Pooling layer

In neural networks, a pooling layer is a kind of network layer that downsamples and aggregates information that is dispersed among many vectors into fewer vectors. It has several uses. It removes redundant information, thus reducing the amount of computation and memory required, which makes the model more robust to small variations in the input; and it increases the receptive field of neurons in later layers in the network. == Convolutional neural network pooling == Pooling is most commonly used in convolutional neural networks (CNN). Below is a description of pooling in 2-dimensional CNNs. The generalization to n-dimensions is immediate. As notation, we consider a tensor x ∈ R H × W × C {\displaystyle x\in \mathbb {R} ^{H\times W\times C}} , where H {\displaystyle H} is height, W {\displaystyle W} is width, and C {\displaystyle C} is the number of channels. A pooling layer outputs a tensor y ∈ R H ′ × W ′ × C ′ {\displaystyle y\in \mathbb {R} ^{H'\times W'\times C'}} . We define two variables f , s {\displaystyle f,s} called "filter size" (aka "kernel size") and "stride". Sometimes, it is necessary to use a different filter size and stride for horizontal and vertical directions. In such cases, we define 4 variables: f H , f W , s H , s W {\displaystyle f_{H},f_{W},s_{H},s_{W}} . The receptive field of an entry in the output tensor, y {\displaystyle y} , are all the entries in x {\displaystyle x} that can affect that entry. === Max pooling === Max Pooling (MaxPool) is commonly used in CNNs to reduce the spatial dimensions of feature maps. Define M a x P o o l ( x | f , s ) 0 , 0 , 0 = max ( x 0 : f − 1 , 0 : f − 1 , 0 ) {\displaystyle \mathrm {MaxPool} (x|f,s)_{0,0,0}=\max(x_{0:f-1,0:f-1,0})} where 0 : f − 1 {\displaystyle 0:f-1} means the range 0 , 1 , … , f − 1 {\displaystyle 0,1,\dots ,f-1} . Note that we need to avoid the off-by-one error. The next input is M a x P o o l ( x | f , s ) 1 , 0 , 0 = max ( x s : s + f − 1 , 0 : f − 1 , 0 ) {\displaystyle \mathrm {MaxPool} (x|f,s)_{1,0,0}=\max(x_{s:s+f-1,0:f-1,0})} and so on. The receptive field of y i , j , c {\displaystyle y_{i,j,c}} is x i s + f − 1 , j s + f − 1 , c {\displaystyle x_{is+f-1,js+f-1,c}} , so in general, M a x P o o l ( x | f , s ) i , j , c = m a x ( x i s : i s + f − 1 , j s : j s + f − 1 , c ) {\displaystyle \mathrm {MaxPool} (x|f,s)_{i,j,c}=\mathrm {max} (x_{is:is+f-1,js:js+f-1,c})} If the horizontal and vertical filter size and strides differ, then in general, M a x P o o l ( x | f , s ) i , j , c = m a x ( x i s H : i s H + f H − 1 , j s W : j s W + f W − 1 , c ) {\displaystyle \mathrm {MaxPool} (x|f,s)_{i,j,c}=\mathrm {max} (x_{is_{H}:is_{H}+f_{H}-1,js_{W}:js_{W}+f_{W}-1,c})} More succinctly, we can write y k = max ( { x k ′ | k ′ in the receptive field of k } ) {\displaystyle y_{k}=\max(\{x_{k'}|k'{\text{ in the receptive field of }}k\})} . If H {\displaystyle H} is not expressible as k s + f {\displaystyle ks+f} where k {\displaystyle k} is an integer, then for computing the entries of the output tensor on the boundaries, max pooling would attempt to take as inputs variables off the tensor. In this case, how those non-existent variables are handled depends on the padding conditions, illustrated on the right. Global Max Pooling (GMP) is a specific kind of max pooling where the output tensor has shape R C {\displaystyle \mathbb {R} ^{C}} and the receptive field of y c {\displaystyle y_{c}} is all of x 0 : H , 0 : W , c {\displaystyle x_{0:H,0:W,c}} . That is, it takes the maximum over each entire channel. It is often used just before the final fully connected layers in a CNN classification head. === Average pooling === Average pooling (AvgPool) is similarly defined A v g P o o l ( x | f , s ) i , j , c = a v e r a g e ( x i s : i s + f − 1 , j s : j s + f − 1 , c ) = 1 f 2 ∑ k ∈ i s : i s + f − 1 ∑ l ∈ j s : j s + f − 1 x k , l , c {\displaystyle \mathrm {AvgPool} (x|f,s)_{i,j,c}=\mathrm {average} (x_{is:is+f-1,js:js+f-1,c})={\frac {1}{f^{2}}}\sum _{k\in is:is+f-1}\sum _{l\in js:js+f-1}x_{k,l,c}} Global Average Pooling (GAP) is defined similarly to GMP. It was first proposed in