@ccb.bbn.com And Rosta: >Why is this an argument for the determiner being the head? Rather, >the examples seem to show the attractiveness of treating the determiner >as *coreferential* with the common noun. Or is _that which is_ actually >*present* at some stage & then deleted? Yes, _that which is_ is actually present. In fact, it never goes away. It is present overtly at an earlier stage of the derivation, and still present in phonemically zero form at the later stage. Thus, all dependencies found in the unreduced form are still present in the reduced form. >What are the grounds for deciding which phrase's head is the root of >the restrictive appo construction? The noun _friend_ is clearly the head of the construction _my friend who is the doctor_, and it is still the head when _who is_ is reduced to zero in _my friend the doctor_. The (indefinite) noun _that_ is clearly the head of the construction _that which is a family doctor_, and it is still the head when _that_ is reduced to _the_ in the environment of _which is_ being reduced to zero, yielding _the family doctor_. > Is the funny prosody of (1a) relevant? > > 1 a. MY friend eLIza > b. my FRIEND the DOCtor Contrastive stress is another matter, available in a pretty unrestricted way. It applies before the reductions, viz.: 1 a. MY friend who is eLIza b. my FRIEND who is the DOCtor For reference, I had said: > In operator grammar, > "the" is analyzed as a noun in apposition to the following noun, > by way of a reduction from "that which is": > > The family doctor is fast disappearing. > That which is a family doctor is fast disappearing. > > Compare other appositional constructions, viz.: > > My friend John > My friend who is John > My friend the doctor > My friend who is the doctor > > For discussion and supporting argument, see Z. Harris, _A Grammar > of English on Mathematical Principles_, Wiley (1982), 236-243. Co-reference is an effect of a metalinguistic assertion that two occurrences of a word are the same. In many treatments of grammar this metalinguistic assertion is in the form of subscript indices in the written representation. But the metalanguage for natural language is demonstrably statable in natural language itself (see below), and must necessarily be included in language. (The metalanguage for logic or mathematics depends upon the background vernacular of language, the metalanguage for language cannot. For discussion, see Z. Harris, _A Theory of Language and Information_ Oxford, Clarendon (1991).) The metalanguage adequate for asserting sameness of two words requires very simple vocabulary and syntax. The assertion of sameness is one of the requirements for the derivation of a relative clause from a secondary sentence: My friend the doctor arrived My friend who is the doctor arrived. My friend--a friend (prior same as mentioned) is the doctor--arrived. The intonation of the interrupting secondary sentence is retained in the relative clause and in the appositional noun phrase. (Contrastive stress on _doctor_ would be retained also.) This account clarifies the difference between "ordinary" dependency of argument words and operators ("predicative" words like verbs, adjectives, adverbs, etc.) and dependency of adjuncts (modifiers) on heads. We have zero allomorphs in many morphophonemic alternations, viz. _John has three sheep-0_. What are commonly called cases of elision involve zero allomorphs, e.g.: I prefer that John should go. I prefer that I go. ---------- I prefer to 0 go. What does that do for dependency representations? prefer that prefer to / \ / \ / \ / \ I go I go / / I 0 The reduction to zero requires a metalinguistic assertion of sameness. The interruption is under a paratactic conjunction (represented in writing by semicolon or dash, and represented in speech by secondary intonation, which is preserved in the reduced form of the adjunct). The fact that it is an interruption means that dependency lines (or projections from nodes to the linear string of words) must cross: ; 0 / \ / \ / same as / 0 / / \ / / \ | prior mentioned | 0 0 | | | | prefer that prefer to / \ / \ / \ / \ I go I go / / I 0 In this example, the paratactic conjunction _;_ is zeroed (the entire metalinguistic assertion has zero form, and there is nothing left to receive reduced intonation). In the case of modifiers from relative clauses, it has phonemic shape still (as subordinated intonation). ; / \ / same as / / \ / prior mentioned ; / \_______________________ / \ hit is red / \ /------------/ John a ball a ball John hit a ball; a ball (prior same as mentioned) is red. In the full relative clause, the paratactic conjunction has additional phonemic content as the wh- of the relative pronoun: wh- / \ / 0 / / \ / 0 0 ; / \_______________________ / \ hit is red / \ /---------/ John a ball -ich John hit a ball which is red. I am now going to use | in place of slanted lines for the sake of compression. A vertical line does represent a dependency below, not a projection from the dependency tree to the linear string. 0 | \ | 0 | / \ ; 0 0 / \ / red hit | / \ 0 John a .. ball John hit a red ball. Graphical representations of word dependencies soon reach the limit of practicability (even without the limitations of diagramming with ASCII characters!). And one does not need them to implement a grammar. The point here is to show that adjuncts, like the adjective "red" in this example and the noun in apposition in the earlier examples, are not dependent upon their head noun, they are dependent upon the paratactic conjunction that makes an interrupting sentence into a secondary aside about a word in the main sentence. That conjunction is still present in zero form. As is the metalinguistic sameness statement. Again, for exemplification, argument, and discussion, see Z. Harris, _A Grammar of English on Mathematical Principles_. Bruce Nevin bn@bbn.com --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From vhahn@nats4.informatik.uni-hamburg.de Wed Jan 13 04:13:11 1993 Return-Path: Received: from deneb.dfn.de by aisun1.ai.uga.edu (4.1) id AA23919; Wed, 13 Jan 93 04:13:11 EST Received: from fbihh.informatik.uni-hamburg.de by deneb.dfn.de (4.1/SMI-4.2) id AA26790; Wed, 13 Jan 93 10:10:52 +0100 Received: from nats4.informatik.uni-hamburg.de by fbihh.informatik.uni-hamburg.de (5.65+/FBIHH-2.26) with SMTP; id AA06275; Wed, 13 Jan 93 10:10:38 +0100 Message-Id: <9301130908.AA04723@nats4.informatik.uni-hamburg.de> Received: from [134.100.5.232] (bomac42) by nats4.informatik.uni-hamburg.de (4.1/FBIHH-2.15); id AA04723; Wed, 13 Jan 93 10:08:04 +0100 Date: Wed, 13 Jan 1993 10:12:41 +0100 To: dg@ai.uga.edu From: vhahn@nats4.informatik.uni-hamburg.de Subject: Sgall's answers Dear Colleagues, Using the opportunity of a short stay in Hamburg, where I can find more quiet time than at home, I would like to express my opinion on some of the questions you have been discussing, applying the advantages of my age, viz. the training in European structural grammar I went through with my Prague teachers in the years around 1950, and my owm long experience with empirical and theoretical research in dependency syntax. (i) As for Chomsky's 'minimalist theory', I cannot say anything concerning the technical shape of his new theory, but at least one point is important, and that is his emphasis on the two 'interface levels' of phonetics and of logical form. In our publications on the 'Functional Generative Description' (see esp. the book The meaning of the sentence, published with Reidel in 1986, with J. Mey as the editor) we have always worked with a single underlying level, conjoining (from a viewpoint closw to Chomsky's "Cartesian Linguistics" from the 1960's) the roles of 'initial P-markers', D-structure and logical form; in recent writings, in our group also the possibility to work without a level of surface syntax was pointed out (see my paper in Wiener Linguistischer Almanach, Sonderband 33 (Festschrift fuer Rozencvejg), Vienna 1992), so that there are clear points of convergence between the two approaches, as far as they can be compared at all. (ii)It may be claimed that dependency trees are much simpler than the usual kinds of P-markers, having a lower number of nodes, working just with a single set of symbols for syntactic relations (all of which, not only those corresponding to theta roles or arguments, but also free adverbials or adjuncts, being justified by the valency grids of the head words, in which parameters of obligatoriness, deletability, role as controllers and many others can be lexically specified). However, it should be admitted that coordination (even if not fully symmetric) is a relation of another kind, so that more than the two dimensions of the dependency tree are needed for a network that completely describes the sentence structure. Strong limitations such as R. Hudson's adjacency (projectivity) make the whole system simple enough (the unprojective consructions can be described as exceptional deviations), treatable e.g. in a linearized form with two kinds of parentheses (for dependency and for cooridnation), see the paper by Petkevic in Theoretical Linguistics 1987). (iii) An analog of the ID/LP rule dichotomy can be seen in our hypothesis of a basic ('systemic') ordering of the kinds of dependency relations (of valency slots, or of theta roles and adjuncts), which is reflected by word order with several 'deviations', the main of which consists in the elements of the topic standing more to the left than what the systemic ordering itself would determine (see Sgall and Hajicova, Ordering principle, in Journal of Pragmatics 1987); other such deviatios concern 'shallow rules' on the position of the verb, of the adjective, of the clitics, etc. (iv) A dependency based parser of English was published in our group by Z. Kirschner in two volumes, which are of an internal character, not accessible on the market, but I can send them to those who are interested; the two drawbacks there are that he used Colmerauer's Q-systems for implementation and that the lexicon used there is narrowly restricted. We are now preparing a more general version of the parser, and the contact person for this issue would be Alexandr Rosen, e-mail:rosen@Praha1.ff.cuni.cs.internet. Since I am here just for a few days, please, do not send replies on the address of this message, but rather to Prague (sgall@cspguk11.bitnet). I hope I'll be able to continue our discussions in February. Cordially Petr Sgall ===================================== ==== Walther v.Hahn ====== = Computer Science Department = = Natural Language Systems Division = = Bodenstedtstr.16 = = D - 2000 HAMBURG 50 = = Tel: (X40) 4123 4529 = = Fax (X40) 4123 6530 = ------------------------------------- vhahn@nats2.informatik.uni-hamburg.de ===================================== --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From mcovingt@aisun3.ai.uga.edu Thu Jan 14 23:41:48 1993 Return-Path: Received: from aisun3.ai.uga.edu by aisun1.ai.uga.edu (4.1) id AA03622; Thu, 14 Jan 93 23:41:48 EST Received: by aisun3.ai.uga.edu (4.1) id AA22675; Thu, 14 Jan 93 23:41:48 EST Date: Thu, 14 Jan 93 23:41:48 EST From: mcovingt@aisun3.ai.uga.edu (Michael Covington) Message-Id: <9301150441.AA22675@aisun3.ai.uga.edu> To: dg@ai.uga.edu Subject: Test message This is a test message to validate the entire DG mailing list. We have just upgraded our mailer to sendmail.mx and should have better success reaching overseas addresses that had caused difficulty. If you receive this message, please ignore it; if you don't receive it, I'll be notified automatically. --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From mcovingt@aisun3.ai.uga.edu Fri Jan 22 11:24:49 1993 Return-Path: Received: from aisun3.ai.uga.edu by aisun1.ai.uga.edu (4.1) id AA27546; Fri, 22 Jan 93 11:24:49 EST Received: by aisun3.ai.uga.edu (4.1) id AA05757; Fri, 22 Jan 93 11:24:47 EST From: mcovingt@aisun3.ai.uga.edu (Michael Covington) Message-Id: <9301221624.AA05757@aisun3.ai.uga.edu> Subject: Test message To: dg@ai.uga.edu Date: Fri, 22 Jan 93 11:24:47 EST X-Mailer: ELM [version 2.3 PL11] Test message; please ignore. --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From mcovingt@aisun3.ai.uga.edu Sun Jan 24 20:38:57 1993 Return-Path: Received: from aisun3.ai.uga.edu by aisun1.ai.uga.edu (4.1) id AA04520; Sun, 24 Jan 93 20:38:57 EST Received: by aisun3.ai.uga.edu (4.1) id AA09560; Sun, 24 Jan 93 20:38:56 EST From: mcovingt@aisun3.ai.uga.edu (Michael Covington) Message-Id: <9301250138.AA09560@aisun3.ai.uga.edu> Subject: Service upgrades on aisun1 (DG host) To: dg@ai.uga.edu Date: Sun, 24 Jan 93 20:38:55 EST X-Mailer: ELM [version 2.3 PL11] The mailer on aisun1.ai.uga.edu (the machine that hosts DG) has just been upgraded, and we should have a