Detmar Meurers (dm@sfs.nphil.uni-tuebingen.de)
Tue, 28 Jan 1997 19:27:36 +0100 (MET)
In the message "adverb extraction and lexical nonproliferation", Ivan discusses the question whether lexical rules lead to a proliferation of lexical entries. To shed some light on this issue, I believe one needs to take a look at how lexical rules are interpreted in HPSG - especially since there are different approaches which lead to different answers to the original question. Two formalizations of lexical rules as used by HPSG linguists have been proposed, the meta-level lexical rules (MLRs) and the description-level lexical rules (DLRs). The former approach is introduced in a paper by Mike Calcagno and presentations by Carl Pollard, the latter in a paper of mine. Another formalization of DLRs is discussed in a paper by Suresh Manandhar. The references are included below - here only the relevant basic ideas are mentioned: In the terminology of both approaches, the linguist specifies the lexicon as a set (or disjunction) of lexical entries, and each lexical entry describes a set of word objects. The MLR approach sees lexical rules in the traditional sense as relations between lexical entries. The set of lexical entries is closed under the application of lexical rules, which results in a (possibly infinite) set of lexical entries. In order to be grammatical, every word object occurring in a sentence has to be described by one of the descriptions in this extended lexicon set. In the MLR setup, lexical rules are thus external to the rest of the theory (in the formal sense), they only serve to provide an extended lexicon set. Licensing grammatical words is then done by this set - the lexical rules play no direct role. The DLR approach formalizes lexical rules as relations between word objects. Lexical rules under this approach are part of the theory, just like any other constraint of the grammar, and they relate the word objects licensed by the original lexical entries to another set of well-formed word objects. The constraint on word which encodes the lexicon is modified to include a lexical rule description which licenses the additional word objects. Thus, no new lexical entries are created, but the modified theory itself licenses more grammatical words. Regarding Ivan's original point this means that the DLR formalization of lexical rules does not lead to 'a proliferation of lexical entries', whereas the MLR formalization in a sense does. However, one can wonder, whether this is the right criterion by which to judge a lexical rule mechanism: One advantage of the traditional view of lexical rules as relating lexical entries (i.e. specifications) which is embodied in the MLR approach is that it provides for a clean separation between the lexicon and syntax. MLRs can only refer to specifications present in lexical entries and `produce' new lexical entries made up of related specifications. What is not specified in the lexical entries therefore clearly cannot be related by a lexical rule and MLRs do not interact with constraints expressed in the theory in any direct way. The DLR approach on the other hand includes lexical rules on a par with all other grammatical generalizations. DLRs therefore can be used to express syntactic generalizations, e.g. about the arguments raised onto the SUBCAT list of an auxiliary by Hinrichs/Nakazawa argument-raising. Now, such an approach of lexicalizing syntax has been taken in several papers - but I think one should be aware that such an approach leaves the traditional idea of lexical rules behind and in essence does away with the lexicon as something separate from syntax. If lexical rules are to play such a powerful role, the development of general, linguistically motivated restrictions on what lexical rules can do probably deserves some attention. Finally, Ivan's discussion touches upon another tradition in HPSG, which has sometimes been referred to as ``lexical underspecification''. The basic idea is to use underspecification of lexical entries together with a sophisticated organization of types to express the relatedness of words in the theory without using lexical rules. As examples for this approach one could probably cite Kathol:94a, Frank:94a, Frank:94b, Oliva:94, Sanfilippo:95a, Sanfilippo:95b. I believe those approaches are very close to the encoding Ivan seems to have in mind - but as far as I know they do not intend to cover recursive application of something like `add adjuncts'. Also, note that a lexical underspecification approach (or in fact any proposal which uses the theory to express generalizations over lexical information) is incompatible with the traditional approach to lexical rules, since the whole purpose of lexical underspecification is to not specify everything in the lexical entries but obtain the restrictions by interaction with the constraints expressed in the theory. Gru3, Detmar ---------------------- Detmar Meurers SFB 340, Project B4/B8 Kleine Wilhelmstr. 113 72074 Tuebingen Germany Tel. +49-(0)7071-2977314 http://www.sfs.nphil.uni-tuebingen.de/~dm/ ---------------------- ---------- References ---------- a) lexical rule formalization @inproceedings{Calcagno:95, author = "Mike Calcagno", title = "Interpreting Lexical Rules", booktitle = "Proceedings of the Formal Grammar Conference", publisher = "Universidad Politecnica de Catalunya, Barcelona", email = "calcagno@ling.ohio-state.edu", homepage = "http://ling.ohio-state.edu/People/MikeCalcagno.html", year = "1995" } @misc{Manandhar:95, author = "Suresh Manandhar", title = "The Update Operation in Feature Logic", note = "Ms., December 1995", email = "suresh@cs.york.ac.uk" } @inproceedings{Meurers:95, author = "Walt Detmar Meurers", title = "Towards a Semantics for Lexical Rules as used in HPSG", booktitle = "Proceedings of the Formal Grammar Conference", publisher = "Universidad Politecnica de Catalunya, Barcelona", email = "dm@sfs.nphil.uni-tuebingen.de", url = "http://www.sfs.nphil.uni-tuebingen.de/~dm/LR/sem.ps.gz", homepage = "http://www.sfs.nphil.uni-tuebingen.de/~dm/", year = "1995" } b) underspecification @incollection{Kathol:94b, address = "Stanford University", author = "Andreas Kathol", booktitle = "German in {Head-Driven Phrase Structure Grammar}", editor = "John Nerbonne and Klaus Netter and Carl Pollard", homepage = "http://grid.let.rug.nl/~kathol/", number = "46", pages = "237--272", publisher = "CSLI Publications", series = "Lecture Notes", title = "Passives without Lexical Rules", year = "1994" } @inproceedings{Frank:94a, author = {Annette Frank}, title = {Verb Second by Underspecification}, booktitle = {KONVENS '94}, year = {1994}, editor = {Harald Trost}, publisher = {Springer-Verlag}, address = {Berlin}, pages = {121--130} } @techreport{Frank:94b, author = {Annette Frank}, title = {Verb Second by Lexical Rule or by Underspecification}, institution = {IMS}, year = {1994}, address = {Stuttgart}, number = {43}, type = {Arbeitsberichte des Sonderforschungsbereichs 340} } @INPROCEEDINGS{Oliva:94, AUTHOR = {Karel Oliva}, ADDRESS = {Kyoto, Japan}, BOOKTITLE = {Proceedings of the 15th Conference on Computational Linguistics}, TITLE = {{\sc hpsg} {L}exicon without {L}exical {R}ules}, YEAR = {1994} } @INPROCEEDINGS{Sanfilippo:95a, AUTHOR = {Antonio Sanfilippo}, TITLE = {Lexical Polymorphism and Word Usage Extensibility}, BOOKTITLE = {Proceedings of the ACQUILEX II Workshop on Lexical Rules}, YEAR = {1995}, ADDRESS = {Cambridge, UK} } @INPROCEEDINGS{Sanfilippo:95b, AUTHOR = {Antonio Sanfilippo}, TITLE = {Lexical Polymorphism and Word Disambiguation}, BOOKTITLE = {Proceedings of the AAAI-95}, YEAR = {1995} }
This archive was generated by hypermail 2.0b3 on Fri Dec 18 1998 - 20:36:03 PST