Network-in-Network. Similarly to GMP, it is often used just before the final fully connected layers in a CNN classification head. === Interpolations === There are some interpolations of max pooling and average pooling. Mixed Pooling is a linear sum of max pooling and average pooling. That is, M i x e d P o o l ( x | f , s , w ) = w M a x P o o l ( x | f , s ) + ( 1 − w ) A v g P o o l ( x | f , s ) {\displaystyle \mathrm {MixedPool} (x|f,s,w)=w\mathrm {MaxPool} (x|f,s)+(1-w)\mathrm {AvgPool} (x|f,s)} where w ∈ [ 0 , 1 ] {\displaystyle w\in [0,1]} is either a hyperparameter, a learnable parameter, or randomly sampled anew every time. Lp Pooling is similar to average pooling, but uses Lp norm average instead of average: y k = ( 1 N ∑ k ′ in the receptive field of k | x k ′ | p ) 1 / p {\displaystyle y_{k}=\left({\frac {1}{N}}\sum _{k'{\text{ in the receptive field of }}k}|x_{k'}|^{p}\right)^{1/p}} where N {\displaystyle N} is the size of receptive field, and p ≥ 1 {\displaystyle p\geq 1} is a hyperparameter. If all activations are non-negative, then average pooling is the case of p = 1 {\displaystyle p=1} , and max pooling is the case of p → ∞ {\displaystyle p\to \infty } . Square-root pooling is the case of p = 2 {\displaystyle p=2} . Stochastic pooling samples a random activation x k ′ {\displaystyle x_{k'}} from the receptive field with probability x k ′ ∑ k ″ x k ″ {\displaystyle {\frac {x_{k'}}{\sum _{k''}x_{k''}}}} . It is the same as average pooling in expectation. Softmax pooling is like max pooling, but uses softmax, i.e. ∑ k ′ e β x k ′ x k ′ ∑ k ″ e β x k ″ {\displaystyle {\frac {\sum _{k'}e^{\beta x_{k'}}x_{k'}}{\sum _{k''}e^{\beta x_{k''}}}}} where β > 0 {\displaystyle \beta >0} . Average pooling is the case of β ↓ 0 {\displaystyle \beta \downarrow 0} , and max pooling is the case of β ↑ ∞ {\displaystyle \beta \uparrow \infty } Local Importance-based Pooling generalizes softmax pooling by ∑ k ′ e g ( x k ′ ) x k ′ ∑ k ″ e g ( x k ″ ) {\displaystyle {\frac {\sum _{k'}e^{g(x_{k'})}x_{k'}}{\sum _{k''}e^{g(x_{k''})}}}} where g {\displaystyle g} is a learnable function. === Other poolings === Spatial pyramidal pooling applies max pooling (or any other form of pooling) in a pyramid structure. That is, it applies global max pooling, then applies max pooling to the image divided into 4 equal parts, then 16, etc. The results are then concatenated. It is a hierarchical form of global pooling, and similar to global pooling, it is often used just before a classification head. Region of Interest Pooling (also known as RoI pooling) is a variant of max pooling used in R-CNNs for object detection. It is designed to take an arbitrarily-sized input matrix, and output a fixed-sized output matrix. Covariance pooling computes the covariance matrix of the vectors { x k , l , 0 : C − 1 } k ∈ i s : i s + f − 1 , l ∈ j s : j s + f − 1 {\displaystyle \{x_{k,l,0:C-1}\}_{k\in is:is+f-1,l\in js:js+f-1}} which is then flattened to a C 2 {\displaystyle C^{2}} -dimensional vector y i , j , 0 : C 2 − 1 {\displaystyle y_{i,j,0:C^{2}-1}} . Global covariance pooling is used similarly to global max pooling. As average pooling computes the average, which is a first-degree statistic, and covariance is a second-degree statistic, covariance pooling is also called "second-order pooling". It can be generalized to higher-order poolings. Blur Pooling means applying a blurring method before downsampling. For example, the Rect-2 blur pooling means taking an average pooling at f = 2 , s = 1 {\displaystyle f=2,s=1} , then taking every second pixel (identity with s = 2 {\displaystyle s=2} ). == Vision Transformer pooling == In Vision Transformers (ViT), there are the following common kinds of poolings. BERT-like pooling uses a dummy [CLS] token, "classification". For classification, the output at [CLS] is the classification token, which is then processed by a LayerNorm-feedforward-softmax module into a probability distribution, which is the network's prediction of class probability distribution. This is the one used by the original ViT and Masked Autoencoder. Global average pooling (GAP) does not use the dummy token, but