much easier time reaching faraway addresses. (Some of you may be hearing from us now for the first time!) There will be further upgrades (with concomitant outages and temporary malfunctions) during the next couple of weeks. -- :- Michael A. Covington internet mcovingt@uga.cc.uga.edu : ***** :- Artificial Intelligence Programs phone 706 542-0358 : ********* :- The University of Georgia fax 706 542-0349 : * * * :- Athens, Georgia 30602-7415 U.S.A. amateur radio N4TMI : ** *** ** --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From mcovingt@aisun3.ai.uga.edu Tue Jan 26 22:57:52 1993 Return-Path: Received: from aisun3.ai.uga.edu by aisun1.ai.uga.edu (4.1) id AA25186; Tue, 26 Jan 93 22:57:52 EST Received: by aisun3.ai.uga.edu (4.1) id AA13465; Tue, 26 Jan 93 22:57:50 EST From: mcovingt@aisun3.ai.uga.edu (Michael Covington) Message-Id: <9301270357.AA13465@aisun3.ai.uga.edu> Subject: DG parsing vs PSG parsing To: dg@ai.uga.edu Date: Tue, 26 Jan 93 22:57:49 EST X-Mailer: ELM [version 2.3 PL11] Is there some useful mapping by which dependency grammars and D-trees can be mapped onto phrase-structure grammars and PS-trees, thereby generating dependency counterparts of well-known phrase-structure parsing algorithms? I can think of some trivial mappings that don't seem to shed any real light on the issue. Has anyone delived more deeply into this? -- :- Michael A. Covington internet mcovingt@uga.cc.uga.edu : ***** :- Artificial Intelligence Programs phone 706 542-0358 : ********* :- The University of Georgia fax 706 542-0349 : * * * :- Athens, Georgia 30602-7415 U.S.A. amateur radio N4TMI : ** *** ** --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From rsharman@vnet.ibm.com Wed Jan 27 05:00:39 1993 Return-Path: Received: from vnet.ibm.com by aisun1.ai.uga.edu (4.1) id AA26561; Wed, 27 Jan 93 05:00:39 EST Message-Id: <9301271000.AA26561@aisun1.ai.uga.edu> Received: from WINVMD by vnet.ibm.com (IBM VM SMTP V2R2) with BSMTP id 9547; Wed, 27 Jan 93 04:58:35 EST Date: Wed, 27 Jan 93 09:49:09 GMT From: rsharman@vnet.ibm.com To: dg@ai.uga.edu Subject: Finding dependencies from Phrase Structures The correspondence between dependency relations and phrase structure analyses is one of the most interesting questions in practical syntactic analysis of real texts. Avoiding the debate of whether dependencies are "fundamental" and phrases "derivative", or vice versa, for the moment, it would a useful practical task to map one to the other. In particular, there is a growing amount of text labelled with phrase structures, which we call a "treebank". However I have not seen the corresponding "dependency bank" anywhere. If someone has one, or can annotate a given text with dependencies to create one, then a preoper study could be sone of the correspondence between dependencies and phrases IN THE SAME TEXT, and so a model which maps between the two could be created. In default of this, it is necessary to resort to some kind of rule-based approach, which is unfortunately beset by difficulties of interpretation, since the underlying assumptions of the phrase structure analysis, and probably also the assumptions of the dependency labelling, need to be taken into account in specifying the rules. It would, I feel, be preferable to take an existing phrase labelled treebank, and do the dependency analysis, then compare the two. This way we would learn the actual mappings, as opposed to hypothetically assumed mappings? Richard Sharman --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From LARSSON@ntcclu.ntc.nokia.com Wed Jan 27 06:19:02 1993 Return-Path: Received: from NTC02.TELE.NOKIA.FI by aisun1.ai.uga.edu (4.1) id AA26691; Wed, 27 Jan 93 06:19:02 EST Date: Wed, 27 Jan 1993 13:19:52 +0300 (EET) From: LARSSON@ntcclu.ntc.nokia.com Message-Id: <930127131952.2581fc7d@ntcclu.ntc.nokia.com> Subject: Mapping PS-structures onto D-structures To: dg@ai.uga.edu X-Vmsmail-To: INET::"dg@ai.uga.edu" In a recent posting Michael Covington asked: Is there some useful mapping by which dependency grammars and D-trees can be mapped onto phrase-structure grammars and PS-trees, thereby generating dependency counterparts of well-known phrase-structure parsing algorithms? This issue was treated by Robert F. Simmons in the book "Computations from the English" (Prentice-Hall; 1984. ISBN 0-13-164640-0). In chapter 4 (p. 57) he gives the following rules for transforming PS-structures into D-structures: "[...] VERBS govern preceding and following NPs, auxiliaries, VPs adverbials, and PPs that are not more immediately governed. NOUNS dominate articles, adjectives, modifying nouns, and PPs. ADJECTIVES dominate modifying adverbials. PREPOSITIONS are dominated by preceding nouns and verbs and dominate following nouns. [...]" In chapter 5, Simmons elaborates the rationale for doing these transformations and presents appropriate grammar rules (p. 89) to accomplish the task. The rules are DCG rules (called 'procedural grammar rules' by the author) intended for processing with HCPRVR, a Horn clause theorem prover written in Lisp (NOT Common Lisp). The grammar is claimed to be symmetric, thus allowing both parsing and generation. I can think of some trivial mappings that don't seem t shed any real [... additional text deleted] Maybe, the above rules are 'trivial' in a certain sense; they are also limited to languages, where the position of constituents (and their order) plays an important role (e.g. the Nordic languages, German, English). In Finnish, for example, ordering rules could partly be applied (i.e. WITHIN noun phrases), but decisions on which NPs are the 'subject' or the 'object' of a sentence must be made using other criteria (morphology etc.). In my present work concerning corpus-based methods in translation studies (including the practice of translation and composition) using bi/multilingual corpora of technical texts, I am facing the problem of finding a working way of comparing structures in several languages. The requirement is that the structures (parse trees, lists or whatever) should themselvs be mutually comparable (and somehow compatible). D-structures with the finite verb as the root node seem to be a 'natural' representation. I would very much appreciate additional opinions, hints and deeper discussion of this issue on the DG list. Implementational details would also be welcome. *-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-* Arne