Re: [lojban] the myth of monoparsing

2015-02-21 2:32 GMT+03:00 ianek <janek37@gmail.com>:

On Friday, February 20, 2015 at 2:24:22 PM UTC+1, la gleki wrote:

2015-02-13 0:04 GMT+03:00 ianek <jan...@gmail.com>:

On Thursday, February 12, 2015 at 8:00:17 PM UTC+1, la gleki wrote:

2015-02-12 21:42 GMT+03:00 ianek <jan...@gmail.com>:

On Thursday, February 12, 2015 at 7:11:12 AM UTC+1, la gleki wrote:

2015-02-12 1:20 GMT+03:00 ianek <jan...@gmail.com>:

On Wednesday, February 11, 2015 at 1:50:49 PM UTC+1, la gleki wrote:

2015-02-09 23:22 GMT+03:00 ianek <jan...@gmail.com>:

On Monday, February 9, 2015 at 11:54:41 AM UTC+1, la gleki wrote:

2015-02-08 4:34 GMT+03:00 ianek <jan...@gmail.com>:

On Friday, February 6, 2015 at 8:13:30 AM UTC+1, la gleki wrote:

2015-02-04 15:45 GMT+03:00 v4hn <m...@v4hn.de>:
On Tue, Feb 03, 2015 at 11:42:32AM +0300, Gleki Arxokuna wrote:
> "Fred saw a plane flying over Zurich" can have several meanings

Yes.
However, for me, the issue here is that we (hopefully..) agree
that there are different parse trees (which yield the different meanings).

No, several trees arise after you interpret the sentence.

But if you had an English parser, it would yield several trees without any interpreting.

Sure! Because English parsers lack the ability to find something common in all of the parse trees.

No. It's because words in an English sentence can be parsed as different syntactic structures. That's what parsing means: determining structures formed by words. Not "finding something common".

You yourself just showed several parses of the same sentence.
This is how usual English parsers are constructed.

However, there is another option to monoparse this English sentence.

You mix English language and one current theory of how to parse it.

Like this:

"Fred saw a plane flying over Zurich"
NAME VERB-PAST ARTICLE COUNTABLE-NOUN VERB-ING PREPOSITION NAME

Some (much simplified) rules could be:

Sentence ::= Noun-Phrase Verb Noun-Phrase
Sentence ::= Noun-Phrase Verb Noun-Phrase Adverbial-Phrase
Noun-Phrase ::= NAME | ARTICLE COUNTABLE-NOUN | Noun-Phrase VERB-ING Prepositional-Clause
Verb ::= VERB-PAST
Adverbial-Phrase ::= VERB-ING Preposition-Clause
Preposition-Clause ::= PREPOSITION Noun-Phrase

This simple grammar yields two parse trees for that sentence:

Sentence
----Noun-Phrase
--------NAME
------------Fred
----Verb
--------VERB-PAST
------------saw
----Noun-Phrase
--------Noun-Phrase
------------ARTICLE
----------------a
------------NOUN
----------------plane
--------VERB-ING
------------flying
--------Prepositional-Clause
------------PROPOSITION
----------------over
------------Noun-Phrase
----------------NAME
--------------------Zurich

Sentence
----Noun-Phrase
--------NAME
------------Fred
----Verb
--------VERB-PAST
------------saw
----Noun-Phrase
--------Noun-Phrase
------------ARTICLE
----------------a
------------NOUN
----------------plane
----Adverbial-Phrase
--------VERB-ING
------------flying
--------Prepositional-Clause
------------PROPOSITION
----------------over
------------Noun-Phrase
----------------NAME
--------------------Zurich

Formal grammars for natural languages do exist, although they're not perfect, but the problem with multiple grammatically sensible parses (often millions of trees and more) is much greater than the problem with nonsensible trees or correct sentences that don't parse at all.