simply takes the average of all output tokens as the classification token. It was mentioned in the original ViT as being equally good. Multihead attention pooling (MAP) applies a multi headed attention block to pooling. Specifically, it takes as input a list of vectors x 1 , x 2 , … , x n {\displaystyle x_{1},x_{2},\dots ,x_{n}} , which might be thought of as the output vectors of a layer of a ViT. It then applies a feedforward layer F F N {\displaystyle \mathrm {FFN} } on each vector, resulting in a matrix V = [ F F N ( v 1 ) , … , F F N ( v n ) ] {\displaystyle V=[\mathrm {FFN} (v_{1}),\dots ,\mathrm {FFN} (v_{n})]} . This is then sent to a multi-headed attention, resulting in M u l t i h e a d e d A

Lesk algorithm

The Lesk algorithm is a classical algorithm for word sense disambiguation introduced by Michael E. Lesk in 1986. It operates on the premise that words within a given context are likely to share a common meaning. This algorithm compares the dictionary definitions of an ambiguous word with the words in its surrounding context to determine the most appropriate sense. Variations, such as the Simplified Lesk algorithm, have demonstrated improved precision and efficiency. However, the Lesk algorithm has faced criticism for its sensitivity to definition wording and its reliance on brief glosses. Researchers have sought to enhance its accuracy by incorporating additional resources like thesauruses and syntactic models. == Overview == The Lesk algorithm is based on the assumption that words in a given "neighborhood" (section of text) will tend to share a common topic. A simplified version of the Lesk algorithm is to compare the dictionary definition of an ambiguous word with the terms contained in its neighborhood. Versions have been adapted to use WordNet. An implementation might look like this: for every sense of the word being disambiguated one should count the number of words that are in both the neighborhood of that word and in the dictionary definition of that sense the sense that is to be chosen is the sense that has the largest number of this count. A frequently used example illustrating this algorithm is for the context "pine cone". The following dictionary definitions are used: PINE 1. kinds of evergreen tree with needle-shaped leaves 2. waste away through sorrow or illness CONE 1. solid body which narrows to a point 2. something of this shape whether solid or hollow 3. fruit of certain evergreen trees As can be seen, the best intersection is Pine #1 ⋂ Cone #3 = 2. == Simplified Lesk algorithm == In Simplified Lesk algorithm, the correct meaning of each word in a given context is determined individually by locating the sense that overlaps the most between its dictionary definition and the given context. Rather than simultaneously determining the meanings of all words in a given context, this approach tackles each word individually, independent of the meaning of the other words occurring in the same context. "A comparative evaluation performed by Vasilescu et al. (2004) has shown that the simplified Lesk algorithm can significantly outperform the original definition of the algorithm, both in terms of precision and efficiency. By evaluating the disambiguation algorithms on the Senseval-2 English all words data, they measure a 58% precision using the simplified Lesk algorithm compared to the only 42% under the original algorithm. Note: Vasilescu et al. implementation considers a back-off strategy for words not covered by the algorithm, consisting of the most frequent sense defined in WordNet. This means that words for which all their possible meanings lead to zero overlap with current context or with other word definitions are by default assigned sense number one in WordNet." Simplified LESK Algorithm with smart default word sense (Vasilescu et al., 2004) The COMPUTEOVERLAP function returns the number of words in common between two sets, ignoring function words or other words on a stop list. The original Lesk algorithm defines the context in a more complex way. == Criticisms == Unfortunately, Lesk’s approach is very sensitive to the exact wording of