Larsson Nokia Telecommunications Translator Transmission Systems, Customer Services larsson@ntc02.tele.nokia.fi P.O. Box 12, SF-02611 Espoo, Finland Phone +358 0 5117476, Fax +358 0 51044287 *-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-* --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From LARSSON@ntcclu.ntc.nokia.com Wed Jan 27 06:43:02 1993 Return-Path: Received: from NTC02.TELE.NOKIA.FI by aisun1.ai.uga.edu (4.1) id AA26739; Wed, 27 Jan 93 06:43:02 EST Date: Wed, 27 Jan 1993 13:44:01 +0300 (EET) From: LARSSON@ntcclu.ntc.nokia.com Message-Id: <930127134401.2581fc7d@ntcclu.ntc.nokia.com> Subject: DG and PSG To: dg@ai.uga.edu X-Vmsmail-To: INET::"dg@ai.uga.edu" As a follow up to Richard Sharman: Victor Sadler of B.S.O/Research (The Netherlands) has a detailed treatment of a structured parallel corpus of bilingual text, called the Bilingual Knowledge Bank (BKB) in his book "Working with Analogical Semantics (Dordrecht 1989. ISBN 90 6765 428 0.) The author proposes full alignment of dependency structures (p. 127 ff). At the time of writing there was a pilot BKB comprising "some 2,500 sentences" in English, French and Esperanto. It would be very interesting to receive information about the present status of this project (the Distributed Language Translation project), provided such information is not company confidential. *-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-* Arne Larsson Nokia Telecommunications Translator Transmission Systems, Customer Services larsson@ntc02.tele.nokia.fi P.O. Box 12, SF-02611 Espoo, Finland Phone +358 0 5117476, Fax +358 0 51044287 *-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-* --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From ellalain@nuscc.nus.sg Thu Jan 28 00:57:49 1993 Return-Path: Received: from nuscc.nus.sg by aisun1.ai.uga.edu (4.1) id AA03800; Thu, 28 Jan 93 00:57:49 EST Received: by nuscc.nus.sg (5.65/1.34) id AA15246; Thu, 28 Jan 93 13:57:43 +0800 Date: Thu, 28 Jan 93 13:57:43 +0800 From: ellalain@nuscc.nus.sg (Alain Polguere) Message-Id: <9301280557.AA15246@nuscc.nus.sg> To: Subject: DG and Chinese I have a colleague here who would like to know if anything has been done on the description of Chinese in a DG approach. Any reference, etc. will be welcome. Thanks. AP --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From @mail-a.bcc.ac.uk:uclyrah@ucl.ac.uk Thu Jan 28 03:08:47 1993 Return-Path: <@mail-a.bcc.ac.uk:uclyrah@ucl.ac.uk> Received: from sun3.nsfnet-relay.ac.uk by aisun1.ai.uga.edu (4.1) id AA04099; Thu, 28 Jan 93 03:08:47 EST Via: uk.ac.bcc.mail-a; Thu, 28 Jan 1993 08:02:17 +0000 Received: from link-1.ts.bcc.ac.uk by mail-a.bcc.ac.uk with SMTP (PP) id <06259-0@mail-a.bcc.ac.uk>; Thu, 28 Jan 1993 08:08:30 +0000 Received: by link-1.ts.bcc.ac.uk (AIX 3.2/UCB 5.64/4.03) id AA71194; Thu, 28 Jan 1993 08:08:28 GMT Message-Id: <9301280808.AA71194@link-1.ts.bcc.ac.uk> To: dg@ai.uga.edu Subject: converting PSG to DG Date: Thu, 28 Jan 93 08:08:28 +0000 From: RichardHudson50 I spent a couple of weeks a few years ago trying to convert a PSG tree bank into a DG one, and found that there were some insuperable problems when working from the tree-bank itself - i.e. from the list of phrase-types that were observed to occur in the corpus. The main problem is precisely that if you know that some larger phrase contains a smaller phrase, you don't know what kind of word that smaller phrase had as its head. E.g. if you know that there were some PPs, consisting of a preposition followed by a noun-phrase, you can't convert this into a dependency analysis in which a preposition heads some word-class because you don't know what word-class headed each of the NPs. Of course you could assume simply that it was a `noun', including proper nouns, common nouns, pronouns, gerunds, etc, to say nothing of coordinated NPs, but most people working on tree banks would probably want to break down word-classes into more specific classes. It is precisely because phrase-categories are so much less specific and revealing than word-classes that a DG tree-bank is likely to be much more useful than a PSG one, so it would be odd if it were easy to convert the latter into the former. These problems shouldn't arise, however, if one were working from the text itself, with PSG annotations added. In that case I feel sure it would be quite easy at least in the vast majority of cases to replace the PSG phrase tags with tags showing which words depended on which. Dick Hudson Dept of Phonetics and Linguistics, University College London, Gower Street, London WC1E 6BT (071) 387 7050 ext 3152 --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From LARSSON@ntcclu.ntc.nokia.com Thu Jan 28 11:11:14 1993 Return-Path: Received: from NTC02.TELE.NOKIA.FI by aisun1.ai.uga.edu (4.1) id AA05308; Thu, 28 Jan 93 11:11:14 EST Date: Thu, 28 Jan 1993 18:12:12 +0300 (EET) From: LARSSON@ntcclu.ntc.nokia.com Message-Id: <930128181212.2582261f@ntcclu.ntc.nokia.com> Subject: cu-Prolog available To: dg@ai.uga.edu X-Vmsmail-To: INET::"dg@ai.uga.edu" Dear Colleagues, It may be of interest to the members of this group to know that MAC and MS-DOS versions of cu-Prolog from ICOT is now available using ftp from csli.stanford.edu in the directory pub/MacCup. Arne Larsson ---------Here is the readme file for your information:----------- Files: MacCupE0.78w.sit.hqx program, Bin Hexed and Stuff It-ed djcup.lzh MS-DOS program, running under DOS-extender (386/486 cpu), compressed by LHA manual.tex manual for older version of MacCup sample.p sample program a la JPSG util.p utility program ------------------------------------------------------------ Note: MacCup is Macintosh version of CU-Prolog DJCup is MS-DOS version of CU-Prolog compiled with DJ's gcc. CU-Prolog has been developed at ICOT, Japan. ------------------------------------------------------------ Brief Introduction to MacCup: MacCup is a Macintosh version of CUP, Constraint Unification Prolog, whose original version has been developed by ICOT. It looks like a usual Prolog system. However there are several differences: (1) use "" to load a program from a file instead of using []. (2) input program directly from keyboard after `_' prompt instead of using [console]. (3) use %d* to list defined predicates instead of :-listing. etc. etc. To get MacCup from MacCupXXX.sit.hqx: Load (or FTP) to your Macintosh, use BinHex to decode it to get a StuffIt archiver form file. Then use StuffIt to get MacCup program. To run MacCup: Click twice the MacCup icon, and wait about 10 seconds. You will get MacCup window, and the following message: ******** MacCup Ver. XXX ******** All Modular mode (help -> %h) To quit: Input %Q, or :-halt., or Command-Q. To load MacCup Programs: Click FILE menu and OPEN item (or use Command-O), select the program file, and OPEN it. To get command lists: Input %h. MacCup has several `commands' which are marked by `%' as in %h. Note. 1. There are several features which are not mentioned in the manual. As I am writing a new manual, you can get it in September. 