Lojban was carefully designed to avoid this problem. And it doesn't have anything to do with {xi PA}. The Lojban grammar specifies XI clauses unambiguously. Parse trees are unique. Monoparsing is not a myth. XI clauses may add semantic ambiguity on a different level then, say, simple {zo'e}, but it doesn't have anything to do with syntactic ambiguity.

It specifies to which head a clause should attach. And since it's {mo'e zo'e} it's vague to which head it attaches. If the parser you use doesn't allow for that the only thing that can be done is to provide several possible trees.

It's a feature of a language, not a parser. If English had a pronoun, say, 'lar', which would mean 'the subject or the object of the main sentence', you could say "Fred saw a plane as lar flew over Zurich", which would be ambiguous semantically, but not syntactically.

Even in current English theory there are a lot of zero morphemes. What I'm proposing is just another zero morpheme.

This is what And agreed with me.

{la fred pu viska lo vinji do'e lo se xi vei mo'e zo'e nei poi vofli ga'u la tsurix} has only one syntax tree, regardless of the number of possible semantic interpretations.

If you applied {mo'e zo'e} to the English sentence you will still get the only syntax tree.

You can't "apply" {mo'e zo'e} to the English sentence, because it's not there. Likewise you don't "apply" {mo'e zo'e} to the Lojban sentence. You just parse it, because it's there.
In English you can have phrases like 'X of Y of Z' which could be parsed as '(X of Y) of Z' or 'X of (Y of Z)'. In Lojban it's not possible, but you can say ''either (X of Y) of Z or X of (Y of Z)", which is not syntactically ambiguous. You can't apply "either... or" to the English sentence, because you can't parse words which aren't there.

As I just said English parsers use this "add words that aren't there" all the time.

I was searching, but I haven't found any English parser (but I know a Polish one). What parsers do you refer to?

Probably most. Since this concept (of adding words and morphemes of zero length) is present in most modern theories:
https://en.wikipedia.org/wiki/Zero_(linguistics)

This doesn't answer my question. Name at least one working English parser. I haven't found any.

Which requirements do you need? Take Stanford's parser.
But if you want an English parser that would insert zero morpheme to reach vague syntax I'm not aware of any although it's obvious (I hope so) that it's possible to create one (although probably useless since no one including me suggested any possible advantages apart from purely theoretical ones).

It's far from obvious. Natural languages have countless types of syntactic ambiguity, and I'm not sure all of them could be overcome on the parser level.

That's why I asked for examples. For those that by far been presented I used an alternative understanding.

But what if there could be infinitely many examples? It's possible, as there's no upper limit to the length of a sentence. You can't have infinite grammar (technically you could, but it's pointless).

Yes, technically it's unfalsifialbe since you can't be sure that example No. 1256 won't fail to prove or disprove monoparsing in English.

However, note that

1. I don't consider syntactic ambiguity and vague syntax (vague '=' operator) as methods complementing each other. In fact they are rival methods. I consider both of them fine. But this is what can make the claim that Lojban is the only monoparsing language false. It of course means that parsers of Lojban are monoparsing.

2. The grammar of English although is large probably not infinite.

Of course, I respect hundreds of years grammarians put into studying English.

But what I respect more is the English language itself that has thousands of years of development.

In Lojban the natural grammar yields a monoparsing parser (no artificial zero words etc.). To make a polyparsing one, you'd have to do some weird stretches. In English it's the other way around.

When one says "I like Lojban for monoparsing" that person doesn't love Lojban but loves the way it is presented.

Monoparsing is natural in Lojban. You'd have to be really stubborn to polyparse Lojban.

No, it won't be stubborn and monoparsing isn't natural to Lojban.

It's natural to textbooks, learning resources and parsers of this language.

Presenting a language as monoparsing creates an illusion that this language is inherently different from others.

However, this is not the time to fight Chomsky's ideas.

OTOH, polyparsing is much more natural for English than monoparsing (it it's even possible, as you claim).

No, again it's only how grammarians describe this language.