definitions, so the absence of a certain word can radically change the results. Further, the algorithm determines overlaps only among the glosses of the senses being considered. This is a significant limitation in that dictionary glosses tend to be fairly short and do not provide sufficient vocabulary to relate fine-grained sense distinctions. A lot of work has appeared offering different modifications of this algorithm. These works use other resources for analysis (thesauruses, synonyms dictionaries or morphological and syntactic models): for instance, it may use such information as synonyms, different derivatives, or words from definitions of words from definitions. == Lesk variants == Original Lesk (Lesk, 1986) Adapted/Extended Lesk (Banerjee and Pederson, 2002/2003): In the adaptive lesk algorithm, a word vector is created corresponds to every content word in the wordnet gloss. Concatenating glosses of related concepts in WordNet can be used to augment this vector. The vector contains the co-occurrence counts of words co-occurring with w in a large corpus. Adding all the word vectors for all the content words in its gloss creates the Gloss vector g for a concept. Relatedness is determined by comparing the gloss vector using the Cosine similarity measure. There are a lot of studies concerning Lesk and its extensions: Wilks and Stevenson, 1998, 1999; Mahesh et al., 1997; Cowie et al., 1992; Yarowsky, 1992; Pook and Catlett, 1988; Kilgarriff and Rosensweig, 2000; Kwong, 2001; Nastase and Szpakowicz, 2001; Gelbukh and Sidorov, 2004.

Bartels–Stewart algorithm

In numerical linear algebra, the Bartels–Stewart algorithm is used to numerically solve the Sylvester matrix equation A X − X B = C {\displaystyle AX-XB=C} . Developed by R.H. Bartels and G.W. Stewart in 1971, it was the first numerically stable method that could be systematically applied to solve such equations. The algorithm works by using the real Schur decompositions of A {\displaystyle A} and B {\displaystyle B} to transform A X − X B = C {\displaystyle AX-XB=C} into a triangular system that can then be solved using forward or backward substitution. In 1979, G. Golub, C. Van Loan and S. Nash introduced an improved version of the algorithm, known as the Hessenberg–Schur algorithm. It remains a standard approach for solving Sylvester equations when X {\displaystyle X} is of small to moderate size. == The algorithm == Let X , C ∈ R m × n {\displaystyle X,C\in \mathbb {R} ^{m\times n}} , and assume that the eigenvalues of A {\displaystyle A} are distinct from the eigenvalues of B {\displaystyle B} . Then, the matrix equation A X − X B = C {\displaystyle AX-XB=C} has a unique solution. The Bartels–Stewart algorithm computes X {\displaystyle X} by applying the following steps: 1.Compute the real Schur decompositions R = U T A U , {\displaystyle R=U^{T}AU,} S = V T B T V . {\displaystyle S=V^{T}B^{T}V.} The matrices R {\displaystyle R} and S {\displaystyle S} are block-upper triangular matrices, with diagonal blocks of size 1 × 1 {\displaystyle 1\times 1} or 2 × 2 {\displaystyle 2\times 2} . 2. Set F = U T C V . {\displaystyle F=U^{T}CV.} 3. Solve the simplified system R Y − Y S T = F {\displaystyle RY-YS^{T}=F} , where Y = U T X V {\displaystyle Y=U^{T}XV} . This can be done using forward substitution on the blocks. Specifically, if s k − 1 , k = 0 {\displaystyle s_{k-1,k}=0} , then ( R − s k k I ) y k = f k + ∑ j = k + 1 n s k j y j , {\displaystyle (R-s_{kk}I)y_{k}=f_{k}+\sum _{j=k+1}^{n}s_{kj}y_{j},} where y k {\displaystyle y_{k}} is the k {\displaystyle k} th column of Y {\displaystyle Y} . When s k − 1 , k ≠ 0 {\displaystyle s_{k-1,k}\neq 0} , columns [ y k − 1 ∣ y k ] {\displaystyle [y_{k-1}\mid y_{k}]} should be concatenated and solved for simultaneously. 4. Set X = U Y V T . {\displaystyle X=UYV^{T}.