2. Thanks to Dr. Emele, I found a bug in MacCupV0.63, and made MacCupV0.70. The difference between them is that the latter will deal with PSTs in constraint transformation better than the former. For example, p({c/C}) :- ab(C). q({d/D,c/{a/D}}). ab({a/1,b/0}). ab({a/0,b/1}). @ p(X),q(X). will produce different results in V0.63 and V0.70. HOWEVER, MacCup still returns incorrect answer for the following: ab({a/1,b/0}). ab({a/0,b/1}). r({c/C});ab(C). s({d/D,c/X});X={a/D}. :-r(X),s(X). Frankly speaking, there is a trade-off between efficiency and getting correct/strict answers. MacCup078S takes efficiency, and MacCup078X takes correctness. (MacCup078X is not distributed for publicity.) 3. You can interrupt MacCup by pressing COMMAND-`.'. If you click the Abort button, the interrupted process will be aborted. I'd appreciate if you give any comments and/or suggestions. ------------------------------------------------------------ Hidetosi SIRAI sirai@csli.stanford.edu, or sirai@sccs.chukyo-u.ac.jp Chukyo University --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From shimizu@hakobera.isct.kyutech.ac.jp Thu Jan 28 23:15:52 1993 Received: from hiko.isct.kyutech.ac.jp by aisun1.ai.uga.edu (4.1) id AA12098; Thu, 28 Jan 93 23:15:52 EST From: shimizu@hakobera.isct.kyutech.ac.jp Received: by hiko.isct.kyutech.ac.jp (5.65/6.4J.6) id AA07469; Fri, 29 Jan 93 13:17:17 +0900 Received: by hakobera.isct.kyutech.ac.jp (5.65/6.4J.6) id AA16379; Fri, 29 Jan 93 13:26:58 +0900 Return-Path: Message-Id: <9301290426.AA16379@hakobera.isct.kyutech.ac.jp> To: dg@ai.uga.edu Cc: shimizu@hakobera.isct.kyutech.ac.jp Subject: more than him have Date: Fri, 29 Jan 93 13:26:57 +0900 The following is the passage I found in a novel called _The Graduate_; 'How many people have done this?' 'Proposed to me?' 'Yes.' 'I don't know,' she said. 'You mean more than him have?' Since I am not a native speaker of English, I am not certain about the grammarticality (or acceptability) of the last sentence (or utterance). Will anyone tell me your opinion? I also would like to know how you would analyse it. \\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\ Makoto SHIMIZU Kyushu Institute of Technology shimizu@hakobera.isct.kyutech.ac.jp \\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\ "I guess I've lost another pupil," said the professor as his glass eye rolled down the sink. --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From rambow@unagi.cis.upenn.edu Sun Jan 31 22:33:18 1993 Received: from linc.cis.upenn.edu by aisun1.ai.uga.edu (4.1) id AA29501; Sun, 31 Jan 93 22:33:18 EST Received: from UNAGI.CIS.UPENN.EDU by linc.cis.upenn.edu id AA25689; Sun, 31 Jan 93 22:33:16 -0500 Return-Path: Received: by unagi.cis.upenn.edu id AA17609; Sun, 31 Jan 93 22:33:15 EST Date: Sun, 31 Jan 93 22:33:15 EST From: rambow@unagi.cis.upenn.edu (Owen Rambow) Posted-Date: Sun, 31 Jan 93 22:33:15 EST Message-Id: <9302010333.AA17609@unagi.cis.upenn.edu> To: mcovingt@aisun3.ai.uga.edu Cc: dg@ai.uga.edu In-Reply-To: Michael Covington's message of Tue, 26 Jan 93 22:57:49 EST <9301270357.AA13465@aisun3.ai.uga.edu> Subject: DG parsing vs PSG parsing Michael Covington some days ago asked about mapping DGs onto PSGs in order to profit from parsing algorithms for PSGs. Several answers, if I have understood them correctly, have addressed the inverse problem, of mapping PSGs (and PS-trees) onto DGs (and D-trees). Thus, the question of how to map a DG onto a PSG (not simply a D-tree onto a PS-tree) remains open. One of the "trivial mappings" that Michael Covington mentions is the following, I assume: each lexical head L is mapped to a phrase-structure rule of the form LP -> (X1)P ... (Xn)P L (Y1)P ... (Yn)P, where the Xi and Yi are the lexical categories of the dependents of L, in the right order. Optional dependents require additional rules. The result is a CFG, which can be parsed using the usual instrumentarium for CFG-parsing (CYK, Early, LR etc.). The problem is that in the case of adjuncts, there may be an unbounded number of dependents, and thus we would need an unbounded number of PS rules, which would not yield a CFG. Evidence from English shows that it is impossible, in general, to write a linguistically plausible CFG with rules of the form given above for a (necessarily finite) CF grammar. The problem is the issue of "lexicalization": in the above form, every rule of the CF grammar is associated with 1 lexical item (thus allowing the easy mapping between DG and PSG). This is not the case for the common CFGs for English. However, it has been shown (Schabes 1989) that in general, CFGs cannot be transferred into a lexicalized form. One way to lexicalize a CFG is to increase the formal power of the underlying mathematical system, and to go to Tree Adjoining Grammars (TAG). In TAG, grammars consist of trees, which are combined to yield derived structures. In a recent paper, Aravind Joshi and I suggest that the fact that TAGs can be lexicalized makes them an ideal interface between PS-based grammars and DGs. So Michael Covington's question could be answered as follows: to apply parsing algorithms for PSGs to DGs, first devise a TAG for the language. A TAG parse can then easily be converted into a DG analysis. Much research has been done on TAG parsing; Early-type algorithms run in O(n^6) time worst case. Writing a TAG for a language is not, alas, a trivial matter. If done in a principled manner, it amounts to devising a theory of syntax. The connection between TAG and DG can also be exploited in the inverse: in a different paper, we propose to use a dependency parser to derive PS-trees. The dependency parser can be "compiled down" from a LR(0) version of a TAG parser. Owen Rambow