No abstract

How do neural language models keep track of number agreement between subject and verb? We show that 'diagnostic classifiers', trained to predict number from the internal states of a language model, provide a detailed understanding of how, when, and where this information is represented. Moreover, they give us insight into when and where number information is corrupted in cases where the language model ends up making agreement errors. To demonstrate the causal role played by the representations we find, we then use agreement information to influence the course of the LSTM during the processing of difficult sentences. Results from such an intervention reveal a large increase in the language model's accuracy. Together, these results show that diagnostic classifiers give us an unrivalled detailed look into the representation of linguistic information in neural models, and demonstrate that this knowledge can be used to improve their performance.

Immobilization of metal-organic frameworks (MOFs) onto flexible polymeric substrates as secondary supports expands the versatility of MOFs for surface coatings for the development of functional materials. In this work, we demonstrate the deposition of copper(II) benzene-1,3,5-tricarboxylate (CuBTC) crystals directly onto the surface of carboxyl-functionalized cotton capable of generating the therapeutic bioagent nitric oxide (NO) from endogenous sources. Characterization of the CuBTC-cotton material by XRD, ATR-IR, and UV-vis indicate that CuBTC is successfully immobilized on the cotton fabric. In addition, SEM imaging reveals excellent surface coverage with well-defined CuBTC crystals. Subsequently, the CuBTC-cotton material was evaluated as a supported heterogeneous catalyst for the generation of NO using S-nitrosocysteamine as the substrate. The resulting reactivity is consistent with the activity observed for unsupported CuBTC particles. Overall, this work demonstrates deposition of MOFs onto a flexible polymeric material with excellent coverage as well as catalytic NO release from S-nitrosocysteamine at therapeutic levels.

The versatile chemical and physical properties of metal organic frameworks (MOFs) have made them unique platforms for the design of biomimetic catalysts, but with only limited success to date due to instability of the MOFs employed in physiological environments. Herein, the use of Cu(II)1,3,5-Benzene-tris-triazole (CuBTTri) is demonstrated for the catalytic generation of the bioactive agent nitric oxide (NO) from endogenous sources, S -nitrosothiols (RSNOs). CuBTTri exhibits structural integrity in aqueous environments, including phosphate buffered saline (76 h, pH 7.4, 37 °C), cell media used for in vitro testing (76 h, pH 7.4, 37 °C), and fresh citrated whole blood (30 min, pH 7.4, 37 °C). The application of CuBTTri for use in polymeric medical devices is explored through the formation of a composite CuBTTripoly by blending CuBTTri into biomedical grade polyurethane matrices. Once prepared, the CuBTTri-poly material retains the catalytic function towards the generation of NO with tunable release kinetics proportional to the total content of CuBTTri embedded into the polymeric material with a surface fl ux corresponding to the therapeutic range of 1-100 n M cm −2 min −1 , which is maintained even following exposure to blood.

The use of metal organic frameworks (MOFs) for the catalytic production of nitric oxide (NO) is reported. In this account we demonstrate the use of Cu(3)(BTC)(2) as a catalyst for the generation of NO from the biologically occurring substrate, S-nitrosocysteine (CysNO). The MOF catalyst was evaluated as an NO generator by monitoring the evolution of NO in real time via chemiluminescence. The addition of 2, 10, and 15-fold excess CysNO to MOF-Cu(II) sites and cysteine (CysH) resulted in catalytic turnover of the active sites and nearly 100% theoretical yield of the NO product. Control experiments without the MOF present did not yield appreciable NO generation. In separate studies the MOF was found to be reusable over successive iterations of CysNO additions without loss of activity. Subsequently, the MOF catalyst was confirmed to remain structurally intact by pXRD and ATR-IR following reaction with CysNO and CysH.