} === Computational cost === Using the QR algorithm, the real Schur decompositions in step 1 require approximately 10 ( m 3 + n 3 ) {\displaystyle 10(m^{3}+n^{3})} flops, so that the overall computational cost is 10 ( m 3 + n 3 ) + 2.5 ( m n 2 + n m 2 ) {\displaystyle 10(m^{3}+n^{3})+2.5(mn^{2}+nm^{2})} . === Simplifications and special cases === In the special case where B = − A T {\displaystyle B=-A^{T}} and C {\displaystyle C} is symmetric, the solution X {\displaystyle X} will also be symmetric. This symmetry can be exploited so that Y {\displaystyle Y} is found more efficiently in step 3 of the algorithm. == The Hessenberg–Schur algorithm == The Hessenberg–Schur algorithm replaces the decomposition R = U T A U {\displaystyle R=U^{T}AU} in step 1 with the decomposition H = Q T A Q {\displaystyle H=Q^{T}AQ} , where H {\displaystyle H} is an upper-Hessenberg matrix. This leads to a system of the form H Y − Y S T = F {\displaystyle HY-YS^{T}=F} that can be solved using forward substitution. The advantage of this approach is that H = Q T A Q {\displaystyle H=Q^{T}AQ} can be found using Householder reflections at a cost of ( 5 / 3 ) m 3 {\displaystyle (5/3)m^{3}} flops, compared to the 10 m 3 {\displaystyle 10m^{3}} flops required to compute the real Schur decomposition of A {\displaystyle A} . == Software and implementation == The subroutines required for the Hessenberg-Schur variant of the Bartels–Stewart algorithm are implemented in the SLICOT library. These are used in the MATLAB control system toolbox. == Alternative approaches == For large systems, the O ( m 3 + n 3 ) {\displaystyle {\mathcal {O}}(m^{3}+n^{3})} cost of the Bartels–Stewart algorithm can be prohibitive. When A {\displaystyle A} and B {\displaystyle B} are sparse or structured, so that linear solves and matrix vector multiplies involving them are efficient, iterative algorithms can potentially perform better. These include projection-based methods, which use Krylov subspace iterations, methods based on the alternating direction implicit (ADI) iteration, and hybridizations that involve both projection and ADI. Iterative methods can also be used to directly construct low rank approximations to X {\displaystyle X} when solving A X − X B = C {\displaystyle AX-XB=C} .

Seismological Facility for the Advancement of Geoscience

The U.S. National Science Foundation's Seismological Facility for the Advancement of Geoscience (NSF SAGE) is a distributed, multi-user national facility that provides support for state of-the-art seismic research. It is operated by EarthScope Consortium. Its previous operator was the Incorporated Research Institutions for Seismology (IRIS), until its merger with UNAVCO to become EarthScope Consortium. NSF SAGE is one of the two premier geophysical facilities in support of geoscience and geoscience education of the National Science Foundation. The other premiere geophysical facility is NSF GAGE, the Geodetic Facility for the Advancement of Geoscience. The services of the facility include support for the Global Seismographic Network (GSN), Data Services, and instrument support via the EarthScope Primary Instrument Center (EPIC), including magnetotelluric (MT) geophysical research. == Global Seismographic Network (GSN) == NSF SAGE manages 40 stations of the 152-station Global Seismographic Network (GSN) for basic global seismicity and Earth structure research. The GSN also enables earthquake hazard mission-related data operations such as: Earthquake location and characterization Tsunami warning Nuclear explosion monitoring == Data Services == SAGE Data Services (DS) is the largest facility for the archiving, curation, and distribution of seismological and other geophysical data in the world. == EarthScope Primary Instrument Center (EPIC) == The EPIC facility maintains the largest open access, shared-use pool of portable seismic sensors in the world. It is located on the campus of New Mexico Tech. == MT == NSF SAGE provides instruments for magnetotelluric (MT) or electromagnetic geophysical research for the recording of our planet's ambient electric and magnetic fields, which allow for the characterization of the conductivity of the area consisting of the shallow crust to upper mantle. This helps with analysis of results obtained from seismic imaging methodologies. The NSF SAGE facility is: Developing open source MT data formatting and processing software. Providing access to proprietary software products.