rambow@unagi.cis.upenn.edu --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From rambow@unagi.cis.upenn.edu Sun Jan 31 22:43:49 1993 Received: from linc.cis.upenn.edu by aisun1.ai.uga.edu (4.1) id AA29530; Sun, 31 Jan 93 22:43:49 EST Received: from UNAGI.CIS.UPENN.EDU by linc.cis.upenn.edu id AA25912; Sun, 31 Jan 93 22:43:45 -0500 Return-Path: Received: by unagi.cis.upenn.edu id AA17703; Sun, 31 Jan 93 22:43:44 EST Date: Sun, 31 Jan 93 22:43:44 EST From: rambow@unagi.cis.upenn.edu (Owen Rambow) Posted-Date: Sun, 31 Jan 93 22:43:44 EST Message-Id: <9302010343.AA17703@unagi.cis.upenn.edu> To: uclyrah@ucl.ac.uk Cc: dg@ai.uga.edu In-Reply-To: RichardHudson50's message of Thu, 28 Jan 93 08:08:28 +0000 <9301280808.AA71194@link-1.ts.bcc.ac.uk> Subject: converting PS corpora to D-corpora Automatically converting the Penn Tree Bank, a PS-corpus, to include D-annotations would be impossible. This is because in sentences like "Yesterday, John ate", "yesterday" is tagged as an NP, and the automatic converter would not be able to determine if it is a temporal adjunct or the direct object. I am not quite sure if this is the same problem Dick Hudson mentioned. Tony Kroch has a project at the U. of Penn to annotate Old and Middle English texts. Though I don't think the representation will actually be a D-notation, it will include sufficient information to unambiguously derive a D-representation (as far as I know). Owen Rambow rambow@unagi.cis.upenn.edu --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From Daniel.Sleator@SPADE.PC.CS.CMU.EDU Sun Jan 31 23:57:40 1993 Return-Path: Received: from SPADE.PC.CS.CMU.EDU by aisun1.ai.uga.edu (4.1) id AA00104; Sun, 31 Jan 93 23:57:40 EST Message-Id: <9302010457.AA00104@aisun1.ai.uga.edu> From: Daniel Sleator Date: Sun, 31 Jan 93 23:56:34 EST To: dg@ai.uga.edu Cc: Davy Temperley Subject: DG parsing, PSG parsing, and link parsing My question to Michael Covington is: Why not parse directly in the dependency formalism? This is what Davy Temperley have done in our link grammar system. Link grammar is a lexically-based dependency-type formalism in which we have written a large and wide coverage English grammar, and for which we have invented and implemented efficient parsing algorithms. Not only that, but the entire system is available via anonymous ftp. (The anonymous ftp site is "spade.pc.cs.cmu.edu", and the directory is "usr/sleator/public", to which you must move with one "cd" command after entering ftp anonymously.) Over 500 people have taken the system. Link grammars are being developed for other languages, as is a commercial grammar checker. Link grammars have also been proposed as a grammatical alternative to the trigram model. For the moment, I'll leave a detailed description of our work to our tech reports (also available via anonymous ftp), and just give some examples. As a very superficial illustration, when our system is given the following NY Times sentence: "An increased supply of civilian goods can restrain our inflation, increase employment, improve public budgets and enhance exports." it constructs the following parsing in three seconds on a decstation 3100. +------------------ +--------------S-------------+ +--------- +--------D--------+ +-------J-------+ | +-------O- | +-----A----+--M--+ +----A---+ | | +-- | | | | | | | | | an increased.v supply.n of civilian.a goods.n can.v restrain.v our --------------------------I-------------------------------------------+ ------------+ +-------------------------------+ -----+ +-------------------------+ +---------O--------+ | -D---+ +------+-----O-----+ +-----+ +----A---+ +-- | | | | | | | | | inflation.n , increase.v employment.n , improve.v public.a budgets.n and ----+----O----+ | | enhance.v exports.n Every word in our system has a formula that describes the requirements which that word imposes on any sentence containing it (actually, it dictates what the combination of links which are required to connect to that word in order to satisfy it.) For example, the word "civilian" (as an adjective) has the following definition in our system: {Ea- or Eb+} & (A+ or ((AI- or IX+ or Ma- or AA+) & {@EV+})); Each word has a similar definition that completely describes the way that word can be used. Daniel Sleator Carnegie Mellon University School of Computer Science 412-268-7563 --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From mcovingt@aisun3.ai.uga.edu Mon Feb 1 00:23:42 1993 Return-Path: Received: from aisun3.ai.uga.edu by aisun1.ai.uga.edu (4.1) id AA00209; Mon, 1 Feb 93 00:23:42 EST Received: by aisun3.ai.uga.edu (4.1) id AA21720; Mon, 1 Feb 93 00:23:38 EST From: mcovingt@aisun3.ai.uga.edu (Michael Covington) Message-Id: <9302010523.AA21720@aisun3.ai.uga.edu> Subject: Dependency parsing To: dg@ai.uga.edu Date: Mon, 1 Feb 93 0:23:38 EST X-Mailer: ELM [version 2.3 PL11] With regard to Daniel Sleator's challenge, "Why not parse directly in the dependency system?" my answer is -- I do! The reason I asked about mappings of dependency grammars onto PS-grammars was as one strategy for doing complexity analysis. As someone else pointed out, it is not necessarily a very promising one, because a PS-rule handles a head and all its arguments (at a particular X-bar level) at once, whereas a D-rule merely says there is *one* argument and doesn't say how many others there may be. Thanks to all for the interesting discussion. I'll be working through people's postings and replying in greater depth soon. (And apologies in advance for any outages that our machine may experience in the next month. A major operating system upgrade is coming.) -- :- Michael A. Covington internet mcovingt@uga.cc.uga.edu : ***** :- Artificial Intelligence Programs phone 706 542-0358 : ********* :- The University of Georgia fax 706 542-0349 : * * * :- Athens, Georgia 30602-7415 U.S.A. amateur radio N4TMI : ** *** ** --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From @mail-a.bcc.ac.uk:uclyrah@ucl.ac.uk Mon Feb 1 03:34:59 1993 Return-Path: <@mail-a.bcc.ac.uk:uclyrah@ucl.ac.uk> Received: from sun3.nsfnet-relay.ac.uk by aisun1.ai.uga.edu (4.1) id AA01114; Mon, 1 Feb 93 03:34:59 EST Via: uk.ac.bcc.mail-a; Mon, 1 Feb 1993 08:23:02 +0000 Received: from link-1.ts.bcc.ac.uk by mail-a.bcc.ac.uk with SMTP (PP) id <00468-0@mail-a.bcc.ac.uk>; Mon, 1 Feb 1993 08:34:01 +0000 Received: by link-1.ts.bcc.ac.uk (AIX 3.2/UCB 5.64/4.03) id AA20410; Mon, 