Sunday, March 31, 2013

Linear Logic for Linguists

Cayman Islands from Space
This post is mostly an excuse to add a link to some old ESSLLI 2000 lectures notes on Linear Logic for Linguists. I hope to make them into a book, adding some new material, but in the meanwhile it would be good to have them in a place that I can find. Some of the time...
The notes were also used in a course in Stanford in 2002 by Ash Asudeh and Dick Crouch, Resource Accounting at the Syntax-Semantics Interface. 
and they were the basis of another ESSLLI course in 2004.
Ash and Dick wrote several papers together, of which I need to read "Coordination and Parallelism in Glue Semantics: Integrating Discourse Cohesion and the Element Constraint".

Saturday, March 30, 2013

The Canon of Tech

It took me forever to find this Canon again. The Atlantic is a great magazine and Alexis Madrigal is really good, but I thought the piece was in Wired. and I couldn't remember Madrigal's name. and Google wasn't very helpful as tech canon sends you to the cameras, all the time. With some reason, of course.
And yes, I do know some tricks with queries, but it usually irks me to use them, so I kept thinking of other ways of getting to the article...Glad I finally found it again. I'm keeping a copy for myself somewhere safe.
BTW the photo is from Col. Chris Hadfield, who's the commandant of the International Space Station right now. His photos are absolutely fantastic.

Thursday, March 21, 2013

Temporality in Portuguese

I know nothing about it and should really try to learn a bit, but the tenses work very differently in Portuguese and English.

Apparently, if you say I have lost my watch in English you cannot continue to say but I found it again. This is to be contrasted with saying I lost my watch, but I have found it again which is just fine. Now in Portuguese, with literal translations Eu tinha perdido meu relogio  and Eu perdi meu relogio, both can be followed by mas o achei de novo.  And actually the first one seems better. In English the theory is that The perfect  seems to involve some notion of persisting causal consequence, absent from the past tense.

Need to ask my friend Patricia what does she make of this, what is the story that we have in Portuguese...

Time takes forever

Continuing from Temporality in Natural Language....
 6. Logical representations: should sentence meanings explicitly quantify over times and/or events, or should they have tense operators akin to the propositional operators familiar from modal logic? (vcvp: which operators? past, present, future? much as I like operators, are they enough?)
7. Aspect: what is the significance of constructions like the perfect I have lost
my watch
and the progressive John was crossing the road? (vcvp: significant as they may be, not now...)
8. Aspectual class and coercion: processes vs. states and events..(vcvp:definitely punting)
9. Temporal and nominal constitution: collective/distributive distinctions (vcvp:the same..)
10. Nominal tense: nominal predicates do not normally carry an explicit temporal marking, like verbal ones do, but an individual can be a child at one time and an adult at another. How is their temporal interpretation determined?
11. Temporal adverbials and connectives: how do adverbials interact with tense, aspect and aspectual class? (vcvp: is there a list? how do know if we have most of them?)
12. Sequence of tense: in John said Mary went to London, why is Mary’s going to London prior to John’s reported utterance? (vcvp: how much of this can the language tell us about?)
13. Modals, conditionals and the future: there is a close connection between time and modality. Conditionals too interact with time and tense in complex ways. (vcvp: how to make the minimum number of mistakes on that?)
14. Generics: does a syntactically present tense, but generic, sentence like Lions are mammals make any temporal reference, or does it express an atemporal truth? (vcvp: Similarly the command Play Avatar! is this atemporal or a present or future tense form?)

Anna and Paola in Dagstuhl.

Temporality in Natural Language

This is a very late continuation of Of Time and Other Disasters. On that occasion I had found and posted the notes on temporality in Natural Language for beginners that the husband taught at ESSLLI'98.
I don't like Quine, or since I haven't read much by him (nao li e nao gostei), I don't like the ideas usually associated with him. But I have always agreed with this quote:
 Our language shows a tiresome bias in its treatment of time. Relations of date are exalted grammatically as relations of position, weight and color are not. This bias is of itself an inelegance, or breach of theoretical simplicity. Moreover, the form that it takes that of requiring every verb form to show a tense is peculiarly productive of needless complications, since it demands lip service to be paid to time even when time is farthest from our thoughts.(Quine1960)

Whether we like it or not time is really that important in language, so we need to make some choices and they'd better be defensible for a while.

Which are these choices? Crouch has a list of 14 items in his introduction.
1. The indexical nature of tense: all well formed English sentences appear to build in some indexical reference to the time at which the sentence is uttered. (vcvp: need to pay attention  to this time - the speech time needs to be part of any model, but maybe not of the representation of the meaning of the sentences themselves?)
2. The anaphoric nature of tense: most utterances are about a particular time, though the identity of this time is often only implicitly given by context. (vcvp: only deal with explicit to begin with?)
3. The temporal structure of narrative.  Lascarides example "Max fell, Bob pushed him" as well as "I got out of bed. I turned on the light. (vcvp: only deal with explicit to begin with?)
4. The structure of time: instants versus intervals, all reals...(vcvp: both are necessary)
5. Events, states, propositions and facts:  does language make an ontological commitment to temporally located entities like events or states? (vcvp: some decisions have been made already, can I punt on the others?)
or does it merely relate the truth of propositions to times?

Thursday, March 14, 2013

Old work archeology...