Lancichinetti–Fortunato–Radicchi benchmark

Lancichinetti–Fortunato–Radicchi benchmark is an algorithm that generates benchmark networks (artificial networks that resemble real-world networks). They have a priori known communities and are used to compare different community detection methods. The advantage of the benchmark over other methods is that it accounts for the heterogeneity in the distributions of node degrees and of community sizes. == The algorithm == The node degrees and the community sizes are distributed according to a power law, with different exponents. The benchmark assumes that both the degree and the community size have power law distributions with different exponents, γ {\displaystyle \gamma } and β {\displaystyle \beta } , respectively. N {\displaystyle N} is the number of nodes and the average degree is ⟨ k ⟩ {\displaystyle \langle k\rangle } . There is a mixing parameter μ {\displaystyle \mu } , which is the average fraction of neighboring nodes of a node that do not belong to any community that the benchmark node belongs to. This parameter controls the fraction of edges that are between communities. Thus, it reflects the amount of noise in the network. At the extremes, when μ = 0 {\displaystyle \mu =0} all links are within community links, if μ = 1 {\displaystyle \mu =1} all links are between nodes belonging to different communities. One can generate the benchmark network using the following steps. Step 1: Generate a network with nodes following a power law distribution with exponent γ {\displaystyle \gamma } and choose extremes of the distribution k min {\displaystyle k_{\min }} and k max {\displaystyle k_{\max }} to get desired average degree is ⟨ k ⟩ {\displaystyle \langle k\rangle } . Step 2: ( 1 − μ ) {\displaystyle (1-\mu )} fraction of links of every node is with nodes of the same community, while fraction μ {\displaystyle \mu } is with the other nodes. Step 3: Generate community sizes from a power law distribution with exponent β {\displaystyle \beta } . The sum of all sizes must be equal to N {\displaystyle N} . The minimal and maximal community sizes s min {\displaystyle s_{\min }} and s max {\displaystyle s_{\max }} must satisfy the definition of community so that every non-isolated node is in at least in one community: s min > k min {\displaystyle s_{\min }>k_{\min }} s max > k max {\displaystyle s_{\max }>k_{\max }} Step 4: Initially, no nodes are assigned to communities. Then, each node is randomly assigned to a community. As long as the number of neighboring nodes within the community does not exceed the community size a new node is added to the community, otherwise stays out. In the following iterations the “homeless” node is randomly assigned to some community. If that community is complete, i.e. the size is exhausted, a randomly selected node of that community must be unlinked. Stop the iteration when all the communities are complete and all the nodes belong to at least one community. Step 5: Implement rewiring of nodes keeping the same node degrees but only affecting the fraction of internal and external links such that the number of links outside the community for each node is approximately equal to the mixing parameter μ {\displaystyle \mu } . == Testing == Consider a partition into communities that do not overlap. The communities of randomly chosen nodes in each iteration follow a p ( C ) {\displaystyle p(C)} distribution that represents the probability that a randomly picked node is from the community C {\displaystyle C} . Consider a partition of the same network that was predicted by some community finding algorithm and has p ( C 2 ) {\displaystyle p(C_{2})} distribution. The benchmark partition has p ( C 1 ) {\displaystyle p(C_{1})} distribution. The joint distribution is p ( C 1 , C 2 ) {\displaystyle p(C_{1},C_{2})} . The similarity of these two partitions is captured by the normalized mutual information. I n = ∑ C 1 , C 2 p ( C 1 , C 2 ) log 2 ⁡ p ( C 1 , C 2 ) p ( C 1 ) p ( C 2 ) 1 2 H ( { p ( C 1 ) } ) + 1 2 H ( { p ( C 2 ) } ) {\displaystyle I_{n}={\frac {\sum _{C_{1},C_{2}}p(C_{1},C_{2})\log _{2}{\frac {p(C_{1},C_{2})}{p(C_{1})p(C_{2})}}}{{\frac {1}{2}}H(\{p(C_{1})\})+{\frac {1}{2}}H(\{p(C_{2})\})}}} If I n = 1 {\displaystyle I_{n}=1} the benchmark and the detected partitions are identical, and if I n = 0 {\displaystyle I_{n}=0} then they are independent of each other.