1 Feb 1993 08:33:59 GMT Message-Id: <9302010833.AA20410@link-1.ts.bcc.ac.uk> To: dg@ai.uga.edu Subject: converting PS corpora to D-corpora Date: Mon, 01 Feb 93 08:33:58 +0000 From: RichardHudson50 Owen Rambow says that sentences like (1) can't be automatically reananlysed from a Penn-style PS analysis into a D-analysis. (1) Yesterday John ate. This is because "yesterday" is just labelled as NP, so it would have the same analysis, presumably, as (2). (2) Beans John ate. If you're only looking for semantically-interpretable dependencies, Owen is right; but it's possible to aim in the first instance at least only at the most superficial "skeleton" of dependencies, on which the rest of the depend- ency structure is built. At this level of abstraction, (1) and (2) are the same - they both contain some kind of "extractee" which has a (surface, but not semantically relevant) dependency on "ate". Interesting to hear about Tony Kroch's Old and Middle English corpus. Dick Hudson Dept of Phonetics and Linguistics, University College London, Gower Street, London WC1E 6BT (071) 387 7050 ext 3152 --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From shimizu@hakobera.isct.kyutech.ac.jp Thu Feb 4 00:01:38 1993 Received: from hiko.isct.kyutech.ac.jp ([150.69.2.6]) by aisun1.ai.uga.edu (4.1) id AA25281; Thu, 4 Feb 93 00:01:38 EST From: shimizu@hakobera.isct.kyutech.ac.jp Received: by hiko.isct.kyutech.ac.jp (5.65/6.4J.6) id AA02750; Thu, 4 Feb 93 14:03:21 +0900 Received: by hakobera.isct.kyutech.ac.jp (5.65/6.4J.6) id AA14876; Thu, 4 Feb 93 14:01:24 +0900 Return-Path: Message-Id: <9302040501.AA14876@hakobera.isct.kyutech.ac.jp> To: dg@ai.uga.edu Cc: shimizu@hakobera.isct.kyutech.ac.jp Subject: Category Conversion Date: Thu, 04 Feb 93 14:01:23 +0900 I posted to sci.lang a writing two years ago, but I didn't receive many responses, so I would like to ask you the same question. ================================================ Hi, everybody. I've been interested in the discrepancy between the form and the function of linguistic expressions lately, such as the examples in (1) noted by Quirk et. al.(1985:658): (1) a. A: When are we going to have the next meeting? B: {On Tuesday /In March/ During the vacation/ Between 6 and 7} {will be fine/ suits me/ is what we decided/ may be convenient}. b. He picked up the gun {from under the table/ from behind the curtain}. c. We didn't meet until after the show. d. Food has been scarce since before the war. e. The weather has been fine except in the north. In (1a), for example, the distribution of the PP 'on Tuesday', is in the position of an NP. I have collected some examples myself. (2) a. Now that most of them have first-hand experience with computers, they approach computer applications without fear or superstition and with considerable understanding of how computers can serve man kind. b. At that time, a lot of American companies were looking for clever young scientists from abroad. In (2b), we see the preposition 'from', which usually requires an NP as its object, takes the lexical item 'abroad', which I would presume no dictionary. When we interpret this sentence, I suppose we anyhow regard 'abroad' as functioning as an NP. ===================================================== A few people responded to me. Rick Wojcik (rwojcik@bcsaic.UUCP) gave me interesting examples; (3) Between 45 minutes and an hour will elapse before he turns. (4) The plan requires from six inches to a foot (of space) here. In (3) a PP is treated as a subject,and in (4) as an object. As far as I know, not many people have discussed this phenomenon. Jackendoff(1973) deals with the issue throughly, but I don't think most of us would regard his treatment as satisfactory because the PS rule in the paper is highly idiosyncratic. I suppose he would resort to his correspondence rule now,but I'm not certain exactly how. In Grimshaw(1982), if I understand her correctly, she changes functional schmata and assign SUBJ and OBJ to French clitics (If my memory is correct. Sorry, I don't have it here right now.) But she does not (does she?) discuss the English language, I'm not sure what to do with the examples I put above. I would like to ask you how you would analyse these examples. Any comments and opinions are welcome. I will summarize the responses. I've got another question to ask. I'm thinking of the possibility to rewrite, say, an PP into an NP with a rule similar to Meta Rule in GPSG, but someone told me Meta Rule is not popular these days. But I'm not sure why. Could anyone direct me to the relevant literature, please? \\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\ Makoto SHIMIZU Kyushu Institute of Technology shimizu@hakobera.isct.kyutech.ac.jp \\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\ What did the cannibal say to the captured missionary? "We would like very much to have you for dinner." --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From mcovingt@aisun3.ai.uga.edu Thu Feb 4 11:20:43 1993 Return-Path: Received: from aisun3.ai.uga.edu by aisun1.ai.uga.edu (4.1) id AA27424; Thu, 4 Feb 93 11:20:43 EST Received: by aisun3.ai.uga.edu (4.1) id AA28156; Thu, 4 Feb 93 11:20:39 EST From: mcovingt@aisun3.ai.uga.edu (Michael Covington) Message-Id: <9302041620.AA28156@aisun3.ai.uga.edu> Subject: Address needed To: dg@ai.uga.edu Date: Thu, 4 Feb 93 11:20:39 EST X-Mailer: ELM [version 2.3 PL11] Does anyone have the email address of Michel Eytan (Strasbourg, France)? The address that I have on the DG list is eytan@dpt-info.u-strasbrg.fr, and isn't working. Thanks. -- :- Michael A. Covington internet mcovingt@uga.cc.uga.edu : ***** :- Artificial Intelligence Programs phone 706 542-0358 : ********* :- The University of Georgia fax 706 542-0349 : * * * :- Athens, Georgia 30602-7415 U.S.A. amateur radio N4TMI : ** *** ** --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From @mail-a.bcc.ac.uk:ucleaar@ucl.ac.uk Thu Feb 4 16:48:52 1993 Return-Path: <@mail-a.bcc.ac.uk:ucleaar@ucl.ac.uk> Received: from sun2.nsfnet-relay.ac.uk by aisun1.ai.uga.edu (4.1) id AA00465; Thu, 4 Feb 93 16:48:52 EST Via: uk.ac.bcc.mail-a; Thu, 4 Feb 1993 21:39:14 +0000 Received: from link-1.ts.bcc.ac.uk by mail-a.bcc.ac.uk with SMTP (PP) id <21484-0@mail-a.bcc.ac.uk>; Thu, 4 Feb 1993 21:39:03 +0000 Received: by link-1.ts.bcc.ac.uk (AIX 3.2/UCB 5.64/4.03) id AA11542; Thu, 4 Feb 1993 21:39:01 GMT