Discussing with Joel and Chris the old papers that might make the Bridge system more understandable I came up with the  list below.Adding it here, as it might be useful for others, including myself...
1. Preventing Existence (with C. Condoravdi, D. Crouch, J. Everett, R. Stolle, D. Bobrow, M. van den Berg), In Proceedings of Formal Ontology in Information Systems, FOIS'01, October 2001 (this is a preliminary version of the paper). And  Entailment, Intensionality and Text Understanding (with Cleo Condoravdi, Richard Crouch, Reinhard Stolle, Daniel G. Bobrow.) Proceedings Human Language Technology Conference (HLT-NAACL-2003), Workshop on Text Meaning, Edmonton, Canada, May 2003. These are about why FOL is not enough for natural language, but maybe you're already OK with this…
2. Precision-focused Textual Inference (with Bobrow, D. G., C. Condoravdi, L. Karttunen, T. H. King, L. Price, R. Nairn, L.Price, A. Zaenen) Proceedings of ACL-PASCAL Workshop on Textual Entailment and Paraphrasing, pp. 16-21, 2007. And PARC's Bridge and Question Answering System (with Daniel G. Bobrow, Bob Cheslow, Cleo Condoravdi, Lauri Karttunen, Tracy H. King, Rowan Nairn, Charlotte Price and Annie Zaenen). Proceedings of Grammar Engineering Across Frameworks, pp 26--45, 2007.
These  are about the systems as a whole, system descriptions…
3. A Basic Logic for Textual inference (with D. Bobrow, C. Condoravdi, R. Crouch, R. Kaplan, L. Karttunen, T. King and A. Zaenen), In Procs. of the AAAI Workshop on Inference for Textual Question Answering, Pittsburgh PA, July 2005.  And Textual Inference Logic: Take Two, (with D. G. Bobrow, C. Condoravdi, R. Crouch,  L. Karttunen, T. H. King, R. Nairn and A. Zaenen)  Proceedings of the Workshop on Contexts and Ontologies, Representation and Reasoning, CONTEXT 2007.
These are about the Logic itself, my main concern.
4. The semantics itself and the Unified Lexicon idea are described in SEMANTICS VIA F-STRUCTURE REWRITING  and Unifying Lexical Resources (to be found in http://logic-forall.blogspot.com/2012/11/dick-tracy.html).
5. The paper on the algorithm to percolate truth-values through contexts is in Nairn, Condoravdi and Kartunnen, Computing Relative Polarity for Textual Inference.

Then there papers on regression, deverbals and other side issues. I need a proper map...

Monday, March 11, 2013

Unilog 2013 is approaching

I have to give three talks, hence prepare at least three decks. Late as usual.

Weyl and trying to do too much

Many times in my life I've tried to do too much and ended up feeling like an idiot. Like the time when I offered to read aloud and record (in a cassette. this was a long time ago...) a book of art history  to a friend of my then boyfriend's grandmother. She had lost her sight and was complaining that she couldn't read the books she wanted, as the libraries for legally blind people only had boring steamy romances. Boy, reading aloud and recording is not easy. The book never ended...and I stopped enjoying very soon...Maybe I should try to read it again.

Another example of biting more than I could chew was accepting to translate into English Giuseppe Longo's article on Hermann Weyl's `The mathematical continuum: from intuition to Logic'. Reading a language well enough to understand a book is in no-way preparation enough to translating it...but yeah, I persevered and these things were done. If only I could do things efficiently...

Saturday, March 9, 2013

Measuring stuff

Sometimes my posts are about stuff I've done that I want to remember. Some times it is about things that I feel strongly about. But mostly is about having conversations with myself, pretending that there is someone else there, in the hopes that the pretense will yield clarity and good arguments...
Yep, I think I am one of these people that reason much more clearly when debating. Don't know if it's the adrenalin that brings about new ideas or what, but I am extremely fond of `mental ping-pong'. I say a bunch of things (and some might be provocative) and you reply with a bunch of stuff that hadn't occurred to me and one way or the other (whether we agree or not) things get a little clarified, a little less fuzzy. And we go on in an infinite approximation of knowledge.

So the post today is about measuring the quality of logics or logical systems, if you prefer.  the question comes about from two very different sides. In one hand, in my real work, I am trying to devise ways of measuring the quality of logical forms. In that case I have a proposal for a system of representations and what I am after is simply an automated system that will compute both the representations and changes to them. Even with very small sets of sentences (where it is feasible to calculate the ideal or golden standard representations) given that you're trying to improve stuff, it's very easy to modify one thing and mess up completely something else that you didn't think was related. Tracy H. King and I had a short paper in Coling 2008 about it, Designing Testsuites for Grammar-based Systems in Applications.

Thus you need systems that will keep you in the straight and narrow and that will work as "forcing functions" to move you forward, not simply in circles...The paper above is about `grammar-based systems' because they bring complications of their own, in terms of storing and comparing representations, which, perhaps, more numerical systems  do not. (for numerical systems maybe you only want to know if you f-score is up or down, or maybe you want precision, recall and f-score, but I think even then I wouldn't be satisfied...)

But the need for these measurements is real and palpable in other systems too. I have now spent months, actually years of my life begging for such systems to be implemented--another story for a rainy day. Anyways adding here another paper of Tracy's on the same subject as I probably need to read it again, Regression Testing For Grammar-Based Systems. 

But on the other hand the question comes from a student of Joao Marcos, on how can we measure the difficulty of derivations or proofs. Of course if one is thinking about different logical systems we can, perhaps, measure how good they are through the number of useful results that one can prove about them. So clearly you want a logic that you can prove sound, completeness is nice and the more complexity facts you know about it the better. Not a huge fan of complexity myself, but being able to prove complexity assertions clearly indicates that the system is amenable to mathematical methods, which is good. Then you can try to show something about expressiveness, either in terms of stuff you could do before or in terms of what you've devised your system to do. and some times you can think of notions of coverage and of precision. do you need long proofs? you must have proofs, systems that are only a model do not count for me. but we're back at the student's question: how do we measure and decide which proofs are good proofs? bad proofs? easy proofs? hard proofs? there are some things one can say, but I sure need to read more and think more about it. 

but not now. now is yoga time.