Language resource

In linguistics and language technology, a language resource is a "[composition] of linguistic material used in the construction, improvement and/or evaluation of language processing applications, (...) in language and language-mediated research studies and applications." According to Bird & Simons (2003), this includes data, i.e. "any information that documents or describes a language, such as a published monograph, a computer data file, or even a shoebox full of handwritten index cards. The information could range in content from unanalyzed sound recordings to fully transcribed and annotated texts to a complete descriptive grammar", tools, i.e., "computational resources that facilitate creating, viewing, querying, or otherwise using language data", and advice, i.e., "any information about what data sources are reliable, what tools are appropriate in a given situation, what practices to follow when creating new data". The latter aspect is usually referred to as "best practices" or "(community) standards". In a narrower sense, language resource is specifically applied to resources that are available in digital form, and then, "encompassing (a) data sets (textual, multimodal/multimedia and lexical data, grammars, language models, etc.) in machine readable form, and (b) tools/technologies/services used for their processing and management". == Typology == As of May 2020, no widely used standard typology of language resources has been established (current proposals include the LREMap, METASHARE, and, for data, the LLOD classification). Important classes of language resources include data lexical resources, e.g., machine-readable dictionaries, linguistic corpora, i.e., digital collections of natural language data, linguistic data bases such as the Cross-Linguistic Linked Data collection, tools linguistic annotations and tools for creating such annotations in a manual or semiautomated fashion (e.g., tools for annotating interlinear glossed text such as Toolbox and FLEx, or other language documentation tools), applications for search and retrieval over such data (corpus management systems), for automated annotation (part-of-speech tagging, syntactic parsing, semantic parsing, etc.), metadata and vocabularies vocabularies, repositories of linguistic terminology and language metadata, e.g., MetaShare (for language resource metadata), the ISO 12620 data category registry (for linguistic features, data structures and annotations within a language resource), or the Glottolog database (identifiers for language varieties and bibliographical database). == Language resource publication, dissemination and creation == A major concern of the language resource community has been to develop infrastructures and platforms to present, discuss and disseminate language resources. Selected contributions in this regard include: a series of International Conferences on Language Resources and Evaluation (LREC), the European Language Resources Association (ELRA, EU-based), and the Linguistic Data Consortium (LDC, US-based), which represent commercial hosting and dissemination platforms for language resources, the Open Languages Archives Community (OLAC), which provides and aggregates language resource metadata, the Language Resources and Evaluation Journal (LREJ), the European Language Grid is a European platform for language technologies (eg services), data and resources. As for the development of standards and best practices for language resources, these are subject of several community groups and standardization efforts, including ISO Technical Committee 37: Terminology and other language and content resources (ISO/TC 37), developing standards for all aspects of language resources, W3C Community Group Best Practices for Multilingual Linked Open Data (BPMLOD), working on best practice recommendations for publishing language resources as Linked Data or in RDF, W3C Community Group Linked Data for Language Technology (LD4LT), working on linguistic annotations on the web and language resource metadata, W3C Community Group Ontology-Lexica (OntoLex), working on lexical resources, the Open Linguistics working group of the Open Knowledge Foundation, working on conventions for publishing and linking open language resources, developing the Linguistic Linked Open Data cloud, the Text Encoding Initiative (TEI), working on XML-based specifications for language resources and digitally edited text.

BRS/Search

BRS/Search is a full-text database and information retrieval system. BRS/Search uses a fully inverted indexing system to store, locate, and retrieve unstructured data. It was the search engine that in 1977 powered Bibliographic Retrieval Services (BRS) commercial operations with 20 databases (including the first national commercial availability of MEDLINE); it has changed ownership several times during its development and is currently sold as Livelink ECM Discovery Server by Open Text Corporation. == Early development == Development on what was to become BRS began as Biomedical Communications Network (BCN) at the State University of New York at Albany (SUNY). BCN, which went online in 1968, provided on-line access to nine databases, including MEDLINE and BIOSIS Previews, to large universities and medical schools primarily in the Northeast of the USA. State funding for the project was withdrawn in 1975, and Bibliographic Retrieval Services (BRS) was formed as a non-profit concern the following year. It was incorporated in May 1976 as a for-profit corporation with Ron Quake as president, Jan Egeland as vice president