Message-Id: <9302042139.AA11542@link-1.ts.bcc.ac.uk> To: dg@ai.uga.edu, Daniel Sleator X-Ungarbled_Sender: And Rosta Subject: coordination Date: Thu, 04 Feb 93 21:38:59 +0000 From: And RostaBM-850 From Daniel Sleator's Link Grammar diagram I see that in coordinate structures the conjunction is the head of the conjuncts (or, at least, all conjuncts are subordinate to (indirectly depend on) the conjunction). This approach contrasts with that of Word Grammar, where coordination is the only area of syntax that involves constituency. An example: <--s--< Sophy swam <-------------<----s----< <<< this is a shared { [ Sophy ] and [ Edgar ] } swam dependency I find both approaches appealing. Has anyone done any work on weighing up the pros and cons of each? ---- And Rosta. --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From mahue@linguistik.uni-erlangen.de Sun Feb 14 12:26:47 1993 Return-Path: Received: from faui45.informatik.uni-erlangen.de by aisun1.ai.uga.edu (4.1) id AA10130; Sun, 14 Feb 93 12:26:47 EST Received: from uranus.linguistik.uni-erlangen.de by uni-erlangen.de with SMTP; id AA21931 (5.65c-5/7.3r-FAU); Sun, 14 Feb 1993 18:26:39 +0100 Received: by linguistik.uni-erlangen.de (16.7/7.3h-FAU) id AA22152; Sun, 14 Feb 93 18:26:37 +0100 Date: Sun, 14 Feb 93 18:26:37 +0100 From: Marc Huesken Message-Id: <9302141726.AA22152@linguistik.uni-erlangen.de> To: dg@ai.uga.edu Subject: Subscribe Please add me to the list. ----------------------------------+----------------------------------- Marc Huesken Ei joh! | CLD University of Erlangen Germany mahue@linguistik.uni-erlangen.de | 09131/85-9252 priv.: 09131/67209 ====================================================================== --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From mcovingt@aisun3.ai.uga.edu Tue Feb 23 00:13:35 1993 Return-Path: Received: from aisun3.ai.uga.edu by aisun1.ai.uga.edu (4.1) id AA05512; Tue, 23 Feb 93 00:13:35 EST Received: by aisun3.ai.uga.edu (4.1) id AA12906; Tue, 23 Feb 93 00:13:33 EST From: mcovingt@aisun3.ai.uga.edu (Michael Covington) Message-Id: <9302230513.AA12906@aisun3.ai.uga.edu> Subject: Possible service interruption To: dg@ai.uga.edu Date: Tue, 23 Feb 1993 00:13:33 -0500 (EST) X-Mailer: ELM [version 2.4 PL20] Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Content-Length: 537 The machine that runs the DG mailing list will be undergoing major upgrades this week, and there will be periods when messages posted to DG may be substantially delayed. Please let me know of any problems, especially if they persist. -- :- Michael A. Covington internet mcovingt@uga.cc.uga.edu : ***** :- Artificial Intelligence Programs phone 706 542-0358 : ********* :- The University of Georgia fax 706 542-0349 : * * * :- Athens, Georgia 30602-7415 U.S.A. amateur radio N4TMI : ** *** ** --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From murzaku@ux1sns.sns.it Tue Feb 23 07:19:39 1993 Return-Path: Received: from ux1sns.sns.it by aisun1.ai.uga.edu (4.1) id AA06343; Tue, 23 Feb 93 07:19:39 EST Received: by ux1sns.sns.it (5.57/Ultrix3.0-C) id AA14354; Tue, 23 Feb 93 13:09:04 GMT Date: Tue, 23 Feb 93 13:09:04 GMT From: murzaku@ux1sns.sns.it (Aleksander Murzaku) Message-Id: <9302231309.AA14354@ux1sns.sns.it> To: dg@ai.uga.edu Subject: Syntax checkers Do you know anything about some approach made by someone and concerning the use of DG in the construction of syntax checkers. This can be very useful, I think, especially for the free word order languages such as Russian, Albanian etc. But these languages, have also a very reach flectional system. So there must be added a special morphological parser. Is there anyone working on such languages and such problems? Thanks s in advance, Aleksander Murzaku Scuola Normale Superiore tel.+39/50/597111 Piazza dei Cavalieri 7 murzaku@ux1sns.sns.it 56126 PISA - Italy murzaku at ipisnsib Home adress: Viale Amelia 15 // 00181 ROMA - Italy // Tel. +39/6/7806295 --------------------------------------------------------------------------- Distributed through DG, the international dependency grammar mailing list. To subscribe or unsubscribe, or for assistance, email mcovingt@ai.uga.edu. --------------------------------------------------------------------------- If you REPLY to this mail, your response will go to the person who sent it. To reach the ENTIRE discussion group, MAIL your response to: dg@ai.uga.edu. --------------------------------------------------------------------------- --------------------------------------------------------------------------- From mcovingt@aisun3.ai.uga.edu Wed Feb 24 19:09:21 1993 Return-Path: Received: from aisun3.ai.uga.edu by aisun1.ai.uga.edu (4.1) id AA00556; Wed, 24 Feb 93 19:09:21 EST Received: by aisun3.ai.uga.edu (4.1) id AA00361; Wed, 24 Feb 93 19:09:21 EST From: mcovingt@aisun3.ai.uga.edu (Michael Covington) Message-Id: <9302250009.AA00361@aisun3.ai.uga.edu> Subject: Czech dependency grammar (Resent by M. Covington) To: dg@ai.uga.edu Date: Wed, 24 Feb 1993 19:09:20 -0500 (EST) X-Mailer: ELM [version 2.4 PL20] Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Content-Length: 2757 Forwarded message: From @UGA.CC.UGA.EDU:SGALL@CSPGUK11.BITNET Wed Feb 24 11:26:57 1993 Message-Id: <9302241626.AA00894@aisun3.ai.uga.edu> Date: Wed, 24 Feb 93 17:11:19 MET From: Petr Sgall Subject: Re: Possible service interruption To: Michael Covington In-Reply-To: Your message of Tue, 23 Feb 1993 00:13:33 -0500 (EST) Dear Colleague, I was not successful in using the shorter address, but I would be grateful if you can distribute this message in the group. I have just a few remarks to the ongoing discussion(s): In Czech linguistics the experience with dependency syntax is rather rich, since a highly detailed and consistentversion has been used here even in school grammars for decades, with success, and has been reformulated in our research group in the shape of a formal framework - a synthetic discussion was published by Sgall, Hajicova and Panevova: The meaning of the sentence in its semantic and pragmatic aspects, ed. by J.Mey, Dordrecht, Holl.:Reidel, 1986. There you find, among other issues, a brief analysis of valency of verbs and nouns. Function words (articles, prepositions, conjunctions, aux.verb forms) are handled here, in accordance with the older tradition, just as indices or parts of the lexical occurrences (node labels), since there is no need to inclu