When you say in Lojban {xi mo'e zo'e}, you conciously express ambiguity, using unambiguous syntax. When you say an English sentence with >1000 natural parses, you're not concious of them and you're usually not willing to express ambiguity.
All of that is really obvious.

So you reduce all your reasoning to zero morphemes? This is the same argument pycyn put earlier. Different languages have different sets of zeros.

However, as shown earlier by Ilmen even without {xi vei mo'e zo'e} a simple {broda i brode} creates a wide range of meanings. Note that {zo'e} here are implicit.

It's also highly doubtful what is said consciously and what is not. Are zeros the only parts of the language expressed unconsciously according to you?

Language is one thing and its description is another one.

At the moment I can't see polyparsing in English except in polysemy but that is said not to be relevant here.

I guess that your votum is separatum.

I don't care how many people for and against here.

E.g And Rosta seems to have a different understanding what grammar is and how it is connected to parsing although Parsing _expression_ Grammar hints at what it is for.

It of course results in the inability of a fair comparison of Lojban and English parsers. But that's acceptable.

In English you can have sentences that are semantically ambiguous due to syntactic ambiguity. In Lojban you can have sentences with (roughly) the same semantic ambiguity as the English ones, but syntactically unambiguous.

> {la fred pu viska lo vinji do'e lo se xi vei mo'e zo'e nei poi vofli ga'u
> la tsurix}

camxes only produces one parse tree for that.

And for English you don't provide any parses at all.
May be someone should just parse the original English sentence as camxes does for Lojban one?
I won't be surprised if such parser for English doesn't exist since those who write them might mix parsing and interpretation of it. The latter would be replacing {mo'e zo'e} with some PA which will immediately lead to several syntactic trees.

So I both disagree and agree with you on whether English sentence has several syntactic trees. If using one term for two operations is stopped the contradiction disappears.

If you think it should produce more then one, raise a bug report.

I'm not aware of any Lojban parsers that perform interpretation operation. In most cases you just need context and one interpretation. But this is semantic analysis. Producing all possible syntactic trees is a task needed more seldom.

Camxes is intended to produce all possible syntactic trees, and there's only one of them for any valid sentence.

You may invent a Lojban parser that won't be able to parse {mo'e zo'e}. Then you will need workarounds to output several trees.

XI clauses have an ambiguous syntax, so I don't see how I'd need workarounfds and several trees. Of course, I could invent a Lojban parser that won't be able to parse anything, but what's the point? {mo'e zo'e} from the parser's view is just MOhE KOhA. If I can't parse it, then I have an incomplete parser.

And this is what I state for English: its current parsers are incomplete and further improvements will make polyparsed sentences monoparsed.

What you mean sounds rather like a semantic analyzer, which is extremely hard for any language, including Lojban.

mu'o mi'e ianek

mu'o mi'e ianek

--
You received this message because you are subscribed to the Google Groups "lojban" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lojban+un...@googlegroups.com.
To post to this group, send email to loj...@googlegroups.com.
Visit this group at http://groups.google.com/group/lojban.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "lojban" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lojban+un...@googlegroups.com.
To post to this group, send email to loj...@googlegroups.com.
Visit this group at http://groups.google.com/group/lojban.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "lojban" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lojban+un...@googlegroups.com.
To post to this group, send email to loj...@googlegroups.com.
Visit this group at http://groups.google.com/group/lojban.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "lojban" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lojban+un...@googlegroups.com.
To post to this group, send email to loj...@googlegroups.com.
Visit this group at http://groups.google.com/group/lojban.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "lojban" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lojban+un...@googlegroups.com.
To post to this group, send email to loj...@googlegroups.com.
Visit this group at http://groups.google.com/group/lojban.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "lojban" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lojban+unsubscribe@googlegroups.com.
To post to this group, send email to lojban@googlegroups.com.
Visit this group at http://groups.google.com/group/lojban.
For more options, visit https://groups.google.com/d/optout.