in charge of marketing and training, and Lloyd Palmer as vice president of systems. == BRS commercial operations == In December 1976, the First BRS User Meeting was held in Syracuse, New York, and by January 1977 BRS started commercial operations with 20 databases (including the first national commercial availability of MEDLINE) and 9 million records, using modified IBM STAIRS (STorage And Information Retrieval System) software, Telenet for telecommunications, and timesharing mainframe computers of Carrier Corporation. In October 1980 BRS was sold by Egeland and Quake to Indian Head, Inc., a subsidiary of the Dutch company Thyssen-Bornemisza Group. == 1989–1993 == In 1989 Robert Maxwell acquired BRS and the BRS/Search software; he announced the planned incorporation of the ORBIT Search Service and BRS Information Technologies and renamed the whole group Maxwell Online, Inc. At that time BRS Information Technologies was serving the medical and academic library marketplace with over 150 databases. Maxwell later bought the publishing company Macmillan and put Maxwell Online under Macmillan. In the same year BRS/LINK (hypertext connection of databases; first application delivering full text) was announced. The initial BRS/LINK application "relates the citation in a bibliographic database to its full-text article in a second database," and "eliminates the need to re-execute a search strategy in the second database in order to find the corresponding full-text article." Initially BRS/LINK supported linking only selected bibliographic databases: MEDLINE, Health Planning and Administration, and MEDLINE References on AIDS to the full-text Comprehensive Core Medical Library. At the time of Robert Maxwell’s death in 1991, Macmillan brought in Andrew Gregory to represent the company during the 2 years that Maxwell’s affairs were being settled and to prepare Maxwell Online to be able to sell the components. Maxwell Online shortly thereafter underwent yet another name change, this time to InfoPro Technologies. == Dataware Technologies ownership of BRS/SEARCH == Early in 1994, InfoPro Technologies, a subsidiary of MHC Inc. (holding company for Macmillan Inc.), the former Maxwell Online service, sold off all its subsidiaries. ORBIT Search Services went to the French-owned Questel, the dial-up BRS Search Services to CD Plus Technologies (later to become OVID), and BRS Software Products (including BRS/SEARCH) to Dataware Technologies. Almost up to the end of InfoPro Technologies, BRS Software had been the fastest growing segment of the company. At the 14th BRS North American Users Group Conference in 1999, Dave Schubmehl of Dataware Technologies presented a paper in which he stated "The purpose of this presentation is to update BRS users on upcoming releases of BRS/Search, NetAnswer, and other Dataware products. BRS/Search 7.0 will include features specifically requested by customers, as well as other enhancements. Earlier this year, Dataware acquired Sovereign Hill Software, makers of InQuery. In light of that acquisition, and Dataware's other development projects, we'll look at Dataware's plans for all products, including BRS/Search and NetAnswer." == Open Text acquisition of BRS/Search == In 2001 BRS/Search was acquired by Open Text and became LiveLink ECM Discovery Server. It is now referred to as Open Text Discovery Server. Open Text still supports both BRS/Search and NetAnswer. The core BRS/Search technology in the Open Text portfolio was augmented with other capabilities through various acquisitions. For example, Dataware's acquisition of Sovereign-Hill brought InQuery, “a probabilistic information retrieval system using an inference network”, which was developed by the University of Massachusetts Amherst Center for Intelligent Information Retrieval] out of the UMass CIIR and into the marketplace. A product re-branding table shows the range of products, their old names and their new names. InQuery is a concept search engine that uses noun phrases, parts of speech and other co-occurrence relationships in overlapping passages of text rather than single term inverted indexes of single words in documents. Open Text's portfolio has grown to include Hummingbird Content Management, and has always included BASIS. == 2003 == BRS/Search North America User's Group (BRSNAUG) website with a June 8, 2003 date listed the following features for BRS/Search. The BRSNAUG also disincorporated in 2003. Cross-references to BRS/Search on the World Wide Web point to Open Text Livelink. Engine features include: Rapid query response time. Numerical data handling and elementary statistical processing (sum, avg, min, max) Search results weighting and relevancy ranking Left- and right-truncation and expansion of search terms Superior data compression – loaded databases typically use only about 1.5 times the input stream size in disk space Large capacity databases – up to 100 million documents, each with up to 65,000 paragraphs Fine control of indexing and searching – right down to the word, sentence, and paragraph level Fine control over data security. Document access can be controlled at the database, document, and paragraph level International language support for all 7/8 bit characters sets and customizable language tables Flexible and customizable stop word lists ANSI-compatible thesauri Hypertext links within and between documents and databases (R6.x) Support for natural language parsing of queries Automatic document summarization tools Client/Server development Programming interfaces for World-Wide Web (HTTP, HTML) access to databases