The first #lsa2019 session I'm attending is "Advances in Categorial Grammar and its Application: In Memory of Richard T. Oehrle", with Richard Campbell, Nancy Frishberg, Robert Levine, and Michael J. Moortgat.
First talk: Michael Moortgat - Modes of composition. Linguistic expressions are multidimensional (phonology, syntax, semantics, ...) Composition in each of these dimensions behaves in a specific way, and Richard Oehrle's ideas revolve around composition across the dimensions.
One inspiration for Oehrle's work was Linear Logic (Girard 1987). LL refines traditional logic and makes it more powerful/expressive.
Oehrle's "Multi-modal type-logical grammar" (2011) defines families of logical operations that work along the same underlying principles, but are realized differently in each of the dimensions involved. Again, no expressivity is lost, compared to the linear-logic approach.
For instance, binary operations can signify a particular relationship between two individual elements. Distinct merge algebras are necessary to realize this relationship in morphology, syntax, or the other dimensions.
(Merging morphologically is different from merging syntactically, but both are merging operations.)
Moortgat giving a bunch of examples and formulas I can't reproduce without LaTeX.
The idea of using categorial grammars for other things than syntax is really interesting to me, but there's no time in this session to dive into the details.
Second talk of the #lsa2019 session "Advances in Categorial Grammar and its Application: In Memory of Richard T. Oehrle": Robert Levine - The great breakthrough: functional prosody in Oehrle's term-labeled deductive calculus.
The great breakthrough relates to Montague's problem (Montague 1973): The existence of different readings for sentences like "Every student read a book"
One way to deal with this ambiguity is the Lambek calculus, a type logic where some strings are functors and other strings are arguments. Each of the readings of the sentence can be logically derived in this calculus.
One uncool thing about this is that quantified expressions (like "every student") get different type descriptions depending on their position in the sentence/formula. This is necessary to make sure the rest of the sentence can be combined with the expression.
Cue Oehrle's breakthrough! He defined linguistic signs as triples of phonology, semantics, and syntactic type, such that each part of the sign can be derived compositionally by the same operation...
... as well as a type hierarchy for the prosodic sector in which higher-order types can be defined as functions from lower-order types.
Oehrle introduces lambda-abstraction over prosodic variables. This gives rise to prosodic functors... and this disambiguates the two readings of the sentence (?)
Quantified expressions can now be inserted into the sentence at different points in the derivation, which changes their scope.
Still problematic: "John admires, and Mary praised, the Prime Minister" - this will resolve to "John admires Mary and Mary praised the Prime Minister".
Oehrle's breakthrough is an important contribution, particularly in the area of quantifiers. Levine closes with "No one shall expel us from the paradise that Oehrle has created"
Third talk in the session "Advances in Categorial Grammar and its Application: In Memory of Richard T. Oehrle": Nancy Frishberg - Linguistics in industry
Frishberg's specialty is ASL. She's about to tell us how Oehrle has influenced the way linguistics is practiced in industry.
In bigger corporations, things are often "extremely hierarchical", and you might not get the chance to get individual recognition for your work. Frishberg explains "managing up" and "managing down".
Oehrle was a manager and faced all the challenges of these different kinds of management. He also valued community and participated in many communities.
Oehrle worked on the technology that made the company successful (NLP). But a company needs not only a technology, but also a product, something that can be sold.
In the case of Cataphora (in which Oehrle was involved), the product wasn't only the software, but also the service that professionals provide in setting up the software and maintaining it.
Fourth talk in the session "Advances in Categorial Grammar and its Application: In Memory of Richard T. Oehrle": Richard Campbell - On Dick Oehrle's work at Cataphora #lsa2019
Oehrle started as Chief Linguist at Cataphora in 2003. The company was sold to EY in 2011 and he worked there for another while, then he moved to Chenope (a relaunched version of another part of Cataphora)
Oehrle was awarded two patents, one in 2014 and one in 2016. The first one was about "continuous anomaly detection".
Customers of Cataphora were usually large corporations involved in some legal matter. In these situations, the company has to hand over all douments that have something to do with the topic, but there's a huge amount of data, some confidential, and it's difficult to filter.
In other cases, a company may be trying to find "bad actors" within it - emotional or secretive language can be an indicator that an employee is trying to hide something.
The process of building the ontology that can then be used for filtering involves meeting up with the customer to find key terms and phrases - "it's like fieldwork"
Specialized ontologies were built for emotional and secretive language. Emotional language can be e.g. cursing, secretive can be e.g. "let's talk about this offline", "can't discuss it here"
Oehrle had the idea to use the ENRON email corpus to build up email-specific ontologies.
Ontologies were also expanded to account for misspellings, orthographic variants, and some syntactic alternations.
Oehrle played several important roles at Cataphora: He directed and managed the linguistics team, convinced the company and the customers that linguistics was useful for these applications, and he contributed his own work to the product.
- End of the Richard Oehrle memory session -
