Thursday, August 15, 2013

Wittgenstein and Mentalese

I first want to say that I edited my last post to replace the phrase "vector types" with "vector categories". I think the NLP community is more comfortable with the term category, as in "syntactic category".  The phrase "vector categories" may have a better change of going viral in the NLP community.

This post is on Wittgenstein and Mentalese.   For those not familiar with phrase, "mentalese"refers to a hypothetical language of thought --- some form of symbolic representation of propositions and beliefs manipulated by the computational process of thought.  Mentalese seems related to the concept of an interlingua in machine translation or logical forms in linguistics.  One interpretation of "semantics" is the relationship between the surface form of language and internal mentalese expressions.  Before going any further, the first sentence of the lead story at as of this writing is:

President Obama said Thursday his government "strongly condemns" violence in Egypt, and he is canceling U.S.-Egyptian military exercises that had been scheduled for next month.

What should we take the internal representation of this sentence to be?  I have argued in previous posts that semantics should be viewed as a relationship between sentences and a database model of reality.  The above sentence presents a new fact (I will assume that it is true) that Obama said what he is claimed to have said.  Reading this sentence might cause me to update my model of reality.  I have argued in previous posts that entities can simply be symbolic tokens such as entity-3687 which occur in relations  such as

entity-3687 is an organization
entity-3687 is named "the Muslim Brotherhood".

We might generate a database representation of the above sentence as follows where I will write entity tokens as symbols starting with *.

*E is an event
*E is the referent of " *Obama said *P "
*E happened on *D
*P states *G strongly condemns *V
*P states *Obama is canceling  *Exc

The reader may already be familiar with the entities *Obama, the day Thursday *D, Obama's government *G, the current violence in Egypt *V, and the scheduled exercises *Exc.  If these entities are well established then perhaps the above relations are all that need to be added to the database. Note that in this case the reference resolution drops most of the verbatim wording of the sentence.

I mention Wittgenstein here because of his emphasis on "public language" as opposed to "private language".  I like the emphasis on public language because it seems plausible to me that the relations of the database model of reality are actually just fragments of public language (of, say, English) with certain phrases replaced by entity tokens.  In this view mentalese is essentially just public language --- there is no mysterious mentalese.  This seems consistent with "shallow" machine translation and the apparent lack of a need for an interlingua.  Translation can be viewed as paraphrase --- a direct substitution of one surface form for another.  Thought itself may take place in a similar manner --- as a direct manipulation of surface forms.

I heard Chris Manning take the position at a panel on language understanding that dependency parses were perhaps an adequate representation of semantics.  I would add reference resolution --- a dependency parse with resolved references seems to go a long way toward meaning.

Of course there  is a large literature on tense, aspect, modality, counterfactuals, and quantification.   Rigorous mathematical thought must involve some representation of mathematically rigorous statements.  I don't know how to reconcile these things with the idea that mentalese might be just fragments of public language.  However I find it plausible that some reconciliation exists --- for example, that precise mathematical thought can be carried out in fragments of English.  I certainly take the public language hypothesis seriously.


  1. I think "formal semantics" set up generations of linguists to confuse syntax with semantics. Semantics is about meaning and how language connects to the world. Syntax is about syntactic relations. Most linguists really just mean rich syntax when they talk about "semantics".

    Dependency and other parses are syntax. If I say "It ran", a dependency parse will tell me that "it" is the subject of the verb "ran" and if I'm lucky, that "ran" is "run" is in the past tense. It doesn't tell me what "It" refers to or which sense of "ran" is intended (tear as in stockings, drip as paint or an ice cream, locomotion as with an animal, etc.).

    Any idea what Chris meant when he said that dependency parses were adequate? They're certainly not adequate for understanding what someone means by a sentence.

    I also think of anaphora as primarily syntactic. If I say "John ran. And he jumped." and you tell me that "he" corefers with "John", it's still not semantics because the language isn't being linked to the world. Now if I told you who or what "John" referred to, that would be semantics. Or if I told you what it meaned to jump.

    1. I wrote the post on Tarski and mentalese largely in response to this comment, but I think I should also respond more directly. There is considerable work these days on cross-document and ever cross-lingual coreference. The current trend in coreference is to work with mention-entity models rather than mention-mention models. In an mention-entity model of co-reference across thousands of documents the "entities" act like semantic referents. This is especially true if we take the entities directly from a database such as freebase as is done in

      I expect that when Chris says that a dependency parse is largely adequate he means that it is an adequate for applications which would seem to require semantics --- such as translation and perhaps even entailment. But for me "semantics" should include reference resolution in the sense of the above URL but where freebase is